fs/ext4/super.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/ext4/super.c
   4  *
   5  * Copyright (C) 1992, 1993, 1994, 1995
   6  * Remy Card ([email protected])
   7  * Laboratoire MASI - Institut Blaise Pascal
   8  * Universite Pierre et Marie Curie (Paris VI)
   9  *
  10  *  from
  11  *
  12  *  linux/fs/minix/inode.c
  13  *
  14  *  Copyright (C) 1991, 1992  Linus Torvalds
  15  *
  16  *  Big-endian to little-endian byte-swapping/bitmaps by
  17  *        David S. Miller ([email protected]), 1995
  18  */
  19
  20 #include <linux/module.h>
  21 #include <linux/string.h>
  22 #include <linux/fs.h>
  23 #include <linux/time.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/backing-dev.h>
  29 #include <linux/parser.h>
  30 #include <linux/buffer_head.h>
  31 #include <linux/exportfs.h>
  32 #include <linux/vfs.h>
  33 #include <linux/random.h>
  34 #include <linux/mount.h>
  35 #include <linux/namei.h>
  36 #include <linux/quotaops.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/dax.h>
  42 #include <linux/cleancache.h>
  43 #include <linux/uaccess.h>
  44 #include <linux/iversion.h>
  45 #include <linux/unicode.h>
  46 #include <linux/part_stat.h>
  47 #include <linux/kthread.h>
  48 #include <linux/freezer.h>
  49
  50 #include "ext4.h"
  51 #include "ext4_extents.h"       /* Needed for trace points definition */
  52 #include "ext4_jbd2.h"
  53 #include "xattr.h"
  54 #include "acl.h"
  55 #include "mballoc.h"
  56 #include "fsmap.h"
  57
  58 #define CREATE_TRACE_POINTS
  59 #include <trace/events/ext4.h>
  60
  61 static struct ext4_lazy_init *ext4_li_info;
  62 static struct mutex ext4_li_mtx;
  63 static struct ratelimit_state ext4_mount_msg_ratelimit;
  64
  65 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  66                              unsigned long journal_devnum);
  67 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  68 static int ext4_commit_super(struct super_block *sb, int sync);
  69 static void ext4_mark_recovery_complete(struct super_block *sb,
  70                                         struct ext4_super_block *es);
  71 static void ext4_clear_journal_err(struct super_block *sb,
  72                                    struct ext4_super_block *es);
  73 static int ext4_sync_fs(struct super_block *sb, int wait);
  74 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  75 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  76 static int ext4_unfreeze(struct super_block *sb);
  77 static int ext4_freeze(struct super_block *sb);
  78 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  79                        const char *dev_name, void *data);
  80 static inline int ext2_feature_set_ok(struct super_block *sb);
  81 static inline int ext3_feature_set_ok(struct super_block *sb);
  82 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  83 static void ext4_destroy_lazyinit_thread(void);
  84 static void ext4_unregister_li_request(struct super_block *sb);
  85 static void ext4_clear_request_list(void);
  86 static struct inode *ext4_get_journal_inode(struct super_block *sb,
  87                                             unsigned int journal_inum);
  88
  89 /*
  90  * Lock ordering
  91  *
  92  * Note the difference between i_mmap_sem (EXT4_I(inode)->i_mmap_sem) and
  93  * i_mmap_rwsem (inode->i_mmap_rwsem)!
  94  *
  95  * page fault path:
  96  * mmap_sem -> sb_start_pagefault -> i_mmap_sem (r) -> transaction start ->
  97  *   page lock -> i_data_sem (rw)
  98  *
  99  * buffered write path:
 100  * sb_start_write -> i_mutex -> mmap_sem
 101  * sb_start_write -> i_mutex -> transaction start -> page lock ->
 102  *   i_data_sem (rw)
 103  *
 104  * truncate:
 105  * sb_start_write -> i_mutex -> i_mmap_sem (w) -> i_mmap_rwsem (w) -> page lock
 106  * sb_start_write -> i_mutex -> i_mmap_sem (w) -> transaction start ->
 107  *   i_data_sem (rw)
 108  *
 109  * direct IO:
 110  * sb_start_write -> i_mutex -> mmap_sem
 111  * sb_start_write -> i_mutex -> transaction start -> i_data_sem (rw)
 112  *
 113  * writepages:
 114  * transaction start -> page lock(s) -> i_data_sem (rw)
 115  */
 116
 117 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
 118 static struct file_system_type ext2_fs_type = {
 119         .owner          = THIS_MODULE,
 120         .name           = "ext2",
 121         .mount          = ext4_mount,
 122         .kill_sb        = kill_block_super,
 123         .fs_flags       = FS_REQUIRES_DEV,
 124 };
 125 MODULE_ALIAS_FS("ext2");
 126 MODULE_ALIAS("ext2");
 127 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
 128 #else
 129 #define IS_EXT2_SB(sb) (0)
 130 #endif
 131
 132
 133 static struct file_system_type ext3_fs_type = {
 134         .owner          = THIS_MODULE,
 135         .name           = "ext3",
 136         .mount          = ext4_mount,
 137         .kill_sb        = kill_block_super,
 138         .fs_flags       = FS_REQUIRES_DEV,
 139 };
 140 MODULE_ALIAS_FS("ext3");
 141 MODULE_ALIAS("ext3");
 142 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 143
 144 /*
 145  * This works like sb_bread() except it uses ERR_PTR for error
 146  * returns.  Currently with sb_bread it's impossible to distinguish
 147  * between ENOMEM and EIO situations (since both result in a NULL
 148  * return.
 149  */
 150 struct buffer_head *
 151 ext4_sb_bread(struct super_block *sb, sector_t block, int op_flags)
 152 {
 153         struct buffer_head *bh = sb_getblk(sb, block);
 154
 155         if (bh == NULL)
 156                 return ERR_PTR(-ENOMEM);
 157         if (ext4_buffer_uptodate(bh))
 158                 return bh;
 159         ll_rw_block(REQ_OP_READ, REQ_META | op_flags, 1, &bh);
 160         wait_on_buffer(bh);
 161         if (buffer_uptodate(bh))
 162                 return bh;
 163         put_bh(bh);
 164         return ERR_PTR(-EIO);
 165 }
 166
 167 static int ext4_verify_csum_type(struct super_block *sb,
 168                                  struct ext4_super_block *es)
 169 {
 170         if (!ext4_has_feature_metadata_csum(sb))
 171                 return 1;
 172
 173         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 174 }
 175
 176 static __le32 ext4_superblock_csum(struct super_block *sb,
 177                                    struct ext4_super_block *es)
 178 {
 179         struct ext4_sb_info *sbi = EXT4_SB(sb);
 180         int offset = offsetof(struct ext4_super_block, s_checksum);
 181         __u32 csum;
 182
 183         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 184
 185         return cpu_to_le32(csum);
 186 }
 187
 188 static int ext4_superblock_csum_verify(struct super_block *sb,
 189                                        struct ext4_super_block *es)
 190 {
 191         if (!ext4_has_metadata_csum(sb))
 192                 return 1;
 193
 194         return es->s_checksum == ext4_superblock_csum(sb, es);
 195 }
 196
 197 void ext4_superblock_csum_set(struct super_block *sb)
 198 {
 199         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 200
 201         if (!ext4_has_metadata_csum(sb))
 202                 return;
 203
 204         es->s_checksum = ext4_superblock_csum(sb, es);
 205 }
 206
 207 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 208                                struct ext4_group_desc *bg)
 209 {
 210         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 211                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 212                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 213 }
 214
 215 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 216                                struct ext4_group_desc *bg)
 217 {
 218         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 219                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 220                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 221 }
 222
 223 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 224                               struct ext4_group_desc *bg)
 225 {
 226         return le32_to_cpu(bg->bg_inode_table_lo) |
 227                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 228                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 229 }
 230
 231 __u32 ext4_free_group_clusters(struct super_block *sb,
 232                                struct ext4_group_desc *bg)
 233 {
 234         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 235                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 236                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 237 }
 238
 239 __u32 ext4_free_inodes_count(struct super_block *sb,
 240                               struct ext4_group_desc *bg)
 241 {
 242         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 243                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 244                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 245 }
 246
 247 __u32 ext4_used_dirs_count(struct super_block *sb,
 248                               struct ext4_group_desc *bg)
 249 {
 250         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 251                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 252                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 253 }
 254
 255 __u32 ext4_itable_unused_count(struct super_block *sb,
 256                               struct ext4_group_desc *bg)
 257 {
 258         return le16_to_cpu(bg->bg_itable_unused_lo) |
 259                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 260                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 261 }
 262
 263 void ext4_block_bitmap_set(struct super_block *sb,
 264                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 265 {
 266         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 267         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 268                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 269 }
 270
 271 void ext4_inode_bitmap_set(struct super_block *sb,
 272                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 273 {
 274         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 275         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 276                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 277 }
 278
 279 void ext4_inode_table_set(struct super_block *sb,
 280                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 281 {
 282         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 283         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 284                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 285 }
 286
 287 void ext4_free_group_clusters_set(struct super_block *sb,
 288                                   struct ext4_group_desc *bg, __u32 count)
 289 {
 290         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 291         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 292                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 293 }
 294
 295 void ext4_free_inodes_set(struct super_block *sb,
 296                           struct ext4_group_desc *bg, __u32 count)
 297 {
 298         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 299         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 300                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 301 }
 302
 303 void ext4_used_dirs_set(struct super_block *sb,
 304                           struct ext4_group_desc *bg, __u32 count)
 305 {
 306         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 307         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 308                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 309 }
 310
 311 void ext4_itable_unused_set(struct super_block *sb,
 312                           struct ext4_group_desc *bg, __u32 count)
 313 {
 314         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 315         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 316                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 317 }
 318
 319 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi)
 320 {
 321         time64_t now = ktime_get_real_seconds();
 322
 323         now = clamp_val(now, 0, (1ull << 40) - 1);
 324
 325         *lo = cpu_to_le32(lower_32_bits(now));
 326         *hi = upper_32_bits(now);
 327 }
 328
 329 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi)
 330 {
 331         return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo);
 332 }
 333 #define ext4_update_tstamp(es, tstamp) \
 334         __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
 335 #define ext4_get_tstamp(es, tstamp) \
 336         __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
 337
 338 static void __save_error_info(struct super_block *sb, int error,
 339                               __u32 ino, __u64 block,
 340                               const char *func, unsigned int line)
 341 {
 342         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 343         int err;
 344
 345         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 346         if (bdev_read_only(sb->s_bdev))
 347                 return;
 348         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 349         ext4_update_tstamp(es, s_last_error_time);
 350         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 351         es->s_last_error_line = cpu_to_le32(line);
 352         es->s_last_error_ino = cpu_to_le32(ino);
 353         es->s_last_error_block = cpu_to_le64(block);
 354         switch (error) {
 355         case EIO:
 356                 err = EXT4_ERR_EIO;
 357                 break;
 358         case ENOMEM:
 359                 err = EXT4_ERR_ENOMEM;
 360                 break;
 361         case EFSBADCRC:
 362                 err = EXT4_ERR_EFSBADCRC;
 363                 break;
 364         case 0:
 365         case EFSCORRUPTED:
 366                 err = EXT4_ERR_EFSCORRUPTED;
 367                 break;
 368         case ENOSPC:
 369                 err = EXT4_ERR_ENOSPC;
 370                 break;
 371         case ENOKEY:
 372                 err = EXT4_ERR_ENOKEY;
 373                 break;
 374         case EROFS:
 375                 err = EXT4_ERR_EROFS;
 376                 break;
 377         case EFBIG:
 378                 err = EXT4_ERR_EFBIG;
 379                 break;
 380         case EEXIST:
 381                 err = EXT4_ERR_EEXIST;
 382                 break;
 383         case ERANGE:
 384                 err = EXT4_ERR_ERANGE;
 385                 break;
 386         case EOVERFLOW:
 387                 err = EXT4_ERR_EOVERFLOW;
 388                 break;
 389         case EBUSY:
 390                 err = EXT4_ERR_EBUSY;
 391                 break;
 392         case ENOTDIR:
 393                 err = EXT4_ERR_ENOTDIR;
 394                 break;
 395         case ENOTEMPTY:
 396                 err = EXT4_ERR_ENOTEMPTY;
 397                 break;
 398         case ESHUTDOWN:
 399                 err = EXT4_ERR_ESHUTDOWN;
 400                 break;
 401         case EFAULT:
 402                 err = EXT4_ERR_EFAULT;
 403                 break;
 404         default:
 405                 err = EXT4_ERR_UNKNOWN;
 406         }
 407         es->s_last_error_errcode = err;
 408         if (!es->s_first_error_time) {
 409                 es->s_first_error_time = es->s_last_error_time;
 410                 es->s_first_error_time_hi = es->s_last_error_time_hi;
 411                 strncpy(es->s_first_error_func, func,
 412                         sizeof(es->s_first_error_func));
 413                 es->s_first_error_line = cpu_to_le32(line);
 414                 es->s_first_error_ino = es->s_last_error_ino;
 415                 es->s_first_error_block = es->s_last_error_block;
 416                 es->s_first_error_errcode = es->s_last_error_errcode;
 417         }
 418         /*
 419          * Start the daily error reporting function if it hasn't been
 420          * started already
 421          */
 422         if (!es->s_error_count)
 423                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 424         le32_add_cpu(&es->s_error_count, 1);
 425 }
 426
 427 static void save_error_info(struct super_block *sb, int error,
 428                             __u32 ino, __u64 block,
 429                             const char *func, unsigned int line)
 430 {
 431         __save_error_info(sb, error, ino, block, func, line);
 432         if (!bdev_read_only(sb->s_bdev))
 433                 ext4_commit_super(sb, 1);
 434 }
 435
 436 /*
 437  * The del_gendisk() function uninitializes the disk-specific data
 438  * structures, including the bdi structure, without telling anyone
 439  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 440  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 441  * This is a kludge to prevent these oops until we can put in a proper
 442  * hook in del_gendisk() to inform the VFS and file system layers.
 443  */
 444 static int block_device_ejected(struct super_block *sb)
 445 {
 446         struct inode *bd_inode = sb->s_bdev->bd_inode;
 447         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 448
 449         return bdi->dev == NULL;
 450 }
 451
 452 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 453 {
 454         struct super_block              *sb = journal->j_private;
 455         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 456         int                             error = is_journal_aborted(journal);
 457         struct ext4_journal_cb_entry    *jce;
 458
 459         BUG_ON(txn->t_state == T_FINISHED);
 460
 461         ext4_process_freed_data(sb, txn->t_tid);
 462
 463         spin_lock(&sbi->s_md_lock);
 464         while (!list_empty(&txn->t_private_list)) {
 465                 jce = list_entry(txn->t_private_list.next,
 466                                  struct ext4_journal_cb_entry, jce_list);
 467                 list_del_init(&jce->jce_list);
 468                 spin_unlock(&sbi->s_md_lock);
 469                 jce->jce_func(sb, jce, error);
 470                 spin_lock(&sbi->s_md_lock);
 471         }
 472         spin_unlock(&sbi->s_md_lock);
 473 }
 474
 475 static bool system_going_down(void)
 476 {
 477         return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF
 478                 || system_state == SYSTEM_RESTART;
 479 }
 480
 481 /* Deal with the reporting of failure conditions on a filesystem such as
 482  * inconsistencies detected or read IO failures.
 483  *
 484  * On ext2, we can store the error state of the filesystem in the
 485  * superblock.  That is not possible on ext4, because we may have other
 486  * write ordering constraints on the superblock which prevent us from
 487  * writing it out straight away; and given that the journal is about to
 488  * be aborted, we can't rely on the current, or future, transactions to
 489  * write out the superblock safely.
 490  *
 491  * We'll just use the jbd2_journal_abort() error code to record an error in
 492  * the journal instead.  On recovery, the journal will complain about
 493  * that error until we've noted it down and cleared it.
 494  */
 495
 496 static void ext4_handle_error(struct super_block *sb)
 497 {
 498         if (test_opt(sb, WARN_ON_ERROR))
 499                 WARN_ON_ONCE(1);
 500
 501         if (sb_rdonly(sb))
 502                 return;
 503
 504         if (!test_opt(sb, ERRORS_CONT)) {
 505                 journal_t *journal = EXT4_SB(sb)->s_journal;
 506
 507                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 508                 if (journal)
 509                         jbd2_journal_abort(journal, -EIO);
 510         }
 511         /*
 512          * We force ERRORS_RO behavior when system is rebooting. Otherwise we
 513          * could panic during 'reboot -f' as the underlying device got already
 514          * disabled.
 515          */
 516         if (test_opt(sb, ERRORS_RO) || system_going_down()) {
 517                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 518                 /*
 519                  * Make sure updated value of ->s_mount_flags will be visible
 520                  * before ->s_flags update
 521                  */
 522                 smp_wmb();
 523                 sb->s_flags |= SB_RDONLY;
 524         } else if (test_opt(sb, ERRORS_PANIC)) {
 525                 if (EXT4_SB(sb)->s_journal &&
 526                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 527                         return;
 528                 panic("EXT4-fs (device %s): panic forced after error\n",
 529                         sb->s_id);
 530         }
 531 }
 532
 533 #define ext4_error_ratelimit(sb)                                        \
 534                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 535                              "EXT4-fs error")
 536
 537 void __ext4_error(struct super_block *sb, const char *function,
 538                   unsigned int line, int error, __u64 block,
 539                   const char *fmt, ...)
 540 {
 541         struct va_format vaf;
 542         va_list args;
 543
 544         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 545                 return;
 546
 547         trace_ext4_error(sb, function, line);
 548         if (ext4_error_ratelimit(sb)) {
 549                 va_start(args, fmt);
 550                 vaf.fmt = fmt;
 551                 vaf.va = &args;
 552                 printk(KERN_CRIT
 553                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 554                        sb->s_id, function, line, current->comm, &vaf);
 555                 va_end(args);
 556         }
 557         save_error_info(sb, error, 0, block, function, line);
 558         ext4_handle_error(sb);
 559 }
 560
 561 void __ext4_error_inode(struct inode *inode, const char *function,
 562                         unsigned int line, ext4_fsblk_t block, int error,
 563                         const char *fmt, ...)
 564 {
 565         va_list args;
 566         struct va_format vaf;
 567
 568         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 569                 return;
 570
 571         trace_ext4_error(inode->i_sb, function, line);
 572         if (ext4_error_ratelimit(inode->i_sb)) {
 573                 va_start(args, fmt);
 574                 vaf.fmt = fmt;
 575                 vaf.va = &args;
 576                 if (block)
 577                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 578                                "inode #%lu: block %llu: comm %s: %pV\n",
 579                                inode->i_sb->s_id, function, line, inode->i_ino,
 580                                block, current->comm, &vaf);
 581                 else
 582                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 583                                "inode #%lu: comm %s: %pV\n",
 584                                inode->i_sb->s_id, function, line, inode->i_ino,
 585                                current->comm, &vaf);
 586                 va_end(args);
 587         }
 588         save_error_info(inode->i_sb, error, inode->i_ino, block,
 589                         function, line);
 590         ext4_handle_error(inode->i_sb);
 591 }
 592
 593 void __ext4_error_file(struct file *file, const char *function,
 594                        unsigned int line, ext4_fsblk_t block,
 595                        const char *fmt, ...)
 596 {
 597         va_list args;
 598         struct va_format vaf;
 599         struct inode *inode = file_inode(file);
 600         char pathname[80], *path;
 601
 602         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 603                 return;
 604
 605         trace_ext4_error(inode->i_sb, function, line);
 606         if (ext4_error_ratelimit(inode->i_sb)) {
 607                 path = file_path(file, pathname, sizeof(pathname));
 608                 if (IS_ERR(path))
 609                         path = "(unknown)";
 610                 va_start(args, fmt);
 611                 vaf.fmt = fmt;
 612                 vaf.va = &args;
 613                 if (block)
 614                         printk(KERN_CRIT
 615                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 616                                "block %llu: comm %s: path %s: %pV\n",
 617                                inode->i_sb->s_id, function, line, inode->i_ino,
 618                                block, current->comm, path, &vaf);
 619                 else
 620                         printk(KERN_CRIT
 621                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 622                                "comm %s: path %s: %pV\n",
 623                                inode->i_sb->s_id, function, line, inode->i_ino,
 624                                current->comm, path, &vaf);
 625                 va_end(args);
 626         }
 627         save_error_info(inode->i_sb, EFSCORRUPTED, inode->i_ino, block,
 628                         function, line);
 629         ext4_handle_error(inode->i_sb);
 630 }
 631
 632 const char *ext4_decode_error(struct super_block *sb, int errno,
 633                               char nbuf[16])
 634 {
 635         char *errstr = NULL;
 636
 637         switch (errno) {
 638         case -EFSCORRUPTED:
 639                 errstr = "Corrupt filesystem";
 640                 break;
 641         case -EFSBADCRC:
 642                 errstr = "Filesystem failed CRC";
 643                 break;
 644         case -EIO:
 645                 errstr = "IO failure";
 646                 break;
 647         case -ENOMEM:
 648                 errstr = "Out of memory";
 649                 break;
 650         case -EROFS:
 651                 if (!sb || (EXT4_SB(sb)->s_journal &&
 652                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 653                         errstr = "Journal has aborted";
 654                 else
 655                         errstr = "Readonly filesystem";
 656                 break;
 657         default:
 658                 /* If the caller passed in an extra buffer for unknown
 659                  * errors, textualise them now.  Else we just return
 660                  * NULL. */
 661                 if (nbuf) {
 662                         /* Check for truncated error codes... */
 663                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 664                                 errstr = nbuf;
 665                 }
 666                 break;
 667         }
 668
 669         return errstr;
 670 }
 671
 672 /* __ext4_std_error decodes expected errors from journaling functions
 673  * automatically and invokes the appropriate error response.  */
 674
 675 void __ext4_std_error(struct super_block *sb, const char *function,
 676                       unsigned int line, int errno)
 677 {
 678         char nbuf[16];
 679         const char *errstr;
 680
 681         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 682                 return;
 683
 684         /* Special case: if the error is EROFS, and we're not already
 685          * inside a transaction, then there's really no point in logging
 686          * an error. */
 687         if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb))
 688                 return;
 689
 690         if (ext4_error_ratelimit(sb)) {
 691                 errstr = ext4_decode_error(sb, errno, nbuf);
 692                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 693                        sb->s_id, function, line, errstr);
 694         }
 695
 696         save_error_info(sb, -errno, 0, 0, function, line);
 697         ext4_handle_error(sb);
 698 }
 699
 700 /*
 701  * ext4_abort is a much stronger failure handler than ext4_error.  The
 702  * abort function may be used to deal with unrecoverable failures such
 703  * as journal IO errors or ENOMEM at a critical moment in log management.
 704  *
 705  * We unconditionally force the filesystem into an ABORT|READONLY state,
 706  * unless the error response on the fs has been set to panic in which
 707  * case we take the easy way out and panic immediately.
 708  */
 709
 710 void __ext4_abort(struct super_block *sb, const char *function,
 711                   unsigned int line, int error, const char *fmt, ...)
 712 {
 713         struct va_format vaf;
 714         va_list args;
 715
 716         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 717                 return;
 718
 719         save_error_info(sb, error, 0, 0, function, line);
 720         va_start(args, fmt);
 721         vaf.fmt = fmt;
 722         vaf.va = &args;
 723         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: %pV\n",
 724                sb->s_id, function, line, &vaf);
 725         va_end(args);
 726
 727         if (sb_rdonly(sb) == 0) {
 728                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 729                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 730                 /*
 731                  * Make sure updated value of ->s_mount_flags will be visible
 732                  * before ->s_flags update
 733                  */
 734                 smp_wmb();
 735                 sb->s_flags |= SB_RDONLY;
 736                 if (EXT4_SB(sb)->s_journal)
 737                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 738         }
 739         if (test_opt(sb, ERRORS_PANIC) && !system_going_down()) {
 740                 if (EXT4_SB(sb)->s_journal &&
 741                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 742                         return;
 743                 panic("EXT4-fs panic from previous error\n");
 744         }
 745 }
 746
 747 void __ext4_msg(struct super_block *sb,
 748                 const char *prefix, const char *fmt, ...)
 749 {
 750         struct va_format vaf;
 751         va_list args;
 752
 753         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 754                 return;
 755
 756         va_start(args, fmt);
 757         vaf.fmt = fmt;
 758         vaf.va = &args;
 759         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 760         va_end(args);
 761 }
 762
 763 #define ext4_warning_ratelimit(sb)                                      \
 764                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
 765                              "EXT4-fs warning")
 766
 767 void __ext4_warning(struct super_block *sb, const char *function,
 768                     unsigned int line, const char *fmt, ...)
 769 {
 770         struct va_format vaf;
 771         va_list args;
 772
 773         if (!ext4_warning_ratelimit(sb))
 774                 return;
 775
 776         va_start(args, fmt);
 777         vaf.fmt = fmt;
 778         vaf.va = &args;
 779         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 780                sb->s_id, function, line, &vaf);
 781         va_end(args);
 782 }
 783
 784 void __ext4_warning_inode(const struct inode *inode, const char *function,
 785                           unsigned int line, const char *fmt, ...)
 786 {
 787         struct va_format vaf;
 788         va_list args;
 789
 790         if (!ext4_warning_ratelimit(inode->i_sb))
 791                 return;
 792
 793         va_start(args, fmt);
 794         vaf.fmt = fmt;
 795         vaf.va = &args;
 796         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 797                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 798                function, line, inode->i_ino, current->comm, &vaf);
 799         va_end(args);
 800 }
 801
 802 void __ext4_grp_locked_error(const char *function, unsigned int line,
 803                              struct super_block *sb, ext4_group_t grp,
 804                              unsigned long ino, ext4_fsblk_t block,
 805                              const char *fmt, ...)
 806 __releases(bitlock)
 807 __acquires(bitlock)
 808 {
 809         struct va_format vaf;
 810         va_list args;
 811
 812         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 813                 return;
 814
 815         trace_ext4_error(sb, function, line);
 816         __save_error_info(sb, EFSCORRUPTED, ino, block, function, line);
 817
 818         if (ext4_error_ratelimit(sb)) {
 819                 va_start(args, fmt);
 820                 vaf.fmt = fmt;
 821                 vaf.va = &args;
 822                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 823                        sb->s_id, function, line, grp);
 824                 if (ino)
 825                         printk(KERN_CONT "inode %lu: ", ino);
 826                 if (block)
 827                         printk(KERN_CONT "block %llu:",
 828                                (unsigned long long) block);
 829                 printk(KERN_CONT "%pV\n", &vaf);
 830                 va_end(args);
 831         }
 832
 833         if (test_opt(sb, WARN_ON_ERROR))
 834                 WARN_ON_ONCE(1);
 835
 836         if (test_opt(sb, ERRORS_CONT)) {
 837                 ext4_commit_super(sb, 0);
 838                 return;
 839         }
 840
 841         ext4_unlock_group(sb, grp);
 842         ext4_commit_super(sb, 1);
 843         ext4_handle_error(sb);
 844         /*
 845          * We only get here in the ERRORS_RO case; relocking the group
 846          * may be dangerous, but nothing bad will happen since the
 847          * filesystem will have already been marked read/only and the
 848          * journal has been aborted.  We return 1 as a hint to callers
 849          * who might what to use the return value from
 850          * ext4_grp_locked_error() to distinguish between the
 851          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 852          * aggressively from the ext4 function in question, with a
 853          * more appropriate error code.
 854          */
 855         ext4_lock_group(sb, grp);
 856         return;
 857 }
 858
 859 void ext4_mark_group_bitmap_corrupted(struct super_block *sb,
 860                                      ext4_group_t group,
 861                                      unsigned int flags)
 862 {
 863         struct ext4_sb_info *sbi = EXT4_SB(sb);
 864         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
 865         struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
 866         int ret;
 867
 868         if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) {
 869                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT,
 870                                             &grp->bb_state);
 871                 if (!ret)
 872                         percpu_counter_sub(&sbi->s_freeclusters_counter,
 873                                            grp->bb_free);
 874         }
 875
 876         if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) {
 877                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT,
 878                                             &grp->bb_state);
 879                 if (!ret && gdp) {
 880                         int count;
 881
 882                         count = ext4_free_inodes_count(sb, gdp);
 883                         percpu_counter_sub(&sbi->s_freeinodes_counter,
 884                                            count);
 885                 }
 886         }
 887 }
 888
 889 void ext4_update_dynamic_rev(struct super_block *sb)
 890 {
 891         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 892
 893         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 894                 return;
 895
 896         ext4_warning(sb,
 897                      "updating to rev %d because of new feature flag, "
 898                      "running e2fsck is recommended",
 899                      EXT4_DYNAMIC_REV);
 900
 901         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 902         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 903         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 904         /* leave es->s_feature_*compat flags alone */
 905         /* es->s_uuid will be set by e2fsck if empty */
 906
 907         /*
 908          * The rest of the superblock fields should be zero, and if not it
 909          * means they are likely already in use, so leave them alone.  We
 910          * can leave it up to e2fsck to clean up any inconsistencies there.
 911          */
 912 }
 913
 914 /*
 915  * Open the external journal device
 916  */
 917 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 918 {
 919         struct block_device *bdev;
 920
 921         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 922         if (IS_ERR(bdev))
 923                 goto fail;
 924         return bdev;
 925
 926 fail:
 927         ext4_msg(sb, KERN_ERR,
 928                  "failed to open journal device unknown-block(%u,%u) %ld",
 929                  MAJOR(dev), MINOR(dev), PTR_ERR(bdev));
 930         return NULL;
 931 }
 932
 933 /*
 934  * Release the journal device
 935  */
 936 static void ext4_blkdev_put(struct block_device *bdev)
 937 {
 938         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 939 }
 940
 941 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 942 {
 943         struct block_device *bdev;
 944         bdev = sbi->journal_bdev;
 945         if (bdev) {
 946                 ext4_blkdev_put(bdev);
 947                 sbi->journal_bdev = NULL;
 948         }
 949 }
 950
 951 static inline struct inode *orphan_list_entry(struct list_head *l)
 952 {
 953         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 954 }
 955
 956 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 957 {
 958         struct list_head *l;
 959
 960         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 961                  le32_to_cpu(sbi->s_es->s_last_orphan));
 962
 963         printk(KERN_ERR "sb_info orphan list:\n");
 964         list_for_each(l, &sbi->s_orphan) {
 965                 struct inode *inode = orphan_list_entry(l);
 966                 printk(KERN_ERR "  "
 967                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 968                        inode->i_sb->s_id, inode->i_ino, inode,
 969                        inode->i_mode, inode->i_nlink,
 970                        NEXT_ORPHAN(inode));
 971         }
 972 }
 973
 974 #ifdef CONFIG_QUOTA
 975 static int ext4_quota_off(struct super_block *sb, int type);
 976
 977 static inline void ext4_quota_off_umount(struct super_block *sb)
 978 {
 979         int type;
 980
 981         /* Use our quota_off function to clear inode flags etc. */
 982         for (type = 0; type < EXT4_MAXQUOTAS; type++)
 983                 ext4_quota_off(sb, type);
 984 }
 985
 986 /*
 987  * This is a helper function which is used in the mount/remount
 988  * codepaths (which holds s_umount) to fetch the quota file name.
 989  */
 990 static inline char *get_qf_name(struct super_block *sb,
 991                                 struct ext4_sb_info *sbi,
 992                                 int type)
 993 {
 994         return rcu_dereference_protected(sbi->s_qf_names[type],
 995                                          lockdep_is_held(&sb->s_umount));
 996 }
 997 #else
 998 static inline void ext4_quota_off_umount(struct super_block *sb)
 999 {
1000 }
1001 #endif
1002
1003 static void ext4_put_super(struct super_block *sb)
1004 {
1005         struct ext4_sb_info *sbi = EXT4_SB(sb);
1006         struct ext4_super_block *es = sbi->s_es;
1007         struct buffer_head **group_desc;
1008         struct flex_groups **flex_groups;
1009         int aborted = 0;
1010         int i, err;
1011
1012         ext4_unregister_li_request(sb);
1013         ext4_quota_off_umount(sb);
1014
1015         destroy_workqueue(sbi->rsv_conversion_wq);
1016
1017         /*
1018          * Unregister sysfs before destroying jbd2 journal.
1019          * Since we could still access attr_journal_task attribute via sysfs
1020          * path which could have sbi->s_journal->j_task as NULL
1021          */
1022         ext4_unregister_sysfs(sb);
1023
1024         if (sbi->s_journal) {
1025                 aborted = is_journal_aborted(sbi->s_journal);
1026                 err = jbd2_journal_destroy(sbi->s_journal);
1027                 sbi->s_journal = NULL;
1028                 if ((err < 0) && !aborted) {
1029                         ext4_abort(sb, -err, "Couldn't clean up the journal");
1030                 }
1031         }
1032
1033         ext4_es_unregister_shrinker(sbi);
1034         del_timer_sync(&sbi->s_err_report);
1035         ext4_release_system_zone(sb);
1036         ext4_mb_release(sb);
1037         ext4_ext_release(sb);
1038
1039         if (!sb_rdonly(sb) && !aborted) {
1040                 ext4_clear_feature_journal_needs_recovery(sb);
1041                 es->s_state = cpu_to_le16(sbi->s_mount_state);
1042         }
1043         if (!sb_rdonly(sb))
1044                 ext4_commit_super(sb, 1);
1045
1046         rcu_read_lock();
1047         group_desc = rcu_dereference(sbi->s_group_desc);
1048         for (i = 0; i < sbi->s_gdb_count; i++)
1049                 brelse(group_desc[i]);
1050         kvfree(group_desc);
1051         flex_groups = rcu_dereference(sbi->s_flex_groups);
1052         if (flex_groups) {
1053                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
1054                         kvfree(flex_groups[i]);
1055                 kvfree(flex_groups);
1056         }
1057         rcu_read_unlock();
1058         percpu_counter_destroy(&sbi->s_freeclusters_counter);
1059         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1060         percpu_counter_destroy(&sbi->s_dirs_counter);
1061         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
1062         percpu_free_rwsem(&sbi->s_writepages_rwsem);
1063 #ifdef CONFIG_QUOTA
1064         for (i = 0; i < EXT4_MAXQUOTAS; i++)
1065                 kfree(get_qf_name(sb, sbi, i));
1066 #endif
1067
1068         /* Debugging code just in case the in-memory inode orphan list
1069          * isn't empty.  The on-disk one can be non-empty if we've
1070          * detected an error and taken the fs readonly, but the
1071          * in-memory list had better be clean by this point. */
1072         if (!list_empty(&sbi->s_orphan))
1073                 dump_orphan_list(sb, sbi);
1074         J_ASSERT(list_empty(&sbi->s_orphan));
1075
1076         sync_blockdev(sb->s_bdev);
1077         invalidate_bdev(sb->s_bdev);
1078         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
1079                 /*
1080                  * Invalidate the journal device's buffers.  We don't want them
1081                  * floating about in memory - the physical journal device may
1082                  * hotswapped, and it breaks the `ro-after' testing code.
1083                  */
1084                 sync_blockdev(sbi->journal_bdev);
1085                 invalidate_bdev(sbi->journal_bdev);
1086                 ext4_blkdev_remove(sbi);
1087         }
1088
1089         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
1090         sbi->s_ea_inode_cache = NULL;
1091
1092         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
1093         sbi->s_ea_block_cache = NULL;
1094
1095         if (sbi->s_mmp_tsk)
1096                 kthread_stop(sbi->s_mmp_tsk);
1097         brelse(sbi->s_sbh);
1098         sb->s_fs_info = NULL;
1099         /*
1100          * Now that we are completely done shutting down the
1101          * superblock, we need to actually destroy the kobject.
1102          */
1103         kobject_put(&sbi->s_kobj);
1104         wait_for_completion(&sbi->s_kobj_unregister);
1105         if (sbi->s_chksum_driver)
1106                 crypto_free_shash(sbi->s_chksum_driver);
1107         kfree(sbi->s_blockgroup_lock);
1108         fs_put_dax(sbi->s_daxdev);
1109 #ifdef CONFIG_UNICODE
1110         utf8_unload(sbi->s_encoding);
1111 #endif
1112         kfree(sbi);
1113 }
1114
1115 static struct kmem_cache *ext4_inode_cachep;
1116
1117 /*
1118  * Called inside transaction, so use GFP_NOFS
1119  */
1120 static struct inode *ext4_alloc_inode(struct super_block *sb)
1121 {
1122         struct ext4_inode_info *ei;
1123
1124         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
1125         if (!ei)
1126                 return NULL;
1127
1128         inode_set_iversion(&ei->vfs_inode, 1);
1129         spin_lock_init(&ei->i_raw_lock);
1130         INIT_LIST_HEAD(&ei->i_prealloc_list);
1131         spin_lock_init(&ei->i_prealloc_lock);
1132         ext4_es_init_tree(&ei->i_es_tree);
1133         rwlock_init(&ei->i_es_lock);
1134         INIT_LIST_HEAD(&ei->i_es_list);
1135         ei->i_es_all_nr = 0;
1136         ei->i_es_shk_nr = 0;
1137         ei->i_es_shrink_lblk = 0;
1138         ei->i_reserved_data_blocks = 0;
1139         spin_lock_init(&(ei->i_block_reservation_lock));
1140         ext4_init_pending_tree(&ei->i_pending_tree);
1141 #ifdef CONFIG_QUOTA
1142         ei->i_reserved_quota = 0;
1143         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
1144 #endif
1145         ei->jinode = NULL;
1146         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
1147         spin_lock_init(&ei->i_completed_io_lock);
1148         ei->i_sync_tid = 0;
1149         ei->i_datasync_tid = 0;
1150         atomic_set(&ei->i_unwritten, 0);
1151         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
1152         return &ei->vfs_inode;
1153 }
1154
1155 static int ext4_drop_inode(struct inode *inode)
1156 {
1157         int drop = generic_drop_inode(inode);
1158
1159         if (!drop)
1160                 drop = fscrypt_drop_inode(inode);
1161
1162         trace_ext4_drop_inode(inode, drop);
1163         return drop;
1164 }
1165
1166 static void ext4_free_in_core_inode(struct inode *inode)
1167 {
1168         fscrypt_free_inode(inode);
1169         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1170 }
1171
1172 static void ext4_destroy_inode(struct inode *inode)
1173 {
1174         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1175                 ext4_msg(inode->i_sb, KERN_ERR,
1176                          "Inode %lu (%p): orphan list check failed!",
1177                          inode->i_ino, EXT4_I(inode));
1178                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1179                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1180                                 true);
1181                 dump_stack();
1182         }
1183 }
1184
1185 static void init_once(void *foo)
1186 {
1187         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1188
1189         INIT_LIST_HEAD(&ei->i_orphan);
1190         init_rwsem(&ei->xattr_sem);
1191         init_rwsem(&ei->i_data_sem);
1192         init_rwsem(&ei->i_mmap_sem);
1193         inode_init_once(&ei->vfs_inode);
1194 }
1195
1196 static int __init init_inodecache(void)
1197 {
1198         ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache",
1199                                 sizeof(struct ext4_inode_info), 0,
1200                                 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
1201                                         SLAB_ACCOUNT),
1202                                 offsetof(struct ext4_inode_info, i_data),
1203                                 sizeof_field(struct ext4_inode_info, i_data),
1204                                 init_once);
1205         if (ext4_inode_cachep == NULL)
1206                 return -ENOMEM;
1207         return 0;
1208 }
1209
1210 static void destroy_inodecache(void)
1211 {
1212         /*
1213          * Make sure all delayed rcu free inodes are flushed before we
1214          * destroy cache.
1215          */
1216         rcu_barrier();
1217         kmem_cache_destroy(ext4_inode_cachep);
1218 }
1219
1220 void ext4_clear_inode(struct inode *inode)
1221 {
1222         invalidate_inode_buffers(inode);
1223         clear_inode(inode);
1224         ext4_discard_preallocations(inode);
1225         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1226         dquot_drop(inode);
1227         if (EXT4_I(inode)->jinode) {
1228                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1229                                                EXT4_I(inode)->jinode);
1230                 jbd2_free_inode(EXT4_I(inode)->jinode);
1231                 EXT4_I(inode)->jinode = NULL;
1232         }
1233         fscrypt_put_encryption_info(inode);
1234         fsverity_cleanup_inode(inode);
1235 }
1236
1237 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1238                                         u64 ino, u32 generation)
1239 {
1240         struct inode *inode;
1241
1242         /*
1243          * Currently we don't know the generation for parent directory, so
1244          * a generation of 0 means "accept any"
1245          */
1246         inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE);
1247         if (IS_ERR(inode))
1248                 return ERR_CAST(inode);
1249         if (generation && inode->i_generation != generation) {
1250                 iput(inode);
1251                 return ERR_PTR(-ESTALE);
1252         }
1253
1254         return inode;
1255 }
1256
1257 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1258                                         int fh_len, int fh_type)
1259 {
1260         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1261                                     ext4_nfs_get_inode);
1262 }
1263
1264 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1265                                         int fh_len, int fh_type)
1266 {
1267         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1268                                     ext4_nfs_get_inode);
1269 }
1270
1271 static int ext4_nfs_commit_metadata(struct inode *inode)
1272 {
1273         struct writeback_control wbc = {
1274                 .sync_mode = WB_SYNC_ALL
1275         };
1276
1277         trace_ext4_nfs_commit_metadata(inode);
1278         return ext4_write_inode(inode, &wbc);
1279 }
1280
1281 /*
1282  * Try to release metadata pages (indirect blocks, directories) which are
1283  * mapped via the block device.  Since these pages could have journal heads
1284  * which would prevent try_to_free_buffers() from freeing them, we must use
1285  * jbd2 layer's try_to_free_buffers() function to release them.
1286  */
1287 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1288                                  gfp_t wait)
1289 {
1290         journal_t *journal = EXT4_SB(sb)->s_journal;
1291
1292         WARN_ON(PageChecked(page));
1293         if (!page_has_buffers(page))
1294                 return 0;
1295         if (journal)
1296                 return jbd2_journal_try_to_free_buffers(journal, page,
1297                                                 wait & ~__GFP_DIRECT_RECLAIM);
1298         return try_to_free_buffers(page);
1299 }
1300
1301 #ifdef CONFIG_FS_ENCRYPTION
1302 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1303 {
1304         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1305                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1306 }
1307
1308 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1309                                                         void *fs_data)
1310 {
1311         handle_t *handle = fs_data;
1312         int res, res2, credits, retries = 0;
1313
1314         /*
1315          * Encrypting the root directory is not allowed because e2fsck expects
1316          * lost+found to exist and be unencrypted, and encrypting the root
1317          * directory would imply encrypting the lost+found directory as well as
1318          * the filename "lost+found" itself.
1319          */
1320         if (inode->i_ino == EXT4_ROOT_INO)
1321                 return -EPERM;
1322
1323         if (WARN_ON_ONCE(IS_DAX(inode) && i_size_read(inode)))
1324                 return -EINVAL;
1325
1326         res = ext4_convert_inline_data(inode);
1327         if (res)
1328                 return res;
1329
1330         /*
1331          * If a journal handle was specified, then the encryption context is
1332          * being set on a new inode via inheritance and is part of a larger
1333          * transaction to create the inode.  Otherwise the encryption context is
1334          * being set on an existing inode in its own transaction.  Only in the
1335          * latter case should the "retry on ENOSPC" logic be used.
1336          */
1337
1338         if (handle) {
1339                 res = ext4_xattr_set_handle(handle, inode,
1340                                             EXT4_XATTR_INDEX_ENCRYPTION,
1341                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1342                                             ctx, len, 0);
1343                 if (!res) {
1344                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1345                         ext4_clear_inode_state(inode,
1346                                         EXT4_STATE_MAY_INLINE_DATA);
1347                         /*
1348                          * Update inode->i_flags - S_ENCRYPTED will be enabled,
1349                          * S_DAX may be disabled
1350                          */
1351                         ext4_set_inode_flags(inode);
1352                 }
1353                 return res;
1354         }
1355
1356         res = dquot_initialize(inode);
1357         if (res)
1358                 return res;
1359 retry:
1360         res = ext4_xattr_set_credits(inode, len, false /* is_create */,
1361                                      &credits);
1362         if (res)
1363                 return res;
1364
1365         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
1366         if (IS_ERR(handle))
1367                 return PTR_ERR(handle);
1368
1369         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1370                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1371                                     ctx, len, 0);
1372         if (!res) {
1373                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1374                 /*
1375                  * Update inode->i_flags - S_ENCRYPTED will be enabled,
1376                  * S_DAX may be disabled
1377                  */
1378                 ext4_set_inode_flags(inode);
1379                 res = ext4_mark_inode_dirty(handle, inode);
1380                 if (res)
1381                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1382         }
1383         res2 = ext4_journal_stop(handle);
1384
1385         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1386                 goto retry;
1387         if (!res)
1388                 res = res2;
1389         return res;
1390 }
1391
1392 static bool ext4_dummy_context(struct inode *inode)
1393 {
1394         return DUMMY_ENCRYPTION_ENABLED(EXT4_SB(inode->i_sb));
1395 }
1396
1397 static bool ext4_has_stable_inodes(struct super_block *sb)
1398 {
1399         return ext4_has_feature_stable_inodes(sb);
1400 }
1401
1402 static void ext4_get_ino_and_lblk_bits(struct super_block *sb,
1403                                        int *ino_bits_ret, int *lblk_bits_ret)
1404 {
1405         *ino_bits_ret = 8 * sizeof(EXT4_SB(sb)->s_es->s_inodes_count);
1406         *lblk_bits_ret = 8 * sizeof(ext4_lblk_t);
1407 }
1408
1409 static const struct fscrypt_operations ext4_cryptops = {
1410         .key_prefix             = "ext4:",
1411         .get_context            = ext4_get_context,
1412         .set_context            = ext4_set_context,
1413         .dummy_context          = ext4_dummy_context,
1414         .empty_dir              = ext4_empty_dir,
1415         .max_namelen            = EXT4_NAME_LEN,
1416         .has_stable_inodes      = ext4_has_stable_inodes,
1417         .get_ino_and_lblk_bits  = ext4_get_ino_and_lblk_bits,
1418 };
1419 #endif
1420
1421 #ifdef CONFIG_QUOTA
1422 static const char * const quotatypes[] = INITQFNAMES;
1423 #define QTYPE2NAME(t) (quotatypes[t])
1424
1425 static int ext4_write_dquot(struct dquot *dquot);
1426 static int ext4_acquire_dquot(struct dquot *dquot);
1427 static int ext4_release_dquot(struct dquot *dquot);
1428 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1429 static int ext4_write_info(struct super_block *sb, int type);
1430 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1431                          const struct path *path);
1432 static int ext4_quota_on_mount(struct super_block *sb, int type);
1433 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1434                                size_t len, loff_t off);
1435 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1436                                 const char *data, size_t len, loff_t off);
1437 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1438                              unsigned int flags);
1439 static int ext4_enable_quotas(struct super_block *sb);
1440
1441 static struct dquot **ext4_get_dquots(struct inode *inode)
1442 {
1443         return EXT4_I(inode)->i_dquot;
1444 }
1445
1446 static const struct dquot_operations ext4_quota_operations = {
1447         .get_reserved_space     = ext4_get_reserved_space,
1448         .write_dquot            = ext4_write_dquot,
1449         .acquire_dquot          = ext4_acquire_dquot,
1450         .release_dquot          = ext4_release_dquot,
1451         .mark_dirty             = ext4_mark_dquot_dirty,
1452         .write_info             = ext4_write_info,
1453         .alloc_dquot            = dquot_alloc,
1454         .destroy_dquot          = dquot_destroy,
1455         .get_projid             = ext4_get_projid,
1456         .get_inode_usage        = ext4_get_inode_usage,
1457         .get_next_id            = dquot_get_next_id,
1458 };
1459
1460 static const struct quotactl_ops ext4_qctl_operations = {
1461         .quota_on       = ext4_quota_on,
1462         .quota_off      = ext4_quota_off,
1463         .quota_sync     = dquot_quota_sync,
1464         .get_state      = dquot_get_state,
1465         .set_info       = dquot_set_dqinfo,
1466         .get_dqblk      = dquot_get_dqblk,
1467         .set_dqblk      = dquot_set_dqblk,
1468         .get_nextdqblk  = dquot_get_next_dqblk,
1469 };
1470 #endif
1471
1472 static const struct super_operations ext4_sops = {
1473         .alloc_inode    = ext4_alloc_inode,
1474         .free_inode     = ext4_free_in_core_inode,
1475         .destroy_inode  = ext4_destroy_inode,
1476         .write_inode    = ext4_write_inode,
1477         .dirty_inode    = ext4_dirty_inode,
1478         .drop_inode     = ext4_drop_inode,
1479         .evict_inode    = ext4_evict_inode,
1480         .put_super      = ext4_put_super,
1481         .sync_fs        = ext4_sync_fs,
1482         .freeze_fs      = ext4_freeze,
1483         .unfreeze_fs    = ext4_unfreeze,
1484         .statfs         = ext4_statfs,
1485         .remount_fs     = ext4_remount,
1486         .show_options   = ext4_show_options,
1487 #ifdef CONFIG_QUOTA
1488         .quota_read     = ext4_quota_read,
1489         .quota_write    = ext4_quota_write,
1490         .get_dquots     = ext4_get_dquots,
1491 #endif
1492         .bdev_try_to_free_page = bdev_try_to_free_page,
1493 };
1494
1495 static const struct export_operations ext4_export_ops = {
1496         .fh_to_dentry = ext4_fh_to_dentry,
1497         .fh_to_parent = ext4_fh_to_parent,
1498         .get_parent = ext4_get_parent,
1499         .commit_metadata = ext4_nfs_commit_metadata,
1500 };
1501
1502 enum {
1503         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1504         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1505         Opt_nouid32, Opt_debug, Opt_removed,
1506         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1507         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1508         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1509         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1510         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1511         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1512         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1513         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1514         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1515         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, Opt_dax,
1516         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
1517         Opt_nowarn_on_error, Opt_mblk_io_submit,
1518         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1519         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1520         Opt_inode_readahead_blks, Opt_journal_ioprio,
1521         Opt_dioread_nolock, Opt_dioread_lock,
1522         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1523         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1524 };
1525
1526 static const match_table_t tokens = {
1527         {Opt_bsd_df, "bsddf"},
1528         {Opt_minix_df, "minixdf"},
1529         {Opt_grpid, "grpid"},
1530         {Opt_grpid, "bsdgroups"},
1531         {Opt_nogrpid, "nogrpid"},
1532         {Opt_nogrpid, "sysvgroups"},
1533         {Opt_resgid, "resgid=%u"},
1534         {Opt_resuid, "resuid=%u"},
1535         {Opt_sb, "sb=%u"},
1536         {Opt_err_cont, "errors=continue"},
1537         {Opt_err_panic, "errors=panic"},
1538         {Opt_err_ro, "errors=remount-ro"},
1539         {Opt_nouid32, "nouid32"},
1540         {Opt_debug, "debug"},
1541         {Opt_removed, "oldalloc"},
1542         {Opt_removed, "orlov"},
1543         {Opt_user_xattr, "user_xattr"},
1544         {Opt_nouser_xattr, "nouser_xattr"},
1545         {Opt_acl, "acl"},
1546         {Opt_noacl, "noacl"},
1547         {Opt_noload, "norecovery"},
1548         {Opt_noload, "noload"},
1549         {Opt_removed, "nobh"},
1550         {Opt_removed, "bh"},
1551         {Opt_commit, "commit=%u"},
1552         {Opt_min_batch_time, "min_batch_time=%u"},
1553         {Opt_max_batch_time, "max_batch_time=%u"},
1554         {Opt_journal_dev, "journal_dev=%u"},
1555         {Opt_journal_path, "journal_path=%s"},
1556         {Opt_journal_checksum, "journal_checksum"},
1557         {Opt_nojournal_checksum, "nojournal_checksum"},
1558         {Opt_journal_async_commit, "journal_async_commit"},
1559         {Opt_abort, "abort"},
1560         {Opt_data_journal, "data=journal"},
1561         {Opt_data_ordered, "data=ordered"},
1562         {Opt_data_writeback, "data=writeback"},
1563         {Opt_data_err_abort, "data_err=abort"},
1564         {Opt_data_err_ignore, "data_err=ignore"},
1565         {Opt_offusrjquota, "usrjquota="},
1566         {Opt_usrjquota, "usrjquota=%s"},
1567         {Opt_offgrpjquota, "grpjquota="},
1568         {Opt_grpjquota, "grpjquota=%s"},
1569         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1570         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1571         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1572         {Opt_grpquota, "grpquota"},
1573         {Opt_noquota, "noquota"},
1574         {Opt_quota, "quota"},
1575         {Opt_usrquota, "usrquota"},
1576         {Opt_prjquota, "prjquota"},
1577         {Opt_barrier, "barrier=%u"},
1578         {Opt_barrier, "barrier"},
1579         {Opt_nobarrier, "nobarrier"},
1580         {Opt_i_version, "i_version"},
1581         {Opt_dax, "dax"},
1582         {Opt_stripe, "stripe=%u"},
1583         {Opt_delalloc, "delalloc"},
1584         {Opt_warn_on_error, "warn_on_error"},
1585         {Opt_nowarn_on_error, "nowarn_on_error"},
1586         {Opt_lazytime, "lazytime"},
1587         {Opt_nolazytime, "nolazytime"},
1588         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1589         {Opt_nodelalloc, "nodelalloc"},
1590         {Opt_removed, "mblk_io_submit"},
1591         {Opt_removed, "nomblk_io_submit"},
1592         {Opt_block_validity, "block_validity"},
1593         {Opt_noblock_validity, "noblock_validity"},
1594         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1595         {Opt_journal_ioprio, "journal_ioprio=%u"},
1596         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1597         {Opt_auto_da_alloc, "auto_da_alloc"},
1598         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1599         {Opt_dioread_nolock, "dioread_nolock"},
1600         {Opt_dioread_lock, "nodioread_nolock"},
1601         {Opt_dioread_lock, "dioread_lock"},
1602         {Opt_discard, "discard"},
1603         {Opt_nodiscard, "nodiscard"},
1604         {Opt_init_itable, "init_itable=%u"},
1605         {Opt_init_itable, "init_itable"},
1606         {Opt_noinit_itable, "noinit_itable"},
1607         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1608         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1609         {Opt_nombcache, "nombcache"},
1610         {Opt_nombcache, "no_mbcache"},  /* for backward compatibility */
1611         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1612         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1613         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1614         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1615         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1616         {Opt_err, NULL},
1617 };
1618
1619 static ext4_fsblk_t get_sb_block(void **data)
1620 {
1621         ext4_fsblk_t    sb_block;
1622         char            *options = (char *) *data;
1623
1624         if (!options || strncmp(options, "sb=", 3) != 0)
1625                 return 1;       /* Default location */
1626
1627         options += 3;
1628         /* TODO: use simple_strtoll with >32bit ext4 */
1629         sb_block = simple_strtoul(options, &options, 0);
1630         if (*options && *options != ',') {
1631                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1632                        (char *) *data);
1633                 return 1;
1634         }
1635         if (*options == ',')
1636                 options++;
1637         *data = (void *) options;
1638
1639         return sb_block;
1640 }
1641
1642 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1643 static const char deprecated_msg[] =
1644         "Mount option \"%s\" will be removed by %s\n"
1645         "Contact [email protected] if you think we should keep it.\n";
1646
1647 #ifdef CONFIG_QUOTA
1648 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1649 {
1650         struct ext4_sb_info *sbi = EXT4_SB(sb);
1651         char *qname, *old_qname = get_qf_name(sb, sbi, qtype);
1652         int ret = -1;
1653
1654         if (sb_any_quota_loaded(sb) && !old_qname) {
1655                 ext4_msg(sb, KERN_ERR,
1656                         "Cannot change journaled "
1657                         "quota options when quota turned on");
1658                 return -1;
1659         }
1660         if (ext4_has_feature_quota(sb)) {
1661                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1662                          "ignored when QUOTA feature is enabled");
1663                 return 1;
1664         }
1665         qname = match_strdup(args);
1666         if (!qname) {
1667                 ext4_msg(sb, KERN_ERR,
1668                         "Not enough memory for storing quotafile name");
1669                 return -1;
1670         }
1671         if (old_qname) {
1672                 if (strcmp(old_qname, qname) == 0)
1673                         ret = 1;
1674                 else
1675                         ext4_msg(sb, KERN_ERR,
1676                                  "%s quota file already specified",
1677                                  QTYPE2NAME(qtype));
1678                 goto errout;
1679         }
1680         if (strchr(qname, '/')) {
1681                 ext4_msg(sb, KERN_ERR,
1682                         "quotafile must be on filesystem root");
1683                 goto errout;
1684         }
1685         rcu_assign_pointer(sbi->s_qf_names[qtype], qname);
1686         set_opt(sb, QUOTA);
1687         return 1;
1688 errout:
1689         kfree(qname);
1690         return ret;
1691 }
1692
1693 static int clear_qf_name(struct super_block *sb, int qtype)
1694 {
1695
1696         struct ext4_sb_info *sbi = EXT4_SB(sb);
1697         char *old_qname = get_qf_name(sb, sbi, qtype);
1698
1699         if (sb_any_quota_loaded(sb) && old_qname) {
1700                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1701                         " when quota turned on");
1702                 return -1;
1703         }
1704         rcu_assign_pointer(sbi->s_qf_names[qtype], NULL);
1705         synchronize_rcu();
1706         kfree(old_qname);
1707         return 1;
1708 }
1709 #endif
1710
1711 #define MOPT_SET        0x0001
1712 #define MOPT_CLEAR      0x0002
1713 #define MOPT_NOSUPPORT  0x0004
1714 #define MOPT_EXPLICIT   0x0008
1715 #define MOPT_CLEAR_ERR  0x0010
1716 #define MOPT_GTE0       0x0020
1717 #ifdef CONFIG_QUOTA
1718 #define MOPT_Q          0
1719 #define MOPT_QFMT       0x0040
1720 #else
1721 #define MOPT_Q          MOPT_NOSUPPORT
1722 #define MOPT_QFMT       MOPT_NOSUPPORT
1723 #endif
1724 #define MOPT_DATAJ      0x0080
1725 #define MOPT_NO_EXT2    0x0100
1726 #define MOPT_NO_EXT3    0x0200
1727 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1728 #define MOPT_STRING     0x0400
1729
1730 static const struct mount_opts {
1731         int     token;
1732         int     mount_opt;
1733         int     flags;
1734 } ext4_mount_opts[] = {
1735         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1736         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1737         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1738         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1739         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1740         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1741         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1742          MOPT_EXT4_ONLY | MOPT_SET},
1743         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1744          MOPT_EXT4_ONLY | MOPT_CLEAR},
1745         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1746         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1747         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1748          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1749         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1750          MOPT_EXT4_ONLY | MOPT_CLEAR},
1751         {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET},
1752         {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR},
1753         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1754          MOPT_EXT4_ONLY | MOPT_CLEAR},
1755         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1756          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1757         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1758                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1759          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1760         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1761         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1762         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1763         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1764         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1765          MOPT_NO_EXT2},
1766         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1767          MOPT_NO_EXT2},
1768         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1769         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1770         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1771         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1772         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1773         {Opt_commit, 0, MOPT_GTE0},
1774         {Opt_max_batch_time, 0, MOPT_GTE0},
1775         {Opt_min_batch_time, 0, MOPT_GTE0},
1776         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1777         {Opt_init_itable, 0, MOPT_GTE0},
1778         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1779         {Opt_stripe, 0, MOPT_GTE0},
1780         {Opt_resuid, 0, MOPT_GTE0},
1781         {Opt_resgid, 0, MOPT_GTE0},
1782         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1783         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1784         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1785         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1786         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1787         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1788          MOPT_NO_EXT2 | MOPT_DATAJ},
1789         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1790         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1791 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1792         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1793         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1794 #else
1795         {Opt_acl, 0, MOPT_NOSUPPORT},
1796         {Opt_noacl, 0, MOPT_NOSUPPORT},
1797 #endif
1798         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1799         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1800         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1801         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1802         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1803                                                         MOPT_SET | MOPT_Q},
1804         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1805                                                         MOPT_SET | MOPT_Q},
1806         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1807                                                         MOPT_SET | MOPT_Q},
1808         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1809                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1810                                                         MOPT_CLEAR | MOPT_Q},
1811         {Opt_usrjquota, 0, MOPT_Q},
1812         {Opt_grpjquota, 0, MOPT_Q},
1813         {Opt_offusrjquota, 0, MOPT_Q},
1814         {Opt_offgrpjquota, 0, MOPT_Q},
1815         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1816         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1817         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1818         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1819         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1820         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1821         {Opt_err, 0, 0}
1822 };
1823
1824 #ifdef CONFIG_UNICODE
1825 static const struct ext4_sb_encodings {
1826         __u16 magic;
1827         char *name;
1828         char *version;
1829 } ext4_sb_encoding_map[] = {
1830         {EXT4_ENC_UTF8_12_1, "utf8", "12.1.0"},
1831 };
1832
1833 static int ext4_sb_read_encoding(const struct ext4_super_block *es,
1834                                  const struct ext4_sb_encodings **encoding,
1835                                  __u16 *flags)
1836 {
1837         __u16 magic = le16_to_cpu(es->s_encoding);
1838         int i;
1839
1840         for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++)
1841                 if (magic == ext4_sb_encoding_map[i].magic)
1842                         break;
1843
1844         if (i >= ARRAY_SIZE(ext4_sb_encoding_map))
1845                 return -EINVAL;
1846
1847         *encoding = &ext4_sb_encoding_map[i];
1848         *flags = le16_to_cpu(es->s_encoding_flags);
1849
1850         return 0;
1851 }
1852 #endif
1853
1854 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1855                             substring_t *args, unsigned long *journal_devnum,
1856                             unsigned int *journal_ioprio, int is_remount)
1857 {
1858         struct ext4_sb_info *sbi = EXT4_SB(sb);
1859         const struct mount_opts *m;
1860         kuid_t uid;
1861         kgid_t gid;
1862         int arg = 0;
1863
1864 #ifdef CONFIG_QUOTA
1865         if (token == Opt_usrjquota)
1866                 return set_qf_name(sb, USRQUOTA, &args[0]);
1867         else if (token == Opt_grpjquota)
1868                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1869         else if (token == Opt_offusrjquota)
1870                 return clear_qf_name(sb, USRQUOTA);
1871         else if (token == Opt_offgrpjquota)
1872                 return clear_qf_name(sb, GRPQUOTA);
1873 #endif
1874         switch (token) {
1875         case Opt_noacl:
1876         case Opt_nouser_xattr:
1877                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1878                 break;
1879         case Opt_sb:
1880                 return 1;       /* handled by get_sb_block() */
1881         case Opt_removed:
1882                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1883                 return 1;
1884         case Opt_abort:
1885                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1886                 return 1;
1887         case Opt_i_version:
1888                 sb->s_flags |= SB_I_VERSION;
1889                 return 1;
1890         case Opt_lazytime:
1891                 sb->s_flags |= SB_LAZYTIME;
1892                 return 1;
1893         case Opt_nolazytime:
1894                 sb->s_flags &= ~SB_LAZYTIME;
1895                 return 1;
1896         }
1897
1898         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1899                 if (token == m->token)
1900                         break;
1901
1902         if (m->token == Opt_err) {
1903                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1904                          "or missing value", opt);
1905                 return -1;
1906         }
1907
1908         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1909                 ext4_msg(sb, KERN_ERR,
1910                          "Mount option \"%s\" incompatible with ext2", opt);
1911                 return -1;
1912         }
1913         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1914                 ext4_msg(sb, KERN_ERR,
1915                          "Mount option \"%s\" incompatible with ext3", opt);
1916                 return -1;
1917         }
1918
1919         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1920                 return -1;
1921         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1922                 return -1;
1923         if (m->flags & MOPT_EXPLICIT) {
1924                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1925                         set_opt2(sb, EXPLICIT_DELALLOC);
1926                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1927                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1928                 } else
1929                         return -1;
1930         }
1931         if (m->flags & MOPT_CLEAR_ERR)
1932                 clear_opt(sb, ERRORS_MASK);
1933         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1934                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1935                          "options when quota turned on");
1936                 return -1;
1937         }
1938
1939         if (m->flags & MOPT_NOSUPPORT) {
1940                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1941         } else if (token == Opt_commit) {
1942                 if (arg == 0)
1943                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1944                 else if (arg > INT_MAX / HZ) {
1945                         ext4_msg(sb, KERN_ERR,
1946                                  "Invalid commit interval %d, "
1947                                  "must be smaller than %d",
1948                                  arg, INT_MAX / HZ);
1949                         return -1;
1950                 }
1951                 sbi->s_commit_interval = HZ * arg;
1952         } else if (token == Opt_debug_want_extra_isize) {
1953                 if ((arg & 1) ||
1954                     (arg < 4) ||
1955                     (arg > (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE))) {
1956                         ext4_msg(sb, KERN_ERR,
1957                                  "Invalid want_extra_isize %d", arg);
1958                         return -1;
1959                 }
1960                 sbi->s_want_extra_isize = arg;
1961         } else if (token == Opt_max_batch_time) {
1962                 sbi->s_max_batch_time = arg;
1963         } else if (token == Opt_min_batch_time) {
1964                 sbi->s_min_batch_time = arg;
1965         } else if (token == Opt_inode_readahead_blks) {
1966                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1967                         ext4_msg(sb, KERN_ERR,
1968                                  "EXT4-fs: inode_readahead_blks must be "
1969                                  "0 or a power of 2 smaller than 2^31");
1970                         return -1;
1971                 }
1972                 sbi->s_inode_readahead_blks = arg;
1973         } else if (token == Opt_init_itable) {
1974                 set_opt(sb, INIT_INODE_TABLE);
1975                 if (!args->from)
1976                         arg = EXT4_DEF_LI_WAIT_MULT;
1977                 sbi->s_li_wait_mult = arg;
1978         } else if (token == Opt_max_dir_size_kb) {
1979                 sbi->s_max_dir_size_kb = arg;
1980         } else if (token == Opt_stripe) {
1981                 sbi->s_stripe = arg;
1982         } else if (token == Opt_resuid) {
1983                 uid = make_kuid(current_user_ns(), arg);
1984                 if (!uid_valid(uid)) {
1985                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1986                         return -1;
1987                 }
1988                 sbi->s_resuid = uid;
1989         } else if (token == Opt_resgid) {
1990                 gid = make_kgid(current_user_ns(), arg);
1991                 if (!gid_valid(gid)) {
1992                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1993                         return -1;
1994                 }
1995                 sbi->s_resgid = gid;
1996         } else if (token == Opt_journal_dev) {
1997                 if (is_remount) {
1998                         ext4_msg(sb, KERN_ERR,
1999                                  "Cannot specify journal on remount");
2000                         return -1;
2001                 }
2002                 *journal_devnum = arg;
2003         } else if (token == Opt_journal_path) {
2004                 char *journal_path;
2005                 struct inode *journal_inode;
2006                 struct path path;
2007                 int error;
2008
2009                 if (is_remount) {
2010                         ext4_msg(sb, KERN_ERR,
2011                                  "Cannot specify journal on remount");
2012                         return -1;
2013                 }
2014                 journal_path = match_strdup(&args[0]);
2015                 if (!journal_path) {
2016                         ext4_msg(sb, KERN_ERR, "error: could not dup "
2017                                 "journal device string");
2018                         return -1;
2019                 }
2020
2021                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
2022                 if (error) {
2023                         ext4_msg(sb, KERN_ERR, "error: could not find "
2024                                 "journal device path: error %d", error);
2025                         kfree(journal_path);
2026                         return -1;
2027                 }
2028
2029                 journal_inode = d_inode(path.dentry);
2030                 if (!S_ISBLK(journal_inode->i_mode)) {
2031                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
2032                                 "is not a block device", journal_path);
2033                         path_put(&path);
2034                         kfree(journal_path);
2035                         return -1;
2036                 }
2037
2038                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
2039                 path_put(&path);
2040                 kfree(journal_path);
2041         } else if (token == Opt_journal_ioprio) {
2042                 if (arg > 7) {
2043                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
2044                                  " (must be 0-7)");
2045                         return -1;
2046                 }
2047                 *journal_ioprio =
2048                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
2049         } else if (token == Opt_test_dummy_encryption) {
2050 #ifdef CONFIG_FS_ENCRYPTION
2051                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
2052                 ext4_msg(sb, KERN_WARNING,
2053                          "Test dummy encryption mode enabled");
2054 #else
2055                 ext4_msg(sb, KERN_WARNING,
2056                          "Test dummy encryption mount option ignored");
2057 #endif
2058         } else if (m->flags & MOPT_DATAJ) {
2059                 if (is_remount) {
2060                         if (!sbi->s_journal)
2061                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
2062                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
2063                                 ext4_msg(sb, KERN_ERR,
2064                                          "Cannot change data mode on remount");
2065                                 return -1;
2066                         }
2067                 } else {
2068                         clear_opt(sb, DATA_FLAGS);
2069                         sbi->s_mount_opt |= m->mount_opt;
2070                 }
2071 #ifdef CONFIG_QUOTA
2072         } else if (m->flags & MOPT_QFMT) {
2073                 if (sb_any_quota_loaded(sb) &&
2074                     sbi->s_jquota_fmt != m->mount_opt) {
2075                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
2076                                  "quota options when quota turned on");
2077                         return -1;
2078                 }
2079                 if (ext4_has_feature_quota(sb)) {
2080                         ext4_msg(sb, KERN_INFO,
2081                                  "Quota format mount options ignored "
2082                                  "when QUOTA feature is enabled");
2083                         return 1;
2084                 }
2085                 sbi->s_jquota_fmt = m->mount_opt;
2086 #endif
2087         } else if (token == Opt_dax) {
2088 #ifdef CONFIG_FS_DAX
2089                 ext4_msg(sb, KERN_WARNING,
2090                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
2091                 sbi->s_mount_opt |= m->mount_opt;
2092 #else
2093                 ext4_msg(sb, KERN_INFO, "dax option not supported");
2094                 return -1;
2095 #endif
2096         } else if (token == Opt_data_err_abort) {
2097                 sbi->s_mount_opt |= m->mount_opt;
2098         } else if (token == Opt_data_err_ignore) {
2099                 sbi->s_mount_opt &= ~m->mount_opt;
2100         } else {
2101                 if (!args->from)
2102                         arg = 1;
2103                 if (m->flags & MOPT_CLEAR)
2104                         arg = !arg;
2105                 else if (unlikely(!(m->flags & MOPT_SET))) {
2106                         ext4_msg(sb, KERN_WARNING,
2107                                  "buggy handling of option %s", opt);
2108                         WARN_ON(1);
2109                         return -1;
2110                 }
2111                 if (arg != 0)
2112                         sbi->s_mount_opt |= m->mount_opt;
2113                 else
2114                         sbi->s_mount_opt &= ~m->mount_opt;
2115         }
2116         return 1;
2117 }
2118
2119 static int parse_options(char *options, struct super_block *sb,
2120                          unsigned long *journal_devnum,
2121                          unsigned int *journal_ioprio,
2122                          int is_remount)
2123 {
2124         struct ext4_sb_info __maybe_unused *sbi = EXT4_SB(sb);
2125         char *p, __maybe_unused *usr_qf_name, __maybe_unused *grp_qf_name;
2126         substring_t args[MAX_OPT_ARGS];
2127         int token;
2128
2129         if (!options)
2130                 return 1;
2131
2132         while ((p = strsep(&options, ",")) != NULL) {
2133                 if (!*p)
2134                         continue;
2135                 /*
2136                  * Initialize args struct so we know whether arg was
2137                  * found; some options take optional arguments.
2138                  */
2139                 args[0].to = args[0].from = NULL;
2140                 token = match_token(p, tokens, args);
2141                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
2142                                      journal_ioprio, is_remount) < 0)
2143                         return 0;
2144         }
2145 #ifdef CONFIG_QUOTA
2146         /*
2147          * We do the test below only for project quotas. 'usrquota' and
2148          * 'grpquota' mount options are allowed even without quota feature
2149          * to support legacy quotas in quota files.
2150          */
2151         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
2152                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
2153                          "Cannot enable project quota enforcement.");
2154                 return 0;
2155         }
2156         usr_qf_name = get_qf_name(sb, sbi, USRQUOTA);
2157         grp_qf_name = get_qf_name(sb, sbi, GRPQUOTA);
2158         if (usr_qf_name || grp_qf_name) {
2159                 if (test_opt(sb, USRQUOTA) && usr_qf_name)
2160                         clear_opt(sb, USRQUOTA);
2161
2162                 if (test_opt(sb, GRPQUOTA) && grp_qf_name)
2163                         clear_opt(sb, GRPQUOTA);
2164
2165                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
2166                         ext4_msg(sb, KERN_ERR, "old and new quota "
2167                                         "format mixing");
2168                         return 0;
2169                 }
2170
2171                 if (!sbi->s_jquota_fmt) {
2172                         ext4_msg(sb, KERN_ERR, "journaled quota format "
2173                                         "not specified");
2174                         return 0;
2175                 }
2176         }
2177 #endif
2178         if (test_opt(sb, DIOREAD_NOLOCK)) {
2179                 int blocksize =
2180                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
2181                 if (blocksize < PAGE_SIZE)
2182                         ext4_msg(sb, KERN_WARNING, "Warning: mounting with an "
2183                                  "experimental mount option 'dioread_nolock' "
2184                                  "for blocksize < PAGE_SIZE");
2185         }
2186         return 1;
2187 }
2188
2189 static inline void ext4_show_quota_options(struct seq_file *seq,
2190                                            struct super_block *sb)
2191 {
2192 #if defined(CONFIG_QUOTA)
2193         struct ext4_sb_info *sbi = EXT4_SB(sb);
2194         char *usr_qf_name, *grp_qf_name;
2195
2196         if (sbi->s_jquota_fmt) {
2197                 char *fmtname = "";
2198
2199                 switch (sbi->s_jquota_fmt) {
2200                 case QFMT_VFS_OLD:
2201                         fmtname = "vfsold";
2202                         break;
2203                 case QFMT_VFS_V0:
2204                         fmtname = "vfsv0";
2205                         break;
2206                 case QFMT_VFS_V1:
2207                         fmtname = "vfsv1";
2208                         break;
2209                 }
2210                 seq_printf(seq, ",jqfmt=%s", fmtname);
2211         }
2212
2213         rcu_read_lock();
2214         usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]);
2215         grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]);
2216         if (usr_qf_name)
2217                 seq_show_option(seq, "usrjquota", usr_qf_name);
2218         if (grp_qf_name)
2219                 seq_show_option(seq, "grpjquota", grp_qf_name);
2220         rcu_read_unlock();
2221 #endif
2222 }
2223
2224 static const char *token2str(int token)
2225 {
2226         const struct match_token *t;
2227
2228         for (t = tokens; t->token != Opt_err; t++)
2229                 if (t->token == token && !strchr(t->pattern, '='))
2230                         break;
2231         return t->pattern;
2232 }
2233
2234 /*
2235  * Show an option if
2236  *  - it's set to a non-default value OR
2237  *  - if the per-sb default is different from the global default
2238  */
2239 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2240                               int nodefs)
2241 {
2242         struct ext4_sb_info *sbi = EXT4_SB(sb);
2243         struct ext4_super_block *es = sbi->s_es;
2244         int def_errors, def_mount_opt = sbi->s_def_mount_opt;
2245         const struct mount_opts *m;
2246         char sep = nodefs ? '\n' : ',';
2247
2248 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2249 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2250
2251         if (sbi->s_sb_block != 1)
2252                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2253
2254         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2255                 int want_set = m->flags & MOPT_SET;
2256                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2257                     (m->flags & MOPT_CLEAR_ERR))
2258                         continue;
2259                 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2260                         continue; /* skip if same as the default */
2261                 if ((want_set &&
2262                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2263                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2264                         continue; /* select Opt_noFoo vs Opt_Foo */
2265                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2266         }
2267
2268         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2269             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2270                 SEQ_OPTS_PRINT("resuid=%u",
2271                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2272         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2273             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2274                 SEQ_OPTS_PRINT("resgid=%u",
2275                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2276         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2277         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2278                 SEQ_OPTS_PUTS("errors=remount-ro");
2279         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2280                 SEQ_OPTS_PUTS("errors=continue");
2281         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2282                 SEQ_OPTS_PUTS("errors=panic");
2283         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2284                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2285         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2286                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2287         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2288                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2289         if (sb->s_flags & SB_I_VERSION)
2290                 SEQ_OPTS_PUTS("i_version");
2291         if (nodefs || sbi->s_stripe)
2292                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2293         if (nodefs || EXT4_MOUNT_DATA_FLAGS &
2294                         (sbi->s_mount_opt ^ def_mount_opt)) {
2295                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2296                         SEQ_OPTS_PUTS("data=journal");
2297                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2298                         SEQ_OPTS_PUTS("data=ordered");
2299                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2300                         SEQ_OPTS_PUTS("data=writeback");
2301         }
2302         if (nodefs ||
2303             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2304                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2305                                sbi->s_inode_readahead_blks);
2306
2307         if (test_opt(sb, INIT_INODE_TABLE) && (nodefs ||
2308                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2309                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2310         if (nodefs || sbi->s_max_dir_size_kb)
2311                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2312         if (test_opt(sb, DATA_ERR_ABORT))
2313                 SEQ_OPTS_PUTS("data_err=abort");
2314         if (DUMMY_ENCRYPTION_ENABLED(sbi))
2315                 SEQ_OPTS_PUTS("test_dummy_encryption");
2316
2317         ext4_show_quota_options(seq, sb);
2318         return 0;
2319 }
2320
2321 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2322 {
2323         return _ext4_show_options(seq, root->d_sb, 0);
2324 }
2325
2326 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2327 {
2328         struct super_block *sb = seq->private;
2329         int rc;
2330
2331         seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
2332         rc = _ext4_show_options(seq, sb, 1);
2333         seq_puts(seq, "\n");
2334         return rc;
2335 }
2336
2337 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2338                             int read_only)
2339 {
2340         struct ext4_sb_info *sbi = EXT4_SB(sb);
2341         int err = 0;
2342
2343         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2344                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2345                          "forcing read-only mode");
2346                 err = -EROFS;
2347         }
2348         if (read_only)
2349                 goto done;
2350         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2351                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2352                          "running e2fsck is recommended");
2353         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2354                 ext4_msg(sb, KERN_WARNING,
2355                          "warning: mounting fs with errors, "
2356                          "running e2fsck is recommended");
2357         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2358                  le16_to_cpu(es->s_mnt_count) >=
2359                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2360                 ext4_msg(sb, KERN_WARNING,
2361                          "warning: maximal mount count reached, "
2362                          "running e2fsck is recommended");
2363         else if (le32_to_cpu(es->s_checkinterval) &&
2364                  (ext4_get_tstamp(es, s_lastcheck) +
2365                   le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds()))
2366                 ext4_msg(sb, KERN_WARNING,
2367                          "warning: checktime reached, "
2368                          "running e2fsck is recommended");
2369         if (!sbi->s_journal)
2370                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2371         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2372                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2373         le16_add_cpu(&es->s_mnt_count, 1);
2374         ext4_update_tstamp(es, s_mtime);
2375         if (sbi->s_journal)
2376                 ext4_set_feature_journal_needs_recovery(sb);
2377
2378         err = ext4_commit_super(sb, 1);
2379 done:
2380         if (test_opt(sb, DEBUG))
2381                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2382                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2383                         sb->s_blocksize,
2384                         sbi->s_groups_count,
2385                         EXT4_BLOCKS_PER_GROUP(sb),
2386                         EXT4_INODES_PER_GROUP(sb),
2387                         sbi->s_mount_opt, sbi->s_mount_opt2);
2388
2389         cleancache_init_fs(sb);
2390         return err;
2391 }
2392
2393 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2394 {
2395         struct ext4_sb_info *sbi = EXT4_SB(sb);
2396         struct flex_groups **old_groups, **new_groups;
2397         int size, i, j;
2398
2399         if (!sbi->s_log_groups_per_flex)
2400                 return 0;
2401
2402         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2403         if (size <= sbi->s_flex_groups_allocated)
2404                 return 0;
2405
2406         new_groups = kvzalloc(roundup_pow_of_two(size *
2407                               sizeof(*sbi->s_flex_groups)), GFP_KERNEL);
2408         if (!new_groups) {
2409                 ext4_msg(sb, KERN_ERR,
2410                          "not enough memory for %d flex group pointers", size);
2411                 return -ENOMEM;
2412         }
2413         for (i = sbi->s_flex_groups_allocated; i < size; i++) {
2414                 new_groups[i] = kvzalloc(roundup_pow_of_two(
2415                                          sizeof(struct flex_groups)),
2416                                          GFP_KERNEL);
2417                 if (!new_groups[i]) {
2418                         for (j = sbi->s_flex_groups_allocated; j < i; j++)
2419                                 kvfree(new_groups[j]);
2420                         kvfree(new_groups);
2421                         ext4_msg(sb, KERN_ERR,
2422                                  "not enough memory for %d flex groups", size);
2423                         return -ENOMEM;
2424                 }
2425         }
2426         rcu_read_lock();
2427         old_groups = rcu_dereference(sbi->s_flex_groups);
2428         if (old_groups)
2429                 memcpy(new_groups, old_groups,
2430                        (sbi->s_flex_groups_allocated *
2431                         sizeof(struct flex_groups *)));
2432         rcu_read_unlock();
2433         rcu_assign_pointer(sbi->s_flex_groups, new_groups);
2434         sbi->s_flex_groups_allocated = size;
2435         if (old_groups)
2436                 ext4_kvfree_array_rcu(old_groups);
2437         return 0;
2438 }
2439
2440 static int ext4_fill_flex_info(struct super_block *sb)
2441 {
2442         struct ext4_sb_info *sbi = EXT4_SB(sb);
2443         struct ext4_group_desc *gdp = NULL;
2444         struct flex_groups *fg;
2445         ext4_group_t flex_group;
2446         int i, err;
2447
2448         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2449         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2450                 sbi->s_log_groups_per_flex = 0;
2451                 return 1;
2452         }
2453
2454         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2455         if (err)
2456                 goto failed;
2457
2458         for (i = 0; i < sbi->s_groups_count; i++) {
2459                 gdp = ext4_get_group_desc(sb, i, NULL);
2460
2461                 flex_group = ext4_flex_group(sbi, i);
2462                 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
2463                 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes);
2464                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2465                              &fg->free_clusters);
2466                 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs);
2467         }
2468
2469         return 1;
2470 failed:
2471         return 0;
2472 }
2473
2474 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2475                                    struct ext4_group_desc *gdp)
2476 {
2477         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2478         __u16 crc = 0;
2479         __le32 le_group = cpu_to_le32(block_group);
2480         struct ext4_sb_info *sbi = EXT4_SB(sb);
2481
2482         if (ext4_has_metadata_csum(sbi->s_sb)) {
2483                 /* Use new metadata_csum algorithm */
2484                 __u32 csum32;
2485                 __u16 dummy_csum = 0;
2486
2487                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2488                                      sizeof(le_group));
2489                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2490                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2491                                      sizeof(dummy_csum));
2492                 offset += sizeof(dummy_csum);
2493                 if (offset < sbi->s_desc_size)
2494                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2495                                              sbi->s_desc_size - offset);
2496
2497                 crc = csum32 & 0xFFFF;
2498                 goto out;
2499         }
2500
2501         /* old crc16 code */
2502         if (!ext4_has_feature_gdt_csum(sb))
2503                 return 0;
2504
2505         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2506         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2507         crc = crc16(crc, (__u8 *)gdp, offset);
2508         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2509         /* for checksum of struct ext4_group_desc do the rest...*/
2510         if (ext4_has_feature_64bit(sb) &&
2511             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2512                 crc = crc16(crc, (__u8 *)gdp + offset,
2513                             le16_to_cpu(sbi->s_es->s_desc_size) -
2514                                 offset);
2515
2516 out:
2517         return cpu_to_le16(crc);
2518 }
2519
2520 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2521                                 struct ext4_group_desc *gdp)
2522 {
2523         if (ext4_has_group_desc_csum(sb) &&
2524             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2525                 return 0;
2526
2527         return 1;
2528 }
2529
2530 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2531                               struct ext4_group_desc *gdp)
2532 {
2533         if (!ext4_has_group_desc_csum(sb))
2534                 return;
2535         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2536 }
2537
2538 /* Called at mount-time, super-block is locked */
2539 static int ext4_check_descriptors(struct super_block *sb,
2540                                   ext4_fsblk_t sb_block,
2541                                   ext4_group_t *first_not_zeroed)
2542 {
2543         struct ext4_sb_info *sbi = EXT4_SB(sb);
2544         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2545         ext4_fsblk_t last_block;
2546         ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0);
2547         ext4_fsblk_t block_bitmap;
2548         ext4_fsblk_t inode_bitmap;
2549         ext4_fsblk_t inode_table;
2550         int flexbg_flag = 0;
2551         ext4_group_t i, grp = sbi->s_groups_count;
2552
2553         if (ext4_has_feature_flex_bg(sb))
2554                 flexbg_flag = 1;
2555
2556         ext4_debug("Checking group descriptors");
2557
2558         for (i = 0; i < sbi->s_groups_count; i++) {
2559                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2560
2561                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2562                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2563                 else
2564                         last_block = first_block +
2565                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2566
2567                 if ((grp == sbi->s_groups_count) &&
2568                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2569                         grp = i;
2570
2571                 block_bitmap = ext4_block_bitmap(sb, gdp);
2572                 if (block_bitmap == sb_block) {
2573                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2574                                  "Block bitmap for group %u overlaps "
2575                                  "superblock", i);
2576                         if (!sb_rdonly(sb))
2577                                 return 0;
2578                 }
2579                 if (block_bitmap >= sb_block + 1 &&
2580                     block_bitmap <= last_bg_block) {
2581                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2582                                  "Block bitmap for group %u overlaps "
2583                                  "block group descriptors", i);
2584                         if (!sb_rdonly(sb))
2585                                 return 0;
2586                 }
2587                 if (block_bitmap < first_block || block_bitmap > last_block) {
2588                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2589                                "Block bitmap for group %u not in group "
2590                                "(block %llu)!", i, block_bitmap);
2591                         return 0;
2592                 }
2593                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2594                 if (inode_bitmap == sb_block) {
2595                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2596                                  "Inode bitmap for group %u overlaps "
2597                                  "superblock", i);
2598                         if (!sb_rdonly(sb))
2599                                 return 0;
2600                 }
2601                 if (inode_bitmap >= sb_block + 1 &&
2602                     inode_bitmap <= last_bg_block) {
2603                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2604                                  "Inode bitmap for group %u overlaps "
2605                                  "block group descriptors", i);
2606                         if (!sb_rdonly(sb))
2607                                 return 0;
2608                 }
2609                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2610                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2611                                "Inode bitmap for group %u not in group "
2612                                "(block %llu)!", i, inode_bitmap);
2613                         return 0;
2614                 }
2615                 inode_table = ext4_inode_table(sb, gdp);
2616                 if (inode_table == sb_block) {
2617                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2618                                  "Inode table for group %u overlaps "
2619                                  "superblock", i);
2620                         if (!sb_rdonly(sb))
2621                                 return 0;
2622                 }
2623                 if (inode_table >= sb_block + 1 &&
2624                     inode_table <= last_bg_block) {
2625                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2626                                  "Inode table for group %u overlaps "
2627                                  "block group descriptors", i);
2628                         if (!sb_rdonly(sb))
2629                                 return 0;
2630                 }
2631                 if (inode_table < first_block ||
2632                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2633                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2634                                "Inode table for group %u not in group "
2635                                "(block %llu)!", i, inode_table);
2636                         return 0;
2637                 }
2638                 ext4_lock_group(sb, i);
2639                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2640                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2641                                  "Checksum for group %u failed (%u!=%u)",
2642                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2643                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2644                         if (!sb_rdonly(sb)) {
2645                                 ext4_unlock_group(sb, i);
2646                                 return 0;
2647                         }
2648                 }
2649                 ext4_unlock_group(sb, i);
2650                 if (!flexbg_flag)
2651                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2652         }
2653         if (NULL != first_not_zeroed)
2654                 *first_not_zeroed = grp;
2655         return 1;
2656 }
2657
2658 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2659  * the superblock) which were deleted from all directories, but held open by
2660  * a process at the time of a crash.  We walk the list and try to delete these
2661  * inodes at recovery time (only with a read-write filesystem).
2662  *
2663  * In order to keep the orphan inode chain consistent during traversal (in
2664  * case of crash during recovery), we link each inode into the superblock
2665  * orphan list_head and handle it the same way as an inode deletion during
2666  * normal operation (which journals the operations for us).
2667  *
2668  * We only do an iget() and an iput() on each inode, which is very safe if we
2669  * accidentally point at an in-use or already deleted inode.  The worst that
2670  * can happen in this case is that we get a "bit already cleared" message from
2671  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2672  * e2fsck was run on this filesystem, and it must have already done the orphan
2673  * inode cleanup for us, so we can safely abort without any further action.
2674  */
2675 static void ext4_orphan_cleanup(struct super_block *sb,
2676                                 struct ext4_super_block *es)
2677 {
2678         unsigned int s_flags = sb->s_flags;
2679         int ret, nr_orphans = 0, nr_truncates = 0;
2680 #ifdef CONFIG_QUOTA
2681         int quota_update = 0;
2682         int i;
2683 #endif
2684         if (!es->s_last_orphan) {
2685                 jbd_debug(4, "no orphan inodes to clean up\n");
2686                 return;
2687         }
2688
2689         if (bdev_read_only(sb->s_bdev)) {
2690                 ext4_msg(sb, KERN_ERR, "write access "
2691                         "unavailable, skipping orphan cleanup");
2692                 return;
2693         }
2694
2695         /* Check if feature set would not allow a r/w mount */
2696         if (!ext4_feature_set_ok(sb, 0)) {
2697                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2698                          "unknown ROCOMPAT features");
2699                 return;
2700         }
2701
2702         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2703                 /* don't clear list on RO mount w/ errors */
2704                 if (es->s_last_orphan && !(s_flags & SB_RDONLY)) {
2705                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2706                                   "clearing orphan list.\n");
2707                         es->s_last_orphan = 0;
2708                 }
2709                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2710                 return;
2711         }
2712
2713         if (s_flags & SB_RDONLY) {
2714                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2715                 sb->s_flags &= ~SB_RDONLY;
2716         }
2717 #ifdef CONFIG_QUOTA
2718         /* Needed for iput() to work correctly and not trash data */
2719         sb->s_flags |= SB_ACTIVE;
2720
2721         /*
2722          * Turn on quotas which were not enabled for read-only mounts if
2723          * filesystem has quota feature, so that they are updated correctly.
2724          */
2725         if (ext4_has_feature_quota(sb) && (s_flags & SB_RDONLY)) {
2726                 int ret = ext4_enable_quotas(sb);
2727
2728                 if (!ret)
2729                         quota_update = 1;
2730                 else
2731                         ext4_msg(sb, KERN_ERR,
2732                                 "Cannot turn on quotas: error %d", ret);
2733         }
2734
2735         /* Turn on journaled quotas used for old sytle */
2736         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2737                 if (EXT4_SB(sb)->s_qf_names[i]) {
2738                         int ret = ext4_quota_on_mount(sb, i);
2739
2740                         if (!ret)
2741                                 quota_update = 1;
2742                         else
2743                                 ext4_msg(sb, KERN_ERR,
2744                                         "Cannot turn on journaled "
2745                                         "quota: type %d: error %d", i, ret);
2746                 }
2747         }
2748 #endif
2749
2750         while (es->s_last_orphan) {
2751                 struct inode *inode;
2752
2753                 /*
2754                  * We may have encountered an error during cleanup; if
2755                  * so, skip the rest.
2756                  */
2757                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2758                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2759                         es->s_last_orphan = 0;
2760                         break;
2761                 }
2762
2763                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2764                 if (IS_ERR(inode)) {
2765                         es->s_last_orphan = 0;
2766                         break;
2767                 }
2768
2769                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2770                 dquot_initialize(inode);
2771                 if (inode->i_nlink) {
2772                         if (test_opt(sb, DEBUG))
2773                                 ext4_msg(sb, KERN_DEBUG,
2774                                         "%s: truncating inode %lu to %lld bytes",
2775                                         __func__, inode->i_ino, inode->i_size);
2776                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2777                                   inode->i_ino, inode->i_size);
2778                         inode_lock(inode);
2779                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2780                         ret = ext4_truncate(inode);
2781                         if (ret)
2782                                 ext4_std_error(inode->i_sb, ret);
2783                         inode_unlock(inode);
2784                         nr_truncates++;
2785                 } else {
2786                         if (test_opt(sb, DEBUG))
2787                                 ext4_msg(sb, KERN_DEBUG,
2788                                         "%s: deleting unreferenced inode %lu",
2789                                         __func__, inode->i_ino);
2790                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2791                                   inode->i_ino);
2792                         nr_orphans++;
2793                 }
2794                 iput(inode);  /* The delete magic happens here! */
2795         }
2796
2797 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2798
2799         if (nr_orphans)
2800                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2801                        PLURAL(nr_orphans));
2802         if (nr_truncates)
2803                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2804                        PLURAL(nr_truncates));
2805 #ifdef CONFIG_QUOTA
2806         /* Turn off quotas if they were enabled for orphan cleanup */
2807         if (quota_update) {
2808                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2809                         if (sb_dqopt(sb)->files[i])
2810                                 dquot_quota_off(sb, i);
2811                 }
2812         }
2813 #endif
2814         sb->s_flags = s_flags; /* Restore SB_RDONLY status */
2815 }
2816
2817 /*
2818  * Maximal extent format file size.
2819  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2820  * extent format containers, within a sector_t, and within i_blocks
2821  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2822  * so that won't be a limiting factor.
2823  *
2824  * However there is other limiting factor. We do store extents in the form
2825  * of starting block and length, hence the resulting length of the extent
2826  * covering maximum file size must fit into on-disk format containers as
2827  * well. Given that length is always by 1 unit bigger than max unit (because
2828  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2829  *
2830  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2831  */
2832 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2833 {
2834         loff_t res;
2835         loff_t upper_limit = MAX_LFS_FILESIZE;
2836
2837         BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64));
2838
2839         if (!has_huge_files) {
2840                 upper_limit = (1LL << 32) - 1;
2841
2842                 /* total blocks in file system block size */
2843                 upper_limit >>= (blkbits - 9);
2844                 upper_limit <<= blkbits;
2845         }
2846
2847         /*
2848          * 32-bit extent-start container, ee_block. We lower the maxbytes
2849          * by one fs block, so ee_len can cover the extent of maximum file
2850          * size
2851          */
2852         res = (1LL << 32) - 1;
2853         res <<= blkbits;
2854
2855         /* Sanity check against vm- & vfs- imposed limits */
2856         if (res > upper_limit)
2857                 res = upper_limit;
2858
2859         return res;
2860 }
2861
2862 /*
2863  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2864  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2865  * We need to be 1 filesystem block less than the 2^48 sector limit.
2866  */
2867 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2868 {
2869         loff_t res = EXT4_NDIR_BLOCKS;
2870         int meta_blocks;
2871         loff_t upper_limit;
2872         /* This is calculated to be the largest file size for a dense, block
2873          * mapped file such that the file's total number of 512-byte sectors,
2874          * including data and all indirect blocks, does not exceed (2^48 - 1).
2875          *
2876          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2877          * number of 512-byte sectors of the file.
2878          */
2879
2880         if (!has_huge_files) {
2881                 /*
2882                  * !has_huge_files or implies that the inode i_block field
2883                  * represents total file blocks in 2^32 512-byte sectors ==
2884                  * size of vfs inode i_blocks * 8
2885                  */
2886                 upper_limit = (1LL << 32) - 1;
2887
2888                 /* total blocks in file system block size */
2889                 upper_limit >>= (bits - 9);
2890
2891         } else {
2892                 /*
2893                  * We use 48 bit ext4_inode i_blocks
2894                  * With EXT4_HUGE_FILE_FL set the i_blocks
2895                  * represent total number of blocks in
2896                  * file system block size
2897                  */
2898                 upper_limit = (1LL << 48) - 1;
2899
2900         }
2901
2902         /* indirect blocks */
2903         meta_blocks = 1;
2904         /* double indirect blocks */
2905         meta_blocks += 1 + (1LL << (bits-2));
2906         /* tripple indirect blocks */
2907         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2908
2909         upper_limit -= meta_blocks;
2910         upper_limit <<= bits;
2911
2912         res += 1LL << (bits-2);
2913         res += 1LL << (2*(bits-2));
2914         res += 1LL << (3*(bits-2));
2915         res <<= bits;
2916         if (res > upper_limit)
2917                 res = upper_limit;
2918
2919         if (res > MAX_LFS_FILESIZE)
2920                 res = MAX_LFS_FILESIZE;
2921
2922         return res;
2923 }
2924
2925 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2926                                    ext4_fsblk_t logical_sb_block, int nr)
2927 {
2928         struct ext4_sb_info *sbi = EXT4_SB(sb);
2929         ext4_group_t bg, first_meta_bg;
2930         int has_super = 0;
2931
2932         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2933
2934         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2935                 return logical_sb_block + nr + 1;
2936         bg = sbi->s_desc_per_block * nr;
2937         if (ext4_bg_has_super(sb, bg))
2938                 has_super = 1;
2939
2940         /*
2941          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2942          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2943          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2944          * compensate.
2945          */
2946         if (sb->s_blocksize == 1024 && nr == 0 &&
2947             le32_to_cpu(sbi->s_es->s_first_data_block) == 0)
2948                 has_super++;
2949
2950         return (has_super + ext4_group_first_block_no(sb, bg));
2951 }
2952
2953 /**
2954  * ext4_get_stripe_size: Get the stripe size.
2955  * @sbi: In memory super block info
2956  *
2957  * If we have specified it via mount option, then
2958  * use the mount option value. If the value specified at mount time is
2959  * greater than the blocks per group use the super block value.
2960  * If the super block value is greater than blocks per group return 0.
2961  * Allocator needs it be less than blocks per group.
2962  *
2963  */
2964 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2965 {
2966         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2967         unsigned long stripe_width =
2968                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2969         int ret;
2970
2971         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2972                 ret = sbi->s_stripe;
2973         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
2974                 ret = stripe_width;
2975         else if (stride && stride <= sbi->s_blocks_per_group)
2976                 ret = stride;
2977         else
2978                 ret = 0;
2979
2980         /*
2981          * If the stripe width is 1, this makes no sense and
2982          * we set it to 0 to turn off stripe handling code.
2983          */
2984         if (ret <= 1)
2985                 ret = 0;
2986
2987         return ret;
2988 }
2989
2990 /*
2991  * Check whether this filesystem can be mounted based on
2992  * the features present and the RDONLY/RDWR mount requested.
2993  * Returns 1 if this filesystem can be mounted as requested,
2994  * 0 if it cannot be.
2995  */
2996 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2997 {
2998         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2999                 ext4_msg(sb, KERN_ERR,
3000                         "Couldn't mount because of "
3001                         "unsupported optional features (%x)",
3002                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
3003                         ~EXT4_FEATURE_INCOMPAT_SUPP));
3004                 return 0;
3005         }
3006
3007 #ifndef CONFIG_UNICODE
3008         if (ext4_has_feature_casefold(sb)) {
3009                 ext4_msg(sb, KERN_ERR,
3010                          "Filesystem with casefold feature cannot be "
3011                          "mounted without CONFIG_UNICODE");
3012                 return 0;
3013         }
3014 #endif
3015
3016         if (readonly)
3017                 return 1;
3018
3019         if (ext4_has_feature_readonly(sb)) {
3020                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
3021                 sb->s_flags |= SB_RDONLY;
3022                 return 1;
3023         }
3024
3025         /* Check that feature set is OK for a read-write mount */
3026         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
3027                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
3028                          "unsupported optional features (%x)",
3029                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
3030                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
3031                 return 0;
3032         }
3033         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
3034                 ext4_msg(sb, KERN_ERR,
3035                          "Can't support bigalloc feature without "
3036                          "extents feature\n");
3037                 return 0;
3038         }
3039
3040 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2)
3041         if (!readonly && (ext4_has_feature_quota(sb) ||
3042                           ext4_has_feature_project(sb))) {
3043                 ext4_msg(sb, KERN_ERR,
3044                          "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2");
3045                 return 0;
3046         }
3047 #endif  /* CONFIG_QUOTA */
3048         return 1;
3049 }
3050
3051 /*
3052  * This function is called once a day if we have errors logged
3053  * on the file system
3054  */
3055 static void print_daily_error_info(struct timer_list *t)
3056 {
3057         struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report);
3058         struct super_block *sb = sbi->s_sb;
3059         struct ext4_super_block *es = sbi->s_es;
3060
3061         if (es->s_error_count)
3062                 /* fsck newer than v1.41.13 is needed to clean this condition. */
3063                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
3064                          le32_to_cpu(es->s_error_count));
3065         if (es->s_first_error_time) {
3066                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d",
3067                        sb->s_id,
3068                        ext4_get_tstamp(es, s_first_error_time),
3069                        (int) sizeof(es->s_first_error_func),
3070                        es->s_first_error_func,
3071                        le32_to_cpu(es->s_first_error_line));
3072                 if (es->s_first_error_ino)
3073                         printk(KERN_CONT ": inode %u",
3074                                le32_to_cpu(es->s_first_error_ino));
3075                 if (es->s_first_error_block)
3076                         printk(KERN_CONT ": block %llu", (unsigned long long)
3077                                le64_to_cpu(es->s_first_error_block));
3078                 printk(KERN_CONT "\n");
3079         }
3080         if (es->s_last_error_time) {
3081                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d",
3082                        sb->s_id,
3083                        ext4_get_tstamp(es, s_last_error_time),
3084                        (int) sizeof(es->s_last_error_func),
3085                        es->s_last_error_func,
3086                        le32_to_cpu(es->s_last_error_line));
3087                 if (es->s_last_error_ino)
3088                         printk(KERN_CONT ": inode %u",
3089                                le32_to_cpu(es->s_last_error_ino));
3090                 if (es->s_last_error_block)
3091                         printk(KERN_CONT ": block %llu", (unsigned long long)
3092                                le64_to_cpu(es->s_last_error_block));
3093                 printk(KERN_CONT "\n");
3094         }
3095         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
3096 }
3097
3098 /* Find next suitable group and run ext4_init_inode_table */
3099 static int ext4_run_li_request(struct ext4_li_request *elr)
3100 {
3101         struct ext4_group_desc *gdp = NULL;
3102         ext4_group_t group, ngroups;
3103         struct super_block *sb;
3104         unsigned long timeout = 0;
3105         int ret = 0;
3106
3107         sb = elr->lr_super;
3108         ngroups = EXT4_SB(sb)->s_groups_count;
3109
3110         for (group = elr->lr_next_group; group < ngroups; group++) {
3111                 gdp = ext4_get_group_desc(sb, group, NULL);
3112                 if (!gdp) {
3113                         ret = 1;
3114                         break;
3115                 }
3116
3117                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3118                         break;
3119         }
3120
3121         if (group >= ngroups)
3122                 ret = 1;
3123
3124         if (!ret) {
3125                 timeout = jiffies;
3126                 ret = ext4_init_inode_table(sb, group,
3127                                             elr->lr_timeout ? 0 : 1);
3128                 if (elr->lr_timeout == 0) {
3129                         timeout = (jiffies - timeout) *
3130                                   elr->lr_sbi->s_li_wait_mult;
3131                         elr->lr_timeout = timeout;
3132                 }
3133                 elr->lr_next_sched = jiffies + elr->lr_timeout;
3134                 elr->lr_next_group = group + 1;
3135         }
3136         return ret;
3137 }
3138
3139 /*
3140  * Remove lr_request from the list_request and free the
3141  * request structure. Should be called with li_list_mtx held
3142  */
3143 static void ext4_remove_li_request(struct ext4_li_request *elr)
3144 {
3145         struct ext4_sb_info *sbi;
3146
3147         if (!elr)
3148                 return;
3149
3150         sbi = elr->lr_sbi;
3151
3152         list_del(&elr->lr_request);
3153         sbi->s_li_request = NULL;
3154         kfree(elr);
3155 }
3156
3157 static void ext4_unregister_li_request(struct super_block *sb)
3158 {
3159         mutex_lock(&ext4_li_mtx);
3160         if (!ext4_li_info) {
3161                 mutex_unlock(&ext4_li_mtx);
3162                 return;
3163         }
3164
3165         mutex_lock(&ext4_li_info->li_list_mtx);
3166         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
3167         mutex_unlock(&ext4_li_info->li_list_mtx);
3168         mutex_unlock(&ext4_li_mtx);
3169 }
3170
3171 static struct task_struct *ext4_lazyinit_task;
3172
3173 /*
3174  * This is the function where ext4lazyinit thread lives. It walks
3175  * through the request list searching for next scheduled filesystem.
3176  * When such a fs is found, run the lazy initialization request
3177  * (ext4_rn_li_request) and keep track of the time spend in this
3178  * function. Based on that time we compute next schedule time of
3179  * the request. When walking through the list is complete, compute
3180  * next waking time and put itself into sleep.
3181  */
3182 static int ext4_lazyinit_thread(void *arg)
3183 {
3184         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
3185         struct list_head *pos, *n;
3186         struct ext4_li_request *elr;
3187         unsigned long next_wakeup, cur;
3188
3189         BUG_ON(NULL == eli);
3190
3191 cont_thread:
3192         while (true) {
3193                 next_wakeup = MAX_JIFFY_OFFSET;
3194
3195                 mutex_lock(&eli->li_list_mtx);
3196                 if (list_empty(&eli->li_request_list)) {
3197                         mutex_unlock(&eli->li_list_mtx);
3198                         goto exit_thread;
3199                 }
3200                 list_for_each_safe(pos, n, &eli->li_request_list) {
3201                         int err = 0;
3202                         int progress = 0;
3203                         elr = list_entry(pos, struct ext4_li_request,
3204                                          lr_request);
3205
3206                         if (time_before(jiffies, elr->lr_next_sched)) {
3207                                 if (time_before(elr->lr_next_sched, next_wakeup))
3208                                         next_wakeup = elr->lr_next_sched;
3209                                 continue;
3210                         }
3211                         if (down_read_trylock(&elr->lr_super->s_umount)) {
3212                                 if (sb_start_write_trylock(elr->lr_super)) {
3213                                         progress = 1;
3214                                         /*
3215                                          * We hold sb->s_umount, sb can not
3216                                          * be removed from the list, it is
3217                                          * now safe to drop li_list_mtx
3218                                          */
3219                                         mutex_unlock(&eli->li_list_mtx);
3220                                         err = ext4_run_li_request(elr);
3221                                         sb_end_write(elr->lr_super);
3222                                         mutex_lock(&eli->li_list_mtx);
3223                                         n = pos->next;
3224                                 }
3225                                 up_read((&elr->lr_super->s_umount));
3226                         }
3227                         /* error, remove the lazy_init job */
3228                         if (err) {
3229                                 ext4_remove_li_request(elr);
3230                                 continue;
3231                         }
3232                         if (!progress) {
3233                                 elr->lr_next_sched = jiffies +
3234                                         (prandom_u32()
3235                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3236                         }
3237                         if (time_before(elr->lr_next_sched, next_wakeup))
3238                                 next_wakeup = elr->lr_next_sched;
3239                 }
3240                 mutex_unlock(&eli->li_list_mtx);
3241
3242                 try_to_freeze();
3243
3244                 cur = jiffies;
3245                 if ((time_after_eq(cur, next_wakeup)) ||
3246                     (MAX_JIFFY_OFFSET == next_wakeup)) {
3247                         cond_resched();
3248                         continue;
3249                 }
3250
3251                 schedule_timeout_interruptible(next_wakeup - cur);
3252
3253                 if (kthread_should_stop()) {
3254                         ext4_clear_request_list();
3255                         goto exit_thread;
3256                 }
3257         }
3258
3259 exit_thread:
3260         /*
3261          * It looks like the request list is empty, but we need
3262          * to check it under the li_list_mtx lock, to prevent any
3263          * additions into it, and of course we should lock ext4_li_mtx
3264          * to atomically free the list and ext4_li_info, because at
3265          * this point another ext4 filesystem could be registering
3266          * new one.
3267          */
3268         mutex_lock(&ext4_li_mtx);
3269         mutex_lock(&eli->li_list_mtx);
3270         if (!list_empty(&eli->li_request_list)) {
3271                 mutex_unlock(&eli->li_list_mtx);
3272                 mutex_unlock(&ext4_li_mtx);
3273                 goto cont_thread;
3274         }
3275         mutex_unlock(&eli->li_list_mtx);
3276         kfree(ext4_li_info);
3277         ext4_li_info = NULL;
3278         mutex_unlock(&ext4_li_mtx);
3279
3280         return 0;
3281 }
3282
3283 static void ext4_clear_request_list(void)
3284 {
3285         struct list_head *pos, *n;
3286         struct ext4_li_request *elr;
3287
3288         mutex_lock(&ext4_li_info->li_list_mtx);
3289         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3290                 elr = list_entry(pos, struct ext4_li_request,
3291                                  lr_request);
3292                 ext4_remove_li_request(elr);
3293         }
3294         mutex_unlock(&ext4_li_info->li_list_mtx);
3295 }
3296
3297 static int ext4_run_lazyinit_thread(void)
3298 {
3299         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3300                                          ext4_li_info, "ext4lazyinit");
3301         if (IS_ERR(ext4_lazyinit_task)) {
3302                 int err = PTR_ERR(ext4_lazyinit_task);
3303                 ext4_clear_request_list();
3304                 kfree(ext4_li_info);
3305                 ext4_li_info = NULL;
3306                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3307                                  "initialization thread\n",
3308                                  err);
3309                 return err;
3310         }
3311         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3312         return 0;
3313 }
3314
3315 /*
3316  * Check whether it make sense to run itable init. thread or not.
3317  * If there is at least one uninitialized inode table, return
3318  * corresponding group number, else the loop goes through all
3319  * groups and return total number of groups.
3320  */
3321 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3322 {
3323         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3324         struct ext4_group_desc *gdp = NULL;
3325
3326         if (!ext4_has_group_desc_csum(sb))
3327                 return ngroups;
3328
3329         for (group = 0; group < ngroups; group++) {
3330                 gdp = ext4_get_group_desc(sb, group, NULL);
3331                 if (!gdp)
3332                         continue;
3333
3334                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3335                         break;
3336         }
3337
3338         return group;
3339 }
3340
3341 static int ext4_li_info_new(void)
3342 {
3343         struct ext4_lazy_init *eli = NULL;
3344
3345         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3346         if (!eli)
3347                 return -ENOMEM;
3348
3349         INIT_LIST_HEAD(&eli->li_request_list);
3350         mutex_init(&eli->li_list_mtx);
3351
3352         eli->li_state |= EXT4_LAZYINIT_QUIT;
3353
3354         ext4_li_info = eli;
3355
3356         return 0;
3357 }
3358
3359 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3360                                             ext4_group_t start)
3361 {
3362         struct ext4_sb_info *sbi = EXT4_SB(sb);
3363         struct ext4_li_request *elr;
3364
3365         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3366         if (!elr)
3367                 return NULL;
3368
3369         elr->lr_super = sb;
3370         elr->lr_sbi = sbi;
3371         elr->lr_next_group = start;
3372
3373         /*
3374          * Randomize first schedule time of the request to
3375          * spread the inode table initialization requests
3376          * better.
3377          */
3378         elr->lr_next_sched = jiffies + (prandom_u32() %
3379                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3380         return elr;
3381 }
3382
3383 int ext4_register_li_request(struct super_block *sb,
3384                              ext4_group_t first_not_zeroed)
3385 {
3386         struct ext4_sb_info *sbi = EXT4_SB(sb);
3387         struct ext4_li_request *elr = NULL;
3388         ext4_group_t ngroups = sbi->s_groups_count;
3389         int ret = 0;
3390
3391         mutex_lock(&ext4_li_mtx);
3392         if (sbi->s_li_request != NULL) {
3393                 /*
3394                  * Reset timeout so it can be computed again, because
3395                  * s_li_wait_mult might have changed.
3396                  */
3397                 sbi->s_li_request->lr_timeout = 0;
3398                 goto out;
3399         }
3400
3401         if (first_not_zeroed == ngroups || sb_rdonly(sb) ||
3402             !test_opt(sb, INIT_INODE_TABLE))
3403                 goto out;
3404
3405         elr = ext4_li_request_new(sb, first_not_zeroed);
3406         if (!elr) {
3407                 ret = -ENOMEM;
3408                 goto out;
3409         }
3410
3411         if (NULL == ext4_li_info) {
3412                 ret = ext4_li_info_new();
3413                 if (ret)
3414                         goto out;
3415         }
3416
3417         mutex_lock(&ext4_li_info->li_list_mtx);
3418         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3419         mutex_unlock(&ext4_li_info->li_list_mtx);
3420
3421         sbi->s_li_request = elr;
3422         /*
3423          * set elr to NULL here since it has been inserted to
3424          * the request_list and the removal and free of it is
3425          * handled by ext4_clear_request_list from now on.
3426          */
3427         elr = NULL;
3428
3429         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3430                 ret = ext4_run_lazyinit_thread();
3431                 if (ret)
3432                         goto out;
3433         }
3434 out:
3435         mutex_unlock(&ext4_li_mtx);
3436         if (ret)
3437                 kfree(elr);
3438         return ret;
3439 }
3440
3441 /*
3442  * We do not need to lock anything since this is called on
3443  * module unload.
3444  */
3445 static void ext4_destroy_lazyinit_thread(void)
3446 {
3447         /*
3448          * If thread exited earlier
3449          * there's nothing to be done.
3450          */
3451         if (!ext4_li_info || !ext4_lazyinit_task)
3452                 return;
3453
3454         kthread_stop(ext4_lazyinit_task);
3455 }
3456
3457 static int set_journal_csum_feature_set(struct super_block *sb)
3458 {
3459         int ret = 1;
3460         int compat, incompat;
3461         struct ext4_sb_info *sbi = EXT4_SB(sb);
3462
3463         if (ext4_has_metadata_csum(sb)) {
3464                 /* journal checksum v3 */
3465                 compat = 0;
3466                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3467         } else {
3468                 /* journal checksum v1 */
3469                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3470                 incompat = 0;
3471         }
3472
3473         jbd2_journal_clear_features(sbi->s_journal,
3474                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3475                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3476                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3477         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3478                 ret = jbd2_journal_set_features(sbi->s_journal,
3479                                 compat, 0,
3480                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3481                                 incompat);
3482         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3483                 ret = jbd2_journal_set_features(sbi->s_journal,
3484                                 compat, 0,
3485                                 incompat);
3486                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3487                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3488         } else {
3489                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3490                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3491         }
3492
3493         return ret;
3494 }
3495
3496 /*
3497  * Note: calculating the overhead so we can be compatible with
3498  * historical BSD practice is quite difficult in the face of
3499  * clusters/bigalloc.  This is because multiple metadata blocks from
3500  * different block group can end up in the same allocation cluster.
3501  * Calculating the exact overhead in the face of clustered allocation
3502  * requires either O(all block bitmaps) in memory or O(number of block
3503  * groups**2) in time.  We will still calculate the superblock for
3504  * older file systems --- and if we come across with a bigalloc file
3505  * system with zero in s_overhead_clusters the estimate will be close to
3506  * correct especially for very large cluster sizes --- but for newer
3507  * file systems, it's better to calculate this figure once at mkfs
3508  * time, and store it in the superblock.  If the superblock value is
3509  * present (even for non-bigalloc file systems), we will use it.
3510  */
3511 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3512                           char *buf)
3513 {
3514         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3515         struct ext4_group_desc  *gdp;
3516         ext4_fsblk_t            first_block, last_block, b;
3517         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3518         int                     s, j, count = 0;
3519
3520         if (!ext4_has_feature_bigalloc(sb))
3521                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3522                         sbi->s_itb_per_group + 2);
3523
3524         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3525                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3526         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3527         for (i = 0; i < ngroups; i++) {
3528                 gdp = ext4_get_group_desc(sb, i, NULL);
3529                 b = ext4_block_bitmap(sb, gdp);
3530                 if (b >= first_block && b <= last_block) {
3531                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3532                         count++;
3533                 }
3534                 b = ext4_inode_bitmap(sb, gdp);
3535                 if (b >= first_block && b <= last_block) {
3536                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3537                         count++;
3538                 }
3539                 b = ext4_inode_table(sb, gdp);
3540                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3541                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3542                                 int c = EXT4_B2C(sbi, b - first_block);
3543                                 ext4_set_bit(c, buf);
3544                                 count++;
3545                         }
3546                 if (i != grp)
3547                         continue;
3548                 s = 0;
3549                 if (ext4_bg_has_super(sb, grp)) {
3550                         ext4_set_bit(s++, buf);
3551                         count++;
3552                 }
3553                 j = ext4_bg_num_gdb(sb, grp);
3554                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3555                         ext4_error(sb, "Invalid number of block group "
3556                                    "descriptor blocks: %d", j);
3557                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3558                 }
3559                 count += j;
3560                 for (; j > 0; j--)
3561                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3562         }
3563         if (!count)
3564                 return 0;
3565         return EXT4_CLUSTERS_PER_GROUP(sb) -
3566                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3567 }
3568
3569 /*
3570  * Compute the overhead and stash it in sbi->s_overhead
3571  */
3572 int ext4_calculate_overhead(struct super_block *sb)
3573 {
3574         struct ext4_sb_info *sbi = EXT4_SB(sb);
3575         struct ext4_super_block *es = sbi->s_es;
3576         struct inode *j_inode;
3577         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3578         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3579         ext4_fsblk_t overhead = 0;
3580         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3581
3582         if (!buf)
3583                 return -ENOMEM;
3584
3585         /*
3586          * Compute the overhead (FS structures).  This is constant
3587          * for a given filesystem unless the number of block groups
3588          * changes so we cache the previous value until it does.
3589          */
3590
3591         /*
3592          * All of the blocks before first_data_block are overhead
3593          */
3594         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3595
3596         /*
3597          * Add the overhead found in each block group
3598          */
3599         for (i = 0; i < ngroups; i++) {
3600                 int blks;
3601
3602                 blks = count_overhead(sb, i, buf);
3603                 overhead += blks;
3604                 if (blks)
3605                         memset(buf, 0, PAGE_SIZE);
3606                 cond_resched();
3607         }
3608
3609         /*
3610          * Add the internal journal blocks whether the journal has been
3611          * loaded or not
3612          */
3613         if (sbi->s_journal && !sbi->journal_bdev)
3614                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3615         else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) {
3616                 /* j_inum for internal journal is non-zero */
3617                 j_inode = ext4_get_journal_inode(sb, j_inum);
3618                 if (j_inode) {
3619                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3620                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
3621                         iput(j_inode);
3622                 } else {
3623                         ext4_msg(sb, KERN_ERR, "can't get journal size");
3624                 }
3625         }
3626         sbi->s_overhead = overhead;
3627         smp_wmb();
3628         free_page((unsigned long) buf);
3629         return 0;
3630 }
3631
3632 static void ext4_set_resv_clusters(struct super_block *sb)
3633 {
3634         ext4_fsblk_t resv_clusters;
3635         struct ext4_sb_info *sbi = EXT4_SB(sb);
3636
3637         /*
3638          * There's no need to reserve anything when we aren't using extents.
3639          * The space estimates are exact, there are no unwritten extents,
3640          * hole punching doesn't need new metadata... This is needed especially
3641          * to keep ext2/3 backward compatibility.
3642          */
3643         if (!ext4_has_feature_extents(sb))
3644                 return;
3645         /*
3646          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3647          * This should cover the situations where we can not afford to run
3648          * out of space like for example punch hole, or converting
3649          * unwritten extents in delalloc path. In most cases such
3650          * allocation would require 1, or 2 blocks, higher numbers are
3651          * very rare.
3652          */
3653         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3654                          sbi->s_cluster_bits);
3655
3656         do_div(resv_clusters, 50);
3657         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3658
3659         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3660 }
3661
3662 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3663 {
3664         struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev);
3665         char *orig_data = kstrdup(data, GFP_KERNEL);
3666         struct buffer_head *bh, **group_desc;
3667         struct ext4_super_block *es = NULL;
3668         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3669         struct flex_groups **flex_groups;
3670         ext4_fsblk_t block;
3671         ext4_fsblk_t sb_block = get_sb_block(&data);
3672         ext4_fsblk_t logical_sb_block;
3673         unsigned long offset = 0;
3674         unsigned long journal_devnum = 0;
3675         unsigned long def_mount_opts;
3676         struct inode *root;
3677         const char *descr;
3678         int ret = -ENOMEM;
3679         int blocksize, clustersize;
3680         unsigned int db_count;
3681         unsigned int i;
3682         int needs_recovery, has_huge_files, has_bigalloc;
3683         __u64 blocks_count;
3684         int err = 0;
3685         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3686         ext4_group_t first_not_zeroed;
3687
3688         if ((data && !orig_data) || !sbi)
3689                 goto out_free_base;
3690
3691         sbi->s_daxdev = dax_dev;
3692         sbi->s_blockgroup_lock =
3693                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3694         if (!sbi->s_blockgroup_lock)
3695                 goto out_free_base;
3696
3697         sb->s_fs_info = sbi;
3698         sbi->s_sb = sb;
3699         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3700         sbi->s_sb_block = sb_block;
3701         if (sb->s_bdev->bd_part)
3702                 sbi->s_sectors_written_start =
3703                         part_stat_read(sb->s_bdev->bd_part, sectors[STAT_WRITE]);
3704
3705         /* Cleanup superblock name */
3706         strreplace(sb->s_id, '/', '!');
3707
3708         /* -EINVAL is default */
3709         ret = -EINVAL;
3710         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3711         if (!blocksize) {
3712                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3713                 goto out_fail;
3714         }
3715
3716         /*
3717          * The ext4 superblock will not be buffer aligned for other than 1kB
3718          * block sizes.  We need to calculate the offset from buffer start.
3719          */
3720         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3721                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3722                 offset = do_div(logical_sb_block, blocksize);
3723         } else {
3724                 logical_sb_block = sb_block;
3725         }
3726
3727         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3728                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3729                 goto out_fail;
3730         }
3731         /*
3732          * Note: s_es must be initialized as soon as possible because
3733          *       some ext4 macro-instructions depend on its value
3734          */
3735         es = (struct ext4_super_block *) (bh->b_data + offset);
3736         sbi->s_es = es;
3737         sb->s_magic = le16_to_cpu(es->s_magic);
3738         if (sb->s_magic != EXT4_SUPER_MAGIC)
3739                 goto cantfind_ext4;
3740         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3741
3742         /* Warn if metadata_csum and gdt_csum are both set. */
3743         if (ext4_has_feature_metadata_csum(sb) &&
3744             ext4_has_feature_gdt_csum(sb))
3745                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3746                              "redundant flags; please run fsck.");
3747
3748         /* Check for a known checksum algorithm */
3749         if (!ext4_verify_csum_type(sb, es)) {
3750                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3751                          "unknown checksum algorithm.");
3752                 silent = 1;
3753                 goto cantfind_ext4;
3754         }
3755
3756         /* Load the checksum driver */
3757         sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3758         if (IS_ERR(sbi->s_chksum_driver)) {
3759                 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3760                 ret = PTR_ERR(sbi->s_chksum_driver);
3761                 sbi->s_chksum_driver = NULL;
3762                 goto failed_mount;
3763         }
3764
3765         /* Check superblock checksum */
3766         if (!ext4_superblock_csum_verify(sb, es)) {
3767                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3768                          "invalid superblock checksum.  Run e2fsck?");
3769                 silent = 1;
3770                 ret = -EFSBADCRC;
3771                 goto cantfind_ext4;
3772         }
3773
3774         /* Precompute checksum seed for all metadata */
3775         if (ext4_has_feature_csum_seed(sb))
3776                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3777         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
3778                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3779                                                sizeof(es->s_uuid));
3780
3781         /* Set defaults before we parse the mount options */
3782         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3783         set_opt(sb, INIT_INODE_TABLE);
3784         if (def_mount_opts & EXT4_DEFM_DEBUG)
3785                 set_opt(sb, DEBUG);
3786         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3787                 set_opt(sb, GRPID);
3788         if (def_mount_opts & EXT4_DEFM_UID16)
3789                 set_opt(sb, NO_UID32);
3790         /* xattr user namespace & acls are now defaulted on */
3791         set_opt(sb, XATTR_USER);
3792 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3793         set_opt(sb, POSIX_ACL);
3794 #endif
3795         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3796         if (ext4_has_metadata_csum(sb))
3797                 set_opt(sb, JOURNAL_CHECKSUM);
3798
3799         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3800                 set_opt(sb, JOURNAL_DATA);
3801         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3802                 set_opt(sb, ORDERED_DATA);
3803         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3804                 set_opt(sb, WRITEBACK_DATA);
3805
3806         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3807                 set_opt(sb, ERRORS_PANIC);
3808         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3809                 set_opt(sb, ERRORS_CONT);
3810         else
3811                 set_opt(sb, ERRORS_RO);
3812         /* block_validity enabled by default; disable with noblock_validity */
3813         set_opt(sb, BLOCK_VALIDITY);
3814         if (def_mount_opts & EXT4_DEFM_DISCARD)
3815                 set_opt(sb, DISCARD);
3816
3817         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3818         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3819         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3820         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3821         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3822
3823         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3824                 set_opt(sb, BARRIER);
3825
3826         /*
3827          * enable delayed allocation by default
3828          * Use -o nodelalloc to turn it off
3829          */
3830         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3831             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3832                 set_opt(sb, DELALLOC);
3833
3834         /*
3835          * set default s_li_wait_mult for lazyinit, for the case there is
3836          * no mount option specified.
3837          */
3838         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3839
3840         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3841
3842         if (blocksize == PAGE_SIZE)
3843                 set_opt(sb, DIOREAD_NOLOCK);
3844
3845         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3846             blocksize > EXT4_MAX_BLOCK_SIZE) {
3847                 ext4_msg(sb, KERN_ERR,
3848                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3849                          blocksize, le32_to_cpu(es->s_log_block_size));
3850                 goto failed_mount;
3851         }
3852
3853         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3854                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3855                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3856         } else {
3857                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3858                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3859                 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) {
3860                         ext4_msg(sb, KERN_ERR, "invalid first ino: %u",
3861                                  sbi->s_first_ino);
3862                         goto failed_mount;
3863                 }
3864                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3865                     (!is_power_of_2(sbi->s_inode_size)) ||
3866                     (sbi->s_inode_size > blocksize)) {
3867                         ext4_msg(sb, KERN_ERR,
3868                                "unsupported inode size: %d",
3869                                sbi->s_inode_size);
3870                         ext4_msg(sb, KERN_ERR, "blocksize: %d", blocksize);
3871                         goto failed_mount;
3872                 }
3873                 /*
3874                  * i_atime_extra is the last extra field available for
3875                  * [acm]times in struct ext4_inode. Checking for that
3876                  * field should suffice to ensure we have extra space
3877                  * for all three.
3878                  */
3879                 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) +
3880                         sizeof(((struct ext4_inode *)0)->i_atime_extra)) {
3881                         sb->s_time_gran = 1;
3882                         sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX;
3883                 } else {
3884                         sb->s_time_gran = NSEC_PER_SEC;
3885                         sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX;
3886                 }
3887                 sb->s_time_min = EXT4_TIMESTAMP_MIN;
3888         }
3889         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3890                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3891                         EXT4_GOOD_OLD_INODE_SIZE;
3892                 if (ext4_has_feature_extra_isize(sb)) {
3893                         unsigned v, max = (sbi->s_inode_size -
3894                                            EXT4_GOOD_OLD_INODE_SIZE);
3895
3896                         v = le16_to_cpu(es->s_want_extra_isize);
3897                         if (v > max) {
3898                                 ext4_msg(sb, KERN_ERR,
3899                                          "bad s_want_extra_isize: %d", v);
3900                                 goto failed_mount;
3901                         }
3902                         if (sbi->s_want_extra_isize < v)
3903                                 sbi->s_want_extra_isize = v;
3904
3905                         v = le16_to_cpu(es->s_min_extra_isize);
3906                         if (v > max) {
3907                                 ext4_msg(sb, KERN_ERR,
3908                                          "bad s_min_extra_isize: %d", v);
3909                                 goto failed_mount;
3910                         }
3911                         if (sbi->s_want_extra_isize < v)
3912                                 sbi->s_want_extra_isize = v;
3913                 }
3914         }
3915
3916         if (sbi->s_es->s_mount_opts[0]) {
3917                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3918                                               sizeof(sbi->s_es->s_mount_opts),
3919                                               GFP_KERNEL);
3920                 if (!s_mount_opts)
3921                         goto failed_mount;
3922                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3923                                    &journal_ioprio, 0)) {
3924                         ext4_msg(sb, KERN_WARNING,
3925                                  "failed to parse options in superblock: %s",
3926                                  s_mount_opts);
3927                 }
3928                 kfree(s_mount_opts);
3929         }
3930         sbi->s_def_mount_opt = sbi->s_mount_opt;
3931         if (!parse_options((char *) data, sb, &journal_devnum,
3932                            &journal_ioprio, 0))
3933                 goto failed_mount;
3934
3935 #ifdef CONFIG_UNICODE
3936         if (ext4_has_feature_casefold(sb) && !sbi->s_encoding) {
3937                 const struct ext4_sb_encodings *encoding_info;
3938                 struct unicode_map *encoding;
3939                 __u16 encoding_flags;
3940
3941                 if (ext4_has_feature_encrypt(sb)) {
3942                         ext4_msg(sb, KERN_ERR,
3943                                  "Can't mount with encoding and encryption");
3944                         goto failed_mount;
3945                 }
3946
3947                 if (ext4_sb_read_encoding(es, &encoding_info,
3948                                           &encoding_flags)) {
3949                         ext4_msg(sb, KERN_ERR,
3950                                  "Encoding requested by superblock is unknown");
3951                         goto failed_mount;
3952                 }
3953
3954                 encoding = utf8_load(encoding_info->version);
3955                 if (IS_ERR(encoding)) {
3956                         ext4_msg(sb, KERN_ERR,
3957                                  "can't mount with superblock charset: %s-%s "
3958                                  "not supported by the kernel. flags: 0x%x.",
3959                                  encoding_info->name, encoding_info->version,
3960                                  encoding_flags);
3961                         goto failed_mount;
3962                 }
3963                 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: "
3964                          "%s-%s with flags 0x%hx", encoding_info->name,
3965                          encoding_info->version?:"\b", encoding_flags);
3966
3967                 sbi->s_encoding = encoding;
3968                 sbi->s_encoding_flags = encoding_flags;
3969         }
3970 #endif
3971
3972         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3973                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with data=journal disables delayed allocation, dioread_nolock, and O_DIRECT support!\n");
3974                 /* can't mount with both data=journal and dioread_nolock. */
3975                 clear_opt(sb, DIOREAD_NOLOCK);
3976                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3977                         ext4_msg(sb, KERN_ERR, "can't mount with "
3978                                  "both data=journal and delalloc");
3979                         goto failed_mount;
3980                 }
3981                 if (test_opt(sb, DAX)) {
3982                         ext4_msg(sb, KERN_ERR, "can't mount with "
3983                                  "both data=journal and dax");
3984                         goto failed_mount;
3985                 }
3986                 if (ext4_has_feature_encrypt(sb)) {
3987                         ext4_msg(sb, KERN_WARNING,
3988                                  "encrypted files will use data=ordered "
3989                                  "instead of data journaling mode");
3990                 }
3991                 if (test_opt(sb, DELALLOC))
3992                         clear_opt(sb, DELALLOC);
3993         } else {
3994                 sb->s_iflags |= SB_I_CGROUPWB;
3995         }
3996
3997         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
3998                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
3999
4000         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
4001             (ext4_has_compat_features(sb) ||
4002              ext4_has_ro_compat_features(sb) ||
4003              ext4_has_incompat_features(sb)))
4004                 ext4_msg(sb, KERN_WARNING,
4005                        "feature flags set on rev 0 fs, "
4006                        "running e2fsck is recommended");
4007
4008         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
4009                 set_opt2(sb, HURD_COMPAT);
4010                 if (ext4_has_feature_64bit(sb)) {
4011                         ext4_msg(sb, KERN_ERR,
4012                                  "The Hurd can't support 64-bit file systems");
4013                         goto failed_mount;
4014                 }
4015
4016                 /*
4017                  * ea_inode feature uses l_i_version field which is not
4018                  * available in HURD_COMPAT mode.
4019                  */
4020                 if (ext4_has_feature_ea_inode(sb)) {
4021                         ext4_msg(sb, KERN_ERR,
4022                                  "ea_inode feature is not supported for Hurd");
4023                         goto failed_mount;
4024                 }
4025         }
4026
4027         if (IS_EXT2_SB(sb)) {
4028                 if (ext2_feature_set_ok(sb))
4029                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
4030                                  "using the ext4 subsystem");
4031                 else {
4032                         /*
4033                          * If we're probing be silent, if this looks like
4034                          * it's actually an ext[34] filesystem.
4035                          */
4036                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4037                                 goto failed_mount;
4038                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
4039                                  "to feature incompatibilities");
4040                         goto failed_mount;
4041                 }
4042         }
4043
4044         if (IS_EXT3_SB(sb)) {
4045                 if (ext3_feature_set_ok(sb))
4046                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
4047                                  "using the ext4 subsystem");
4048                 else {
4049                         /*
4050                          * If we're probing be silent, if this looks like
4051                          * it's actually an ext4 filesystem.
4052                          */
4053                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4054                                 goto failed_mount;
4055                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
4056                                  "to feature incompatibilities");
4057                         goto failed_mount;
4058                 }
4059         }
4060
4061         /*
4062          * Check feature flags regardless of the revision level, since we
4063          * previously didn't change the revision level when setting the flags,
4064          * so there is a chance incompat flags are set on a rev 0 filesystem.
4065          */
4066         if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
4067                 goto failed_mount;
4068
4069         if (le32_to_cpu(es->s_log_block_size) >
4070             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4071                 ext4_msg(sb, KERN_ERR,
4072                          "Invalid log block size: %u",
4073                          le32_to_cpu(es->s_log_block_size));
4074                 goto failed_mount;
4075         }
4076         if (le32_to_cpu(es->s_log_cluster_size) >
4077             (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4078                 ext4_msg(sb, KERN_ERR,
4079                          "Invalid log cluster size: %u",
4080                          le32_to_cpu(es->s_log_cluster_size));
4081                 goto failed_mount;
4082         }
4083
4084         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
4085                 ext4_msg(sb, KERN_ERR,
4086                          "Number of reserved GDT blocks insanely large: %d",
4087                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
4088                 goto failed_mount;
4089         }
4090
4091         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
4092                 if (ext4_has_feature_inline_data(sb)) {
4093                         ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
4094                                         " that may contain inline data");
4095                         goto failed_mount;
4096                 }
4097                 if (!bdev_dax_supported(sb->s_bdev, blocksize)) {
4098                         ext4_msg(sb, KERN_ERR,
4099                                 "DAX unsupported by block device.");
4100                         goto failed_mount;
4101                 }
4102         }
4103
4104         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
4105                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
4106                          es->s_encryption_level);
4107                 goto failed_mount;
4108         }
4109
4110         if (sb->s_blocksize != blocksize) {
4111                 /* Validate the filesystem blocksize */
4112                 if (!sb_set_blocksize(sb, blocksize)) {
4113                         ext4_msg(sb, KERN_ERR, "bad block size %d",
4114                                         blocksize);
4115                         goto failed_mount;
4116                 }
4117
4118                 brelse(bh);
4119                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
4120                 offset = do_div(logical_sb_block, blocksize);
4121                 bh = sb_bread_unmovable(sb, logical_sb_block);
4122                 if (!bh) {
4123                         ext4_msg(sb, KERN_ERR,
4124                                "Can't read superblock on 2nd try");
4125                         goto failed_mount;
4126                 }
4127                 es = (struct ext4_super_block *)(bh->b_data + offset);
4128                 sbi->s_es = es;
4129                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
4130                         ext4_msg(sb, KERN_ERR,
4131                                "Magic mismatch, very weird!");
4132                         goto failed_mount;
4133                 }
4134         }
4135
4136         has_huge_files = ext4_has_feature_huge_file(sb);
4137         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
4138                                                       has_huge_files);
4139         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
4140
4141         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
4142         if (ext4_has_feature_64bit(sb)) {
4143                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
4144                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
4145                     !is_power_of_2(sbi->s_desc_size)) {
4146                         ext4_msg(sb, KERN_ERR,
4147                                "unsupported descriptor size %lu",
4148                                sbi->s_desc_size);
4149                         goto failed_mount;
4150                 }
4151         } else
4152                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
4153
4154         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
4155         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
4156
4157         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
4158         if (sbi->s_inodes_per_block == 0)
4159                 goto cantfind_ext4;
4160         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
4161             sbi->s_inodes_per_group > blocksize * 8) {
4162                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
4163                          sbi->s_inodes_per_group);
4164                 goto failed_mount;
4165         }
4166         sbi->s_itb_per_group = sbi->s_inodes_per_group /
4167                                         sbi->s_inodes_per_block;
4168         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
4169         sbi->s_sbh = bh;
4170         sbi->s_mount_state = le16_to_cpu(es->s_state);
4171         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
4172         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
4173
4174         for (i = 0; i < 4; i++)
4175                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
4176         sbi->s_def_hash_version = es->s_def_hash_version;
4177         if (ext4_has_feature_dir_index(sb)) {
4178                 i = le32_to_cpu(es->s_flags);
4179                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
4180                         sbi->s_hash_unsigned = 3;
4181                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
4182 #ifdef __CHAR_UNSIGNED__
4183                         if (!sb_rdonly(sb))
4184                                 es->s_flags |=
4185                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
4186                         sbi->s_hash_unsigned = 3;
4187 #else
4188                         if (!sb_rdonly(sb))
4189                                 es->s_flags |=
4190                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
4191 #endif
4192                 }
4193         }
4194
4195         /* Handle clustersize */
4196         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
4197         has_bigalloc = ext4_has_feature_bigalloc(sb);
4198         if (has_bigalloc) {
4199                 if (clustersize < blocksize) {
4200                         ext4_msg(sb, KERN_ERR,
4201                                  "cluster size (%d) smaller than "
4202                                  "block size (%d)", clustersize, blocksize);
4203                         goto failed_mount;
4204                 }
4205                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
4206                         le32_to_cpu(es->s_log_block_size);
4207                 sbi->s_clusters_per_group =
4208                         le32_to_cpu(es->s_clusters_per_group);
4209                 if (sbi->s_clusters_per_group > blocksize * 8) {
4210                         ext4_msg(sb, KERN_ERR,
4211                                  "#clusters per group too big: %lu",
4212                                  sbi->s_clusters_per_group);
4213                         goto failed_mount;
4214                 }
4215                 if (sbi->s_blocks_per_group !=
4216                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
4217                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
4218                                  "clusters per group (%lu) inconsistent",
4219                                  sbi->s_blocks_per_group,
4220                                  sbi->s_clusters_per_group);
4221                         goto failed_mount;
4222                 }
4223         } else {
4224                 if (clustersize != blocksize) {
4225                         ext4_msg(sb, KERN_ERR,
4226                                  "fragment/cluster size (%d) != "
4227                                  "block size (%d)", clustersize, blocksize);
4228                         goto failed_mount;
4229                 }
4230                 if (sbi->s_blocks_per_group > blocksize * 8) {
4231                         ext4_msg(sb, KERN_ERR,
4232                                  "#blocks per group too big: %lu",
4233                                  sbi->s_blocks_per_group);
4234                         goto failed_mount;
4235                 }
4236                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
4237                 sbi->s_cluster_bits = 0;
4238         }
4239         sbi->s_cluster_ratio = clustersize / blocksize;
4240
4241         /* Do we have standard group size of clustersize * 8 blocks ? */
4242         if (sbi->s_blocks_per_group == clustersize << 3)
4243                 set_opt2(sb, STD_GROUP_SIZE);
4244
4245         /*
4246          * Test whether we have more sectors than will fit in sector_t,
4247          * and whether the max offset is addressable by the page cache.
4248          */
4249         err = generic_check_addressable(sb->s_blocksize_bits,
4250                                         ext4_blocks_count(es));
4251         if (err) {
4252                 ext4_msg(sb, KERN_ERR, "filesystem"
4253                          " too large to mount safely on this system");
4254                 goto failed_mount;
4255         }
4256
4257         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
4258                 goto cantfind_ext4;
4259
4260         /* check blocks count against device size */
4261         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
4262         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
4263                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
4264                        "exceeds size of device (%llu blocks)",
4265                        ext4_blocks_count(es), blocks_count);
4266                 goto failed_mount;
4267         }
4268
4269         /*
4270          * It makes no sense for the first data block to be beyond the end
4271          * of the filesystem.
4272          */
4273         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
4274                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4275                          "block %u is beyond end of filesystem (%llu)",
4276                          le32_to_cpu(es->s_first_data_block),
4277                          ext4_blocks_count(es));
4278                 goto failed_mount;
4279         }
4280         if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) &&
4281             (sbi->s_cluster_ratio == 1)) {
4282                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4283                          "block is 0 with a 1k block and cluster size");
4284                 goto failed_mount;
4285         }
4286
4287         blocks_count = (ext4_blocks_count(es) -
4288                         le32_to_cpu(es->s_first_data_block) +
4289                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
4290         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
4291         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
4292                 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu "
4293                        "(block count %llu, first data block %u, "
4294                        "blocks per group %lu)", blocks_count,
4295                        ext4_blocks_count(es),
4296                        le32_to_cpu(es->s_first_data_block),
4297                        EXT4_BLOCKS_PER_GROUP(sb));
4298                 goto failed_mount;
4299         }
4300         sbi->s_groups_count = blocks_count;
4301         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
4302                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
4303         if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) !=
4304             le32_to_cpu(es->s_inodes_count)) {
4305                 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu",
4306                          le32_to_cpu(es->s_inodes_count),
4307                          ((u64)sbi->s_groups_count * sbi->s_inodes_per_group));
4308                 ret = -EINVAL;
4309                 goto failed_mount;
4310         }
4311         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
4312                    EXT4_DESC_PER_BLOCK(sb);
4313         if (ext4_has_feature_meta_bg(sb)) {
4314                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
4315                         ext4_msg(sb, KERN_WARNING,
4316                                  "first meta block group too large: %u "
4317                                  "(group descriptor block count %u)",
4318                                  le32_to_cpu(es->s_first_meta_bg), db_count);
4319                         goto failed_mount;
4320                 }
4321         }
4322         rcu_assign_pointer(sbi->s_group_desc,
4323                            kvmalloc_array(db_count,
4324                                           sizeof(struct buffer_head *),
4325                                           GFP_KERNEL));
4326         if (sbi->s_group_desc == NULL) {
4327                 ext4_msg(sb, KERN_ERR, "not enough memory");
4328                 ret = -ENOMEM;
4329                 goto failed_mount;
4330         }
4331
4332         bgl_lock_init(sbi->s_blockgroup_lock);
4333
4334         /* Pre-read the descriptors into the buffer cache */
4335         for (i = 0; i < db_count; i++) {
4336                 block = descriptor_loc(sb, logical_sb_block, i);
4337                 sb_breadahead_unmovable(sb, block);
4338         }
4339
4340         for (i = 0; i < db_count; i++) {
4341                 struct buffer_head *bh;
4342
4343                 block = descriptor_loc(sb, logical_sb_block, i);
4344                 bh = sb_bread_unmovable(sb, block);
4345                 if (!bh) {
4346                         ext4_msg(sb, KERN_ERR,
4347                                "can't read group descriptor %d", i);
4348                         db_count = i;
4349                         goto failed_mount2;
4350                 }
4351                 rcu_read_lock();
4352                 rcu_dereference(sbi->s_group_desc)[i] = bh;
4353                 rcu_read_unlock();
4354         }
4355         sbi->s_gdb_count = db_count;
4356         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
4357                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
4358                 ret = -EFSCORRUPTED;
4359                 goto failed_mount2;
4360         }
4361
4362         timer_setup(&sbi->s_err_report, print_daily_error_info, 0);
4363
4364         /* Register extent status tree shrinker */
4365         if (ext4_es_register_shrinker(sbi))
4366                 goto failed_mount3;
4367
4368         sbi->s_stripe = ext4_get_stripe_size(sbi);
4369         sbi->s_extent_max_zeroout_kb = 32;
4370
4371         /*
4372          * set up enough so that it can read an inode
4373          */
4374         sb->s_op = &ext4_sops;
4375         sb->s_export_op = &ext4_export_ops;
4376         sb->s_xattr = ext4_xattr_handlers;
4377 #ifdef CONFIG_FS_ENCRYPTION
4378         sb->s_cop = &ext4_cryptops;
4379 #endif
4380 #ifdef CONFIG_FS_VERITY
4381         sb->s_vop = &ext4_verityops;
4382 #endif
4383 #ifdef CONFIG_QUOTA
4384         sb->dq_op = &ext4_quota_operations;
4385         if (ext4_has_feature_quota(sb))
4386                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4387         else
4388                 sb->s_qcop = &ext4_qctl_operations;
4389         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
4390 #endif
4391         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
4392
4393         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
4394         mutex_init(&sbi->s_orphan_lock);
4395
4396         sb->s_root = NULL;
4397
4398         needs_recovery = (es->s_last_orphan != 0 ||
4399                           ext4_has_feature_journal_needs_recovery(sb));
4400
4401         if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb))
4402                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
4403                         goto failed_mount3a;
4404
4405         /*
4406          * The first inode we look at is the journal inode.  Don't try
4407          * root first: it may be modified in the journal!
4408          */
4409         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
4410                 err = ext4_load_journal(sb, es, journal_devnum);
4411                 if (err)
4412                         goto failed_mount3a;
4413         } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
4414                    ext4_has_feature_journal_needs_recovery(sb)) {
4415                 ext4_msg(sb, KERN_ERR, "required journal recovery "
4416                        "suppressed and not mounted read-only");
4417                 goto failed_mount_wq;
4418         } else {
4419                 /* Nojournal mode, all journal mount options are illegal */
4420                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
4421                         ext4_msg(sb, KERN_ERR, "can't mount with "
4422                                  "journal_checksum, fs mounted w/o journal");
4423                         goto failed_mount_wq;
4424                 }
4425                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4426                         ext4_msg(sb, KERN_ERR, "can't mount with "
4427                                  "journal_async_commit, fs mounted w/o journal");
4428                         goto failed_mount_wq;
4429                 }
4430                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
4431                         ext4_msg(sb, KERN_ERR, "can't mount with "
4432                                  "commit=%lu, fs mounted w/o journal",
4433                                  sbi->s_commit_interval / HZ);
4434                         goto failed_mount_wq;
4435                 }
4436                 if (EXT4_MOUNT_DATA_FLAGS &
4437                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
4438                         ext4_msg(sb, KERN_ERR, "can't mount with "
4439                                  "data=, fs mounted w/o journal");
4440                         goto failed_mount_wq;
4441                 }
4442                 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM;
4443                 clear_opt(sb, JOURNAL_CHECKSUM);
4444                 clear_opt(sb, DATA_FLAGS);
4445                 sbi->s_journal = NULL;
4446                 needs_recovery = 0;
4447                 goto no_journal;
4448         }
4449
4450         if (ext4_has_feature_64bit(sb) &&
4451             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4452                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4453                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4454                 goto failed_mount_wq;
4455         }
4456
4457         if (!set_journal_csum_feature_set(sb)) {
4458                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4459                          "feature set");
4460                 goto failed_mount_wq;
4461         }
4462
4463         /* We have now updated the journal if required, so we can
4464          * validate the data journaling mode. */
4465         switch (test_opt(sb, DATA_FLAGS)) {
4466         case 0:
4467                 /* No mode set, assume a default based on the journal
4468                  * capabilities: ORDERED_DATA if the journal can
4469                  * cope, else JOURNAL_DATA
4470                  */
4471                 if (jbd2_journal_check_available_features
4472                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4473                         set_opt(sb, ORDERED_DATA);
4474                         sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
4475                 } else {
4476                         set_opt(sb, JOURNAL_DATA);
4477                         sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
4478                 }
4479                 break;
4480
4481         case EXT4_MOUNT_ORDERED_DATA:
4482         case EXT4_MOUNT_WRITEBACK_DATA:
4483                 if (!jbd2_journal_check_available_features
4484                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4485                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4486                                "requested data journaling mode");
4487                         goto failed_mount_wq;
4488                 }
4489         default:
4490                 break;
4491         }
4492
4493         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4494             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4495                 ext4_msg(sb, KERN_ERR, "can't mount with "
4496                         "journal_async_commit in data=ordered mode");
4497                 goto failed_mount_wq;
4498         }
4499
4500         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4501
4502         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4503
4504 no_journal:
4505         if (!test_opt(sb, NO_MBCACHE)) {
4506                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
4507                 if (!sbi->s_ea_block_cache) {
4508                         ext4_msg(sb, KERN_ERR,
4509                                  "Failed to create ea_block_cache");
4510                         goto failed_mount_wq;
4511                 }
4512
4513                 if (ext4_has_feature_ea_inode(sb)) {
4514                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
4515                         if (!sbi->s_ea_inode_cache) {
4516                                 ext4_msg(sb, KERN_ERR,
4517                                          "Failed to create ea_inode_cache");
4518                                 goto failed_mount_wq;
4519                         }
4520                 }
4521         }
4522
4523         if (ext4_has_feature_verity(sb) && blocksize != PAGE_SIZE) {
4524                 ext4_msg(sb, KERN_ERR, "Unsupported blocksize for fs-verity");
4525                 goto failed_mount_wq;
4526         }
4527
4528         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) &&
4529             !ext4_has_feature_encrypt(sb)) {
4530                 ext4_set_feature_encrypt(sb);
4531                 ext4_commit_super(sb, 1);
4532         }
4533
4534         /*
4535          * Get the # of file system overhead blocks from the
4536          * superblock if present.
4537          */
4538         if (es->s_overhead_clusters)
4539                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4540         else {
4541                 err = ext4_calculate_overhead(sb);
4542                 if (err)
4543                         goto failed_mount_wq;
4544         }
4545
4546         /*
4547          * The maximum number of concurrent works can be high and
4548          * concurrency isn't really necessary.  Limit it to 1.
4549          */
4550         EXT4_SB(sb)->rsv_conversion_wq =
4551                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4552         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4553                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4554                 ret = -ENOMEM;
4555                 goto failed_mount4;
4556         }
4557
4558         /*
4559          * The jbd2_journal_load will have done any necessary log recovery,
4560          * so we can safely mount the rest of the filesystem now.
4561          */
4562
4563         root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL);
4564         if (IS_ERR(root)) {
4565                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4566                 ret = PTR_ERR(root);
4567                 root = NULL;
4568                 goto failed_mount4;
4569         }
4570         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4571                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4572                 iput(root);
4573                 goto failed_mount4;
4574         }
4575
4576 #ifdef CONFIG_UNICODE
4577         if (sbi->s_encoding)
4578                 sb->s_d_op = &ext4_dentry_ops;
4579 #endif
4580
4581         sb->s_root = d_make_root(root);
4582         if (!sb->s_root) {
4583                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4584                 ret = -ENOMEM;
4585                 goto failed_mount4;
4586         }
4587
4588         ret = ext4_setup_super(sb, es, sb_rdonly(sb));
4589         if (ret == -EROFS) {
4590                 sb->s_flags |= SB_RDONLY;
4591                 ret = 0;
4592         } else if (ret)
4593                 goto failed_mount4a;
4594
4595         ext4_set_resv_clusters(sb);
4596
4597         err = ext4_setup_system_zone(sb);
4598         if (err) {
4599                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4600                          "zone (%d)", err);
4601                 goto failed_mount4a;
4602         }
4603
4604         ext4_ext_init(sb);
4605         err = ext4_mb_init(sb);
4606         if (err) {
4607                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4608                          err);
4609                 goto failed_mount5;
4610         }
4611
4612         block = ext4_count_free_clusters(sb);
4613         ext4_free_blocks_count_set(sbi->s_es,
4614                                    EXT4_C2B(sbi, block));
4615         ext4_superblock_csum_set(sb);
4616         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4617                                   GFP_KERNEL);
4618         if (!err) {
4619                 unsigned long freei = ext4_count_free_inodes(sb);
4620                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4621                 ext4_superblock_csum_set(sb);
4622                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4623                                           GFP_KERNEL);
4624         }
4625         if (!err)
4626                 err = percpu_counter_init(&sbi->s_dirs_counter,
4627                                           ext4_count_dirs(sb), GFP_KERNEL);
4628         if (!err)
4629                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4630                                           GFP_KERNEL);
4631         if (!err)
4632                 err = percpu_init_rwsem(&sbi->s_writepages_rwsem);
4633
4634         if (err) {
4635                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4636                 goto failed_mount6;
4637         }
4638
4639         if (ext4_has_feature_flex_bg(sb))
4640                 if (!ext4_fill_flex_info(sb)) {
4641                         ext4_msg(sb, KERN_ERR,
4642                                "unable to initialize "
4643                                "flex_bg meta info!");
4644                         goto failed_mount6;
4645                 }
4646
4647         err = ext4_register_li_request(sb, first_not_zeroed);
4648         if (err)
4649                 goto failed_mount6;
4650
4651         err = ext4_register_sysfs(sb);
4652         if (err)
4653                 goto failed_mount7;
4654
4655 #ifdef CONFIG_QUOTA
4656         /* Enable quota usage during mount. */
4657         if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
4658                 err = ext4_enable_quotas(sb);
4659                 if (err)
4660                         goto failed_mount8;
4661         }
4662 #endif  /* CONFIG_QUOTA */
4663
4664         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4665         ext4_orphan_cleanup(sb, es);
4666         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4667         if (needs_recovery) {
4668                 ext4_msg(sb, KERN_INFO, "recovery complete");
4669                 ext4_mark_recovery_complete(sb, es);
4670         }
4671         if (EXT4_SB(sb)->s_journal) {
4672                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4673                         descr = " journalled data mode";
4674                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4675                         descr = " ordered data mode";
4676                 else
4677                         descr = " writeback data mode";
4678         } else
4679                 descr = "out journal";
4680
4681         if (test_opt(sb, DISCARD)) {
4682                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4683                 if (!blk_queue_discard(q))
4684                         ext4_msg(sb, KERN_WARNING,
4685                                  "mounting with \"discard\" option, but "
4686                                  "the device does not support discard");
4687         }
4688
4689         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4690                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4691                          "Opts: %.*s%s%s", descr,
4692                          (int) sizeof(sbi->s_es->s_mount_opts),
4693                          sbi->s_es->s_mount_opts,
4694                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4695
4696         if (es->s_error_count)
4697                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4698
4699         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4700         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4701         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4702         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4703
4704         kfree(orig_data);
4705         return 0;
4706
4707 cantfind_ext4:
4708         if (!silent)
4709                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4710         goto failed_mount;
4711
4712 #ifdef CONFIG_QUOTA
4713 failed_mount8:
4714         ext4_unregister_sysfs(sb);
4715 #endif
4716 failed_mount7:
4717         ext4_unregister_li_request(sb);
4718 failed_mount6:
4719         ext4_mb_release(sb);
4720         rcu_read_lock();
4721         flex_groups = rcu_dereference(sbi->s_flex_groups);
4722         if (flex_groups) {
4723                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
4724                         kvfree(flex_groups[i]);
4725                 kvfree(flex_groups);
4726         }
4727         rcu_read_unlock();
4728         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4729         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4730         percpu_counter_destroy(&sbi->s_dirs_counter);
4731         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4732         percpu_free_rwsem(&sbi->s_writepages_rwsem);
4733 failed_mount5:
4734         ext4_ext_release(sb);
4735         ext4_release_system_zone(sb);
4736 failed_mount4a:
4737         dput(sb->s_root);
4738         sb->s_root = NULL;
4739 failed_mount4:
4740         ext4_msg(sb, KERN_ERR, "mount failed");
4741         if (EXT4_SB(sb)->rsv_conversion_wq)
4742                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4743 failed_mount_wq:
4744         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
4745         sbi->s_ea_inode_cache = NULL;
4746
4747         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
4748         sbi->s_ea_block_cache = NULL;
4749
4750         if (sbi->s_journal) {
4751                 jbd2_journal_destroy(sbi->s_journal);
4752                 sbi->s_journal = NULL;
4753         }
4754 failed_mount3a:
4755         ext4_es_unregister_shrinker(sbi);
4756 failed_mount3:
4757         del_timer_sync(&sbi->s_err_report);
4758         if (sbi->s_mmp_tsk)
4759                 kthread_stop(sbi->s_mmp_tsk);
4760 failed_mount2:
4761         rcu_read_lock();
4762         group_desc = rcu_dereference(sbi->s_group_desc);
4763         for (i = 0; i < db_count; i++)
4764                 brelse(group_desc[i]);
4765         kvfree(group_desc);
4766         rcu_read_unlock();
4767 failed_mount:
4768         if (sbi->s_chksum_driver)
4769                 crypto_free_shash(sbi->s_chksum_driver);
4770
4771 #ifdef CONFIG_UNICODE
4772         utf8_unload(sbi->s_encoding);
4773 #endif
4774
4775 #ifdef CONFIG_QUOTA
4776         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4777                 kfree(get_qf_name(sb, sbi, i));
4778 #endif
4779         ext4_blkdev_remove(sbi);
4780         brelse(bh);
4781 out_fail:
4782         sb->s_fs_info = NULL;
4783         kfree(sbi->s_blockgroup_lock);
4784 out_free_base:
4785         kfree(sbi);
4786         kfree(orig_data);
4787         fs_put_dax(dax_dev);
4788         return err ? err : ret;
4789 }
4790
4791 /*
4792  * Setup any per-fs journal parameters now.  We'll do this both on
4793  * initial mount, once the journal has been initialised but before we've
4794  * done any recovery; and again on any subsequent remount.
4795  */
4796 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4797 {
4798         struct ext4_sb_info *sbi = EXT4_SB(sb);
4799
4800         journal->j_commit_interval = sbi->s_commit_interval;
4801         journal->j_min_batch_time = sbi->s_min_batch_time;
4802         journal->j_max_batch_time = sbi->s_max_batch_time;
4803
4804         write_lock(&journal->j_state_lock);
4805         if (test_opt(sb, BARRIER))
4806                 journal->j_flags |= JBD2_BARRIER;
4807         else
4808                 journal->j_flags &= ~JBD2_BARRIER;
4809         if (test_opt(sb, DATA_ERR_ABORT))
4810                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4811         else
4812                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4813         write_unlock(&journal->j_state_lock);
4814 }
4815
4816 static struct inode *ext4_get_journal_inode(struct super_block *sb,
4817                                              unsigned int journal_inum)
4818 {
4819         struct inode *journal_inode;
4820
4821         /*
4822          * Test for the existence of a valid inode on disk.  Bad things
4823          * happen if we iget() an unused inode, as the subsequent iput()
4824          * will try to delete it.
4825          */
4826         journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL);
4827         if (IS_ERR(journal_inode)) {
4828                 ext4_msg(sb, KERN_ERR, "no journal found");
4829                 return NULL;
4830         }
4831         if (!journal_inode->i_nlink) {
4832                 make_bad_inode(journal_inode);
4833                 iput(journal_inode);
4834                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4835                 return NULL;
4836         }
4837
4838         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4839                   journal_inode, journal_inode->i_size);
4840         if (!S_ISREG(journal_inode->i_mode)) {
4841                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4842                 iput(journal_inode);
4843                 return NULL;
4844         }
4845         return journal_inode;
4846 }
4847
4848 static journal_t *ext4_get_journal(struct super_block *sb,
4849                                    unsigned int journal_inum)
4850 {
4851         struct inode *journal_inode;
4852         journal_t *journal;
4853
4854         BUG_ON(!ext4_has_feature_journal(sb));
4855
4856         journal_inode = ext4_get_journal_inode(sb, journal_inum);
4857         if (!journal_inode)
4858                 return NULL;
4859
4860         journal = jbd2_journal_init_inode(journal_inode);
4861         if (!journal) {
4862                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4863                 iput(journal_inode);
4864                 return NULL;
4865         }
4866         journal->j_private = sb;
4867         ext4_init_journal_params(sb, journal);
4868         return journal;
4869 }
4870
4871 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4872                                        dev_t j_dev)
4873 {
4874         struct buffer_head *bh;
4875         journal_t *journal;
4876         ext4_fsblk_t start;
4877         ext4_fsblk_t len;
4878         int hblock, blocksize;
4879         ext4_fsblk_t sb_block;
4880         unsigned long offset;
4881         struct ext4_super_block *es;
4882         struct block_device *bdev;
4883
4884         BUG_ON(!ext4_has_feature_journal(sb));
4885
4886         bdev = ext4_blkdev_get(j_dev, sb);
4887         if (bdev == NULL)
4888                 return NULL;
4889
4890         blocksize = sb->s_blocksize;
4891         hblock = bdev_logical_block_size(bdev);
4892         if (blocksize < hblock) {
4893                 ext4_msg(sb, KERN_ERR,
4894                         "blocksize too small for journal device");
4895                 goto out_bdev;
4896         }
4897
4898         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4899         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4900         set_blocksize(bdev, blocksize);
4901         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4902                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4903                        "external journal");
4904                 goto out_bdev;
4905         }
4906
4907         es = (struct ext4_super_block *) (bh->b_data + offset);
4908         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4909             !(le32_to_cpu(es->s_feature_incompat) &
4910               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4911                 ext4_msg(sb, KERN_ERR, "external journal has "
4912                                         "bad superblock");
4913                 brelse(bh);
4914                 goto out_bdev;
4915         }
4916
4917         if ((le32_to_cpu(es->s_feature_ro_compat) &
4918              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4919             es->s_checksum != ext4_superblock_csum(sb, es)) {
4920                 ext4_msg(sb, KERN_ERR, "external journal has "
4921                                        "corrupt superblock");
4922                 brelse(bh);
4923                 goto out_bdev;
4924         }
4925
4926         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4927                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4928                 brelse(bh);
4929                 goto out_bdev;
4930         }
4931
4932         len = ext4_blocks_count(es);
4933         start = sb_block + 1;
4934         brelse(bh);     /* we're done with the superblock */
4935
4936         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4937                                         start, len, blocksize);
4938         if (!journal) {
4939                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4940                 goto out_bdev;
4941         }
4942         journal->j_private = sb;
4943         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4944         wait_on_buffer(journal->j_sb_buffer);
4945         if (!buffer_uptodate(journal->j_sb_buffer)) {
4946                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4947                 goto out_journal;
4948         }
4949         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4950                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4951                                         "user (unsupported) - %d",
4952                         be32_to_cpu(journal->j_superblock->s_nr_users));
4953                 goto out_journal;
4954         }
4955         EXT4_SB(sb)->journal_bdev = bdev;
4956         ext4_init_journal_params(sb, journal);
4957         return journal;
4958
4959 out_journal:
4960         jbd2_journal_destroy(journal);
4961 out_bdev:
4962         ext4_blkdev_put(bdev);
4963         return NULL;
4964 }
4965
4966 static int ext4_load_journal(struct super_block *sb,
4967                              struct ext4_super_block *es,
4968                              unsigned long journal_devnum)
4969 {
4970         journal_t *journal;
4971         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4972         dev_t journal_dev;
4973         int err = 0;
4974         int really_read_only;
4975
4976         BUG_ON(!ext4_has_feature_journal(sb));
4977
4978         if (journal_devnum &&
4979             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4980                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4981                         "numbers have changed");
4982                 journal_dev = new_decode_dev(journal_devnum);
4983         } else
4984                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4985
4986         really_read_only = bdev_read_only(sb->s_bdev);
4987
4988         /*
4989          * Are we loading a blank journal or performing recovery after a
4990          * crash?  For recovery, we need to check in advance whether we
4991          * can get read-write access to the device.
4992          */
4993         if (ext4_has_feature_journal_needs_recovery(sb)) {
4994                 if (sb_rdonly(sb)) {
4995                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4996                                         "required on readonly filesystem");
4997                         if (really_read_only) {
4998                                 ext4_msg(sb, KERN_ERR, "write access "
4999                                         "unavailable, cannot proceed "
5000                                         "(try mounting with noload)");
5001                                 return -EROFS;
5002                         }
5003                         ext4_msg(sb, KERN_INFO, "write access will "
5004                                "be enabled during recovery");
5005                 }
5006         }
5007
5008         if (journal_inum && journal_dev) {
5009                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
5010                        "and inode journals!");
5011                 return -EINVAL;
5012         }
5013
5014         if (journal_inum) {
5015                 if (!(journal = ext4_get_journal(sb, journal_inum)))
5016                         return -EINVAL;
5017         } else {
5018                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
5019                         return -EINVAL;
5020         }
5021
5022         if (!(journal->j_flags & JBD2_BARRIER))
5023                 ext4_msg(sb, KERN_INFO, "barriers disabled");
5024
5025         if (!ext4_has_feature_journal_needs_recovery(sb))
5026                 err = jbd2_journal_wipe(journal, !really_read_only);
5027         if (!err) {
5028                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
5029                 if (save)
5030                         memcpy(save, ((char *) es) +
5031                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
5032                 err = jbd2_journal_load(journal);
5033                 if (save)
5034                         memcpy(((char *) es) + EXT4_S_ERR_START,
5035                                save, EXT4_S_ERR_LEN);
5036                 kfree(save);
5037         }
5038
5039         if (err) {
5040                 ext4_msg(sb, KERN_ERR, "error loading journal");
5041                 jbd2_journal_destroy(journal);
5042                 return err;
5043         }
5044
5045         EXT4_SB(sb)->s_journal = journal;
5046         ext4_clear_journal_err(sb, es);
5047
5048         if (!really_read_only && journal_devnum &&
5049             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5050                 es->s_journal_dev = cpu_to_le32(journal_devnum);
5051
5052                 /* Make sure we flush the recovery flag to disk. */
5053                 ext4_commit_super(sb, 1);
5054         }
5055
5056         return 0;
5057 }
5058
5059 static int ext4_commit_super(struct super_block *sb, int sync)
5060 {
5061         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
5062         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
5063         int error = 0;
5064
5065         if (!sbh || block_device_ejected(sb))
5066                 return error;
5067
5068         /*
5069          * The superblock bh should be mapped, but it might not be if the
5070          * device was hot-removed. Not much we can do but fail the I/O.
5071          */
5072         if (!buffer_mapped(sbh))
5073                 return error;
5074
5075         /*
5076          * If the file system is mounted read-only, don't update the
5077          * superblock write time.  This avoids updating the superblock
5078          * write time when we are mounting the root file system
5079          * read/only but we need to replay the journal; at that point,
5080          * for people who are east of GMT and who make their clock
5081          * tick in localtime for Windows bug-for-bug compatibility,
5082          * the clock is set in the future, and this will cause e2fsck
5083          * to complain and force a full file system check.
5084          */
5085         if (!(sb->s_flags & SB_RDONLY))
5086                 ext4_update_tstamp(es, s_wtime);
5087         if (sb->s_bdev->bd_part)
5088                 es->s_kbytes_written =
5089                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
5090                             ((part_stat_read(sb->s_bdev->bd_part,
5091                                              sectors[STAT_WRITE]) -
5092                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
5093         else
5094                 es->s_kbytes_written =
5095                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
5096         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
5097                 ext4_free_blocks_count_set(es,
5098                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
5099                                 &EXT4_SB(sb)->s_freeclusters_counter)));
5100         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
5101                 es->s_free_inodes_count =
5102                         cpu_to_le32(percpu_counter_sum_positive(
5103                                 &EXT4_SB(sb)->s_freeinodes_counter));
5104         BUFFER_TRACE(sbh, "marking dirty");
5105         ext4_superblock_csum_set(sb);
5106         if (sync)
5107                 lock_buffer(sbh);
5108         if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
5109                 /*
5110                  * Oh, dear.  A previous attempt to write the
5111                  * superblock failed.  This could happen because the
5112                  * USB device was yanked out.  Or it could happen to
5113                  * be a transient write error and maybe the block will
5114                  * be remapped.  Nothing we can do but to retry the
5115                  * write and hope for the best.
5116                  */
5117                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
5118                        "superblock detected");
5119                 clear_buffer_write_io_error(sbh);
5120                 set_buffer_uptodate(sbh);
5121         }
5122         mark_buffer_dirty(sbh);
5123         if (sync) {
5124                 unlock_buffer(sbh);
5125                 error = __sync_dirty_buffer(sbh,
5126                         REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
5127                 if (buffer_write_io_error(sbh)) {
5128                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
5129                                "superblock");
5130                         clear_buffer_write_io_error(sbh);
5131                         set_buffer_uptodate(sbh);
5132                 }
5133         }
5134         return error;
5135 }
5136
5137 /*
5138  * Have we just finished recovery?  If so, and if we are mounting (or
5139  * remounting) the filesystem readonly, then we will end up with a
5140  * consistent fs on disk.  Record that fact.
5141  */
5142 static void ext4_mark_recovery_complete(struct super_block *sb,
5143                                         struct ext4_super_block *es)
5144 {
5145         journal_t *journal = EXT4_SB(sb)->s_journal;
5146
5147         if (!ext4_has_feature_journal(sb)) {
5148                 BUG_ON(journal != NULL);
5149                 return;
5150         }
5151         jbd2_journal_lock_updates(journal);
5152         if (jbd2_journal_flush(journal) < 0)
5153                 goto out;
5154
5155         if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) {
5156                 ext4_clear_feature_journal_needs_recovery(sb);
5157                 ext4_commit_super(sb, 1);
5158         }
5159
5160 out:
5161         jbd2_journal_unlock_updates(journal);
5162 }
5163
5164 /*
5165  * If we are mounting (or read-write remounting) a filesystem whose journal
5166  * has recorded an error from a previous lifetime, move that error to the
5167  * main filesystem now.
5168  */
5169 static void ext4_clear_journal_err(struct super_block *sb,
5170                                    struct ext4_super_block *es)
5171 {
5172         journal_t *journal;
5173         int j_errno;
5174         const char *errstr;
5175
5176         BUG_ON(!ext4_has_feature_journal(sb));
5177
5178         journal = EXT4_SB(sb)->s_journal;
5179
5180         /*
5181          * Now check for any error status which may have been recorded in the
5182          * journal by a prior ext4_error() or ext4_abort()
5183          */
5184
5185         j_errno = jbd2_journal_errno(journal);
5186         if (j_errno) {
5187                 char nbuf[16];
5188
5189                 errstr = ext4_decode_error(sb, j_errno, nbuf);
5190                 ext4_warning(sb, "Filesystem error recorded "
5191                              "from previous mount: %s", errstr);
5192                 ext4_warning(sb, "Marking fs in need of filesystem check.");
5193
5194                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
5195                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
5196                 ext4_commit_super(sb, 1);
5197
5198                 jbd2_journal_clear_err(journal);
5199                 jbd2_journal_update_sb_errno(journal);
5200         }
5201 }
5202
5203 /*
5204  * Force the running and committing transactions to commit,
5205  * and wait on the commit.
5206  */
5207 int ext4_force_commit(struct super_block *sb)
5208 {
5209         journal_t *journal;
5210
5211         if (sb_rdonly(sb))
5212                 return 0;
5213
5214         journal = EXT4_SB(sb)->s_journal;
5215         return ext4_journal_force_commit(journal);
5216 }
5217
5218 static int ext4_sync_fs(struct super_block *sb, int wait)
5219 {
5220         int ret = 0;
5221         tid_t target;
5222         bool needs_barrier = false;
5223         struct ext4_sb_info *sbi = EXT4_SB(sb);
5224
5225         if (unlikely(ext4_forced_shutdown(sbi)))
5226                 return 0;
5227
5228         trace_ext4_sync_fs(sb, wait);
5229         flush_workqueue(sbi->rsv_conversion_wq);
5230         /*
5231          * Writeback quota in non-journalled quota case - journalled quota has
5232          * no dirty dquots
5233          */
5234         dquot_writeback_dquots(sb, -1);
5235         /*
5236          * Data writeback is possible w/o journal transaction, so barrier must
5237          * being sent at the end of the function. But we can skip it if
5238          * transaction_commit will do it for us.
5239          */
5240         if (sbi->s_journal) {
5241                 target = jbd2_get_latest_transaction(sbi->s_journal);
5242                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
5243                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
5244                         needs_barrier = true;
5245
5246                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
5247                         if (wait)
5248                                 ret = jbd2_log_wait_commit(sbi->s_journal,
5249                                                            target);
5250                 }
5251         } else if (wait && test_opt(sb, BARRIER))
5252                 needs_barrier = true;
5253         if (needs_barrier) {
5254                 int err;
5255                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
5256                 if (!ret)
5257                         ret = err;
5258         }
5259
5260         return ret;
5261 }
5262
5263 /*
5264  * LVM calls this function before a (read-only) snapshot is created.  This
5265  * gives us a chance to flush the journal completely and mark the fs clean.
5266  *
5267  * Note that only this function cannot bring a filesystem to be in a clean
5268  * state independently. It relies on upper layer to stop all data & metadata
5269  * modifications.
5270  */
5271 static int ext4_freeze(struct super_block *sb)
5272 {
5273         int error = 0;
5274         journal_t *journal;
5275
5276         if (sb_rdonly(sb))
5277                 return 0;
5278
5279         journal = EXT4_SB(sb)->s_journal;
5280
5281         if (journal) {
5282                 /* Now we set up the journal barrier. */
5283                 jbd2_journal_lock_updates(journal);
5284
5285                 /*
5286                  * Don't clear the needs_recovery flag if we failed to
5287                  * flush the journal.
5288                  */
5289                 error = jbd2_journal_flush(journal);
5290                 if (error < 0)
5291                         goto out;
5292
5293                 /* Journal blocked and flushed, clear needs_recovery flag. */
5294                 ext4_clear_feature_journal_needs_recovery(sb);
5295         }
5296
5297         error = ext4_commit_super(sb, 1);
5298 out:
5299         if (journal)
5300                 /* we rely on upper layer to stop further updates */
5301                 jbd2_journal_unlock_updates(journal);
5302         return error;
5303 }
5304
5305 /*
5306  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
5307  * flag here, even though the filesystem is not technically dirty yet.
5308  */
5309 static int ext4_unfreeze(struct super_block *sb)
5310 {
5311         if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
5312                 return 0;
5313
5314         if (EXT4_SB(sb)->s_journal) {
5315                 /* Reset the needs_recovery flag before the fs is unlocked. */
5316                 ext4_set_feature_journal_needs_recovery(sb);
5317         }
5318
5319         ext4_commit_super(sb, 1);
5320         return 0;
5321 }
5322
5323 /*
5324  * Structure to save mount options for ext4_remount's benefit
5325  */
5326 struct ext4_mount_options {
5327         unsigned long s_mount_opt;
5328         unsigned long s_mount_opt2;
5329         kuid_t s_resuid;
5330         kgid_t s_resgid;
5331         unsigned long s_commit_interval;
5332         u32 s_min_batch_time, s_max_batch_time;
5333 #ifdef CONFIG_QUOTA
5334         int s_jquota_fmt;
5335         char *s_qf_names[EXT4_MAXQUOTAS];
5336 #endif
5337 };
5338
5339 static int ext4_remount(struct super_block *sb, int *flags, char *data)
5340 {
5341         struct ext4_super_block *es;
5342         struct ext4_sb_info *sbi = EXT4_SB(sb);
5343         unsigned long old_sb_flags;
5344         struct ext4_mount_options old_opts;
5345         int enable_quota = 0;
5346         ext4_group_t g;
5347         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
5348         int err = 0;
5349 #ifdef CONFIG_QUOTA
5350         int i, j;
5351         char *to_free[EXT4_MAXQUOTAS];
5352 #endif
5353         char *orig_data = kstrdup(data, GFP_KERNEL);
5354
5355         if (data && !orig_data)
5356                 return -ENOMEM;
5357
5358         /* Store the original options */
5359         old_sb_flags = sb->s_flags;
5360         old_opts.s_mount_opt = sbi->s_mount_opt;
5361         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
5362         old_opts.s_resuid = sbi->s_resuid;
5363         old_opts.s_resgid = sbi->s_resgid;
5364         old_opts.s_commit_interval = sbi->s_commit_interval;
5365         old_opts.s_min_batch_time = sbi->s_min_batch_time;
5366         old_opts.s_max_batch_time = sbi->s_max_batch_time;
5367 #ifdef CONFIG_QUOTA
5368         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
5369         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5370                 if (sbi->s_qf_names[i]) {
5371                         char *qf_name = get_qf_name(sb, sbi, i);
5372
5373                         old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL);
5374                         if (!old_opts.s_qf_names[i]) {
5375                                 for (j = 0; j < i; j++)
5376                                         kfree(old_opts.s_qf_names[j]);
5377                                 kfree(orig_data);
5378                                 return -ENOMEM;
5379                         }
5380                 } else
5381                         old_opts.s_qf_names[i] = NULL;
5382 #endif
5383         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
5384                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
5385
5386         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
5387                 err = -EINVAL;
5388                 goto restore_opts;
5389         }
5390
5391         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
5392             test_opt(sb, JOURNAL_CHECKSUM)) {
5393                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
5394                          "during remount not supported; ignoring");
5395                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
5396         }
5397
5398         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
5399                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
5400                         ext4_msg(sb, KERN_ERR, "can't mount with "
5401                                  "both data=journal and delalloc");
5402                         err = -EINVAL;
5403                         goto restore_opts;
5404                 }
5405                 if (test_opt(sb, DIOREAD_NOLOCK)) {
5406                         ext4_msg(sb, KERN_ERR, "can't mount with "
5407                                  "both data=journal and dioread_nolock");
5408                         err = -EINVAL;
5409                         goto restore_opts;
5410                 }
5411                 if (test_opt(sb, DAX)) {
5412                         ext4_msg(sb, KERN_ERR, "can't mount with "
5413                                  "both data=journal and dax");
5414                         err = -EINVAL;
5415                         goto restore_opts;
5416                 }
5417         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
5418                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5419                         ext4_msg(sb, KERN_ERR, "can't mount with "
5420                                 "journal_async_commit in data=ordered mode");
5421                         err = -EINVAL;
5422                         goto restore_opts;
5423                 }
5424         }
5425
5426         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
5427                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
5428                 err = -EINVAL;
5429                 goto restore_opts;
5430         }
5431
5432         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
5433                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
5434                         "dax flag with busy inodes while remounting");
5435                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
5436         }
5437
5438         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
5439                 ext4_abort(sb, EXT4_ERR_ESHUTDOWN, "Abort forced by user");
5440
5441         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
5442                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
5443
5444         es = sbi->s_es;
5445
5446         if (sbi->s_journal) {
5447                 ext4_init_journal_params(sb, sbi->s_journal);
5448                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
5449         }
5450
5451         if (*flags & SB_LAZYTIME)
5452                 sb->s_flags |= SB_LAZYTIME;
5453
5454         if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) {
5455                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
5456                         err = -EROFS;
5457                         goto restore_opts;
5458                 }
5459
5460                 if (*flags & SB_RDONLY) {
5461                         err = sync_filesystem(sb);
5462                         if (err < 0)
5463                                 goto restore_opts;
5464                         err = dquot_suspend(sb, -1);
5465                         if (err < 0)
5466                                 goto restore_opts;
5467
5468                         /*
5469                          * First of all, the unconditional stuff we have to do
5470                          * to disable replay of the journal when we next remount
5471                          */
5472                         sb->s_flags |= SB_RDONLY;
5473
5474                         /*
5475                          * OK, test if we are remounting a valid rw partition
5476                          * readonly, and if so set the rdonly flag and then
5477                          * mark the partition as valid again.
5478                          */
5479                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5480                             (sbi->s_mount_state & EXT4_VALID_FS))
5481                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5482
5483                         if (sbi->s_journal)
5484                                 ext4_mark_recovery_complete(sb, es);
5485                         if (sbi->s_mmp_tsk)
5486                                 kthread_stop(sbi->s_mmp_tsk);
5487                 } else {
5488                         /* Make sure we can mount this feature set readwrite */
5489                         if (ext4_has_feature_readonly(sb) ||
5490                             !ext4_feature_set_ok(sb, 0)) {
5491                                 err = -EROFS;
5492                                 goto restore_opts;
5493                         }
5494                         /*
5495                          * Make sure the group descriptor checksums
5496                          * are sane.  If they aren't, refuse to remount r/w.
5497                          */
5498                         for (g = 0; g < sbi->s_groups_count; g++) {
5499                                 struct ext4_group_desc *gdp =
5500                                         ext4_get_group_desc(sb, g, NULL);
5501
5502                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5503                                         ext4_msg(sb, KERN_ERR,
5504                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5505                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
5506                                                le16_to_cpu(gdp->bg_checksum));
5507                                         err = -EFSBADCRC;
5508                                         goto restore_opts;
5509                                 }
5510                         }
5511
5512                         /*
5513                          * If we have an unprocessed orphan list hanging
5514                          * around from a previously readonly bdev mount,
5515                          * require a full umount/remount for now.
5516                          */
5517                         if (es->s_last_orphan) {
5518                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5519                                        "remount RDWR because of unprocessed "
5520                                        "orphan inode list.  Please "
5521                                        "umount/remount instead");
5522                                 err = -EINVAL;
5523                                 goto restore_opts;
5524                         }
5525
5526                         /*
5527                          * Mounting a RDONLY partition read-write, so reread
5528                          * and store the current valid flag.  (It may have
5529                          * been changed by e2fsck since we originally mounted
5530                          * the partition.)
5531                          */
5532                         if (sbi->s_journal)
5533                                 ext4_clear_journal_err(sb, es);
5534                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5535
5536                         err = ext4_setup_super(sb, es, 0);
5537                         if (err)
5538                                 goto restore_opts;
5539
5540                         sb->s_flags &= ~SB_RDONLY;
5541                         if (ext4_has_feature_mmp(sb))
5542                                 if (ext4_multi_mount_protect(sb,
5543                                                 le64_to_cpu(es->s_mmp_block))) {
5544                                         err = -EROFS;
5545                                         goto restore_opts;
5546                                 }
5547                         enable_quota = 1;
5548                 }
5549         }
5550
5551         /*
5552          * Reinitialize lazy itable initialization thread based on
5553          * current settings
5554          */
5555         if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
5556                 ext4_unregister_li_request(sb);
5557         else {
5558                 ext4_group_t first_not_zeroed;
5559                 first_not_zeroed = ext4_has_uninit_itable(sb);
5560                 ext4_register_li_request(sb, first_not_zeroed);
5561         }
5562
5563         ext4_setup_system_zone(sb);
5564         if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) {
5565                 err = ext4_commit_super(sb, 1);
5566                 if (err)
5567                         goto restore_opts;
5568         }
5569
5570 #ifdef CONFIG_QUOTA
5571         /* Release old quota file names */
5572         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5573                 kfree(old_opts.s_qf_names[i]);
5574         if (enable_quota) {
5575                 if (sb_any_quota_suspended(sb))
5576                         dquot_resume(sb, -1);
5577                 else if (ext4_has_feature_quota(sb)) {
5578                         err = ext4_enable_quotas(sb);
5579                         if (err)
5580                                 goto restore_opts;
5581                 }
5582         }
5583 #endif
5584
5585         *flags = (*flags & ~SB_LAZYTIME) | (sb->s_flags & SB_LAZYTIME);
5586         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5587         kfree(orig_data);
5588         return 0;
5589
5590 restore_opts:
5591         sb->s_flags = old_sb_flags;
5592         sbi->s_mount_opt = old_opts.s_mount_opt;
5593         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5594         sbi->s_resuid = old_opts.s_resuid;
5595         sbi->s_resgid = old_opts.s_resgid;
5596         sbi->s_commit_interval = old_opts.s_commit_interval;
5597         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5598         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5599 #ifdef CONFIG_QUOTA
5600         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5601         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5602                 to_free[i] = get_qf_name(sb, sbi, i);
5603                 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]);
5604         }
5605         synchronize_rcu();
5606         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5607                 kfree(to_free[i]);
5608 #endif
5609         kfree(orig_data);
5610         return err;
5611 }
5612
5613 #ifdef CONFIG_QUOTA
5614 static int ext4_statfs_project(struct super_block *sb,
5615                                kprojid_t projid, struct kstatfs *buf)
5616 {
5617         struct kqid qid;
5618         struct dquot *dquot;
5619         u64 limit;
5620         u64 curblock;
5621
5622         qid = make_kqid_projid(projid);
5623         dquot = dqget(sb, qid);
5624         if (IS_ERR(dquot))
5625                 return PTR_ERR(dquot);
5626         spin_lock(&dquot->dq_dqb_lock);
5627
5628         limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit,
5629                              dquot->dq_dqb.dqb_bhardlimit);
5630         limit >>= sb->s_blocksize_bits;
5631
5632         if (limit && buf->f_blocks > limit) {
5633                 curblock = (dquot->dq_dqb.dqb_curspace +
5634                             dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits;
5635                 buf->f_blocks = limit;
5636                 buf->f_bfree = buf->f_bavail =
5637                         (buf->f_blocks > curblock) ?
5638                          (buf->f_blocks - curblock) : 0;
5639         }
5640
5641         limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit,
5642                              dquot->dq_dqb.dqb_ihardlimit);
5643         if (limit && buf->f_files > limit) {
5644                 buf->f_files = limit;
5645                 buf->f_ffree =
5646                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
5647                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
5648         }
5649
5650         spin_unlock(&dquot->dq_dqb_lock);
5651         dqput(dquot);
5652         return 0;
5653 }
5654 #endif
5655
5656 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5657 {
5658         struct super_block *sb = dentry->d_sb;
5659         struct ext4_sb_info *sbi = EXT4_SB(sb);
5660         struct ext4_super_block *es = sbi->s_es;
5661         ext4_fsblk_t overhead = 0, resv_blocks;
5662         u64 fsid;
5663         s64 bfree;
5664         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5665
5666         if (!test_opt(sb, MINIX_DF))
5667                 overhead = sbi->s_overhead;
5668
5669         buf->f_type = EXT4_SUPER_MAGIC;
5670         buf->f_bsize = sb->s_blocksize;
5671         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5672         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5673                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5674         /* prevent underflow in case that few free space is available */
5675         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5676         buf->f_bavail = buf->f_bfree -
5677                         (ext4_r_blocks_count(es) + resv_blocks);
5678         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5679                 buf->f_bavail = 0;
5680         buf->f_files = le32_to_cpu(es->s_inodes_count);
5681         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5682         buf->f_namelen = EXT4_NAME_LEN;
5683         fsid = le64_to_cpup((void *)es->s_uuid) ^
5684                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5685         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5686         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5687
5688 #ifdef CONFIG_QUOTA
5689         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
5690             sb_has_quota_limits_enabled(sb, PRJQUOTA))
5691                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
5692 #endif
5693         return 0;
5694 }
5695
5696
5697 #ifdef CONFIG_QUOTA
5698
5699 /*
5700  * Helper functions so that transaction is started before we acquire dqio_sem
5701  * to keep correct lock ordering of transaction > dqio_sem
5702  */
5703 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5704 {
5705         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5706 }
5707
5708 static int ext4_write_dquot(struct dquot *dquot)
5709 {
5710         int ret, err;
5711         handle_t *handle;
5712         struct inode *inode;
5713
5714         inode = dquot_to_inode(dquot);
5715         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5716                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5717         if (IS_ERR(handle))
5718                 return PTR_ERR(handle);
5719         ret = dquot_commit(dquot);
5720         err = ext4_journal_stop(handle);
5721         if (!ret)
5722                 ret = err;
5723         return ret;
5724 }
5725
5726 static int ext4_acquire_dquot(struct dquot *dquot)
5727 {
5728         int ret, err;
5729         handle_t *handle;
5730
5731         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5732                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5733         if (IS_ERR(handle))
5734                 return PTR_ERR(handle);
5735         ret = dquot_acquire(dquot);
5736         err = ext4_journal_stop(handle);
5737         if (!ret)
5738                 ret = err;
5739         return ret;
5740 }
5741
5742 static int ext4_release_dquot(struct dquot *dquot)
5743 {
5744         int ret, err;
5745         handle_t *handle;
5746
5747         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5748                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5749         if (IS_ERR(handle)) {
5750                 /* Release dquot anyway to avoid endless cycle in dqput() */
5751                 dquot_release(dquot);
5752                 return PTR_ERR(handle);
5753         }
5754         ret = dquot_release(dquot);
5755         err = ext4_journal_stop(handle);
5756         if (!ret)
5757                 ret = err;
5758         return ret;
5759 }
5760
5761 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5762 {
5763         struct super_block *sb = dquot->dq_sb;
5764         struct ext4_sb_info *sbi = EXT4_SB(sb);
5765
5766         /* Are we journaling quotas? */
5767         if (ext4_has_feature_quota(sb) ||
5768             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5769                 dquot_mark_dquot_dirty(dquot);
5770                 return ext4_write_dquot(dquot);
5771         } else {
5772                 return dquot_mark_dquot_dirty(dquot);
5773         }
5774 }
5775
5776 static int ext4_write_info(struct super_block *sb, int type)
5777 {
5778         int ret, err;
5779         handle_t *handle;
5780
5781         /* Data block + inode block */
5782         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5783         if (IS_ERR(handle))
5784                 return PTR_ERR(handle);
5785         ret = dquot_commit_info(sb, type);
5786         err = ext4_journal_stop(handle);
5787         if (!ret)
5788                 ret = err;
5789         return ret;
5790 }
5791
5792 /*
5793  * Turn on quotas during mount time - we need to find
5794  * the quota file and such...
5795  */
5796 static int ext4_quota_on_mount(struct super_block *sb, int type)
5797 {
5798         return dquot_quota_on_mount(sb, get_qf_name(sb, EXT4_SB(sb), type),
5799                                         EXT4_SB(sb)->s_jquota_fmt, type);
5800 }
5801
5802 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5803 {
5804         struct ext4_inode_info *ei = EXT4_I(inode);
5805
5806         /* The first argument of lockdep_set_subclass has to be
5807          * *exactly* the same as the argument to init_rwsem() --- in
5808          * this case, in init_once() --- or lockdep gets unhappy
5809          * because the name of the lock is set using the
5810          * stringification of the argument to init_rwsem().
5811          */
5812         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5813         lockdep_set_subclass(&ei->i_data_sem, subclass);
5814 }
5815
5816 /*
5817  * Standard function to be called on quota_on
5818  */
5819 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5820                          const struct path *path)
5821 {
5822         int err;
5823
5824         if (!test_opt(sb, QUOTA))
5825                 return -EINVAL;
5826
5827         /* Quotafile not on the same filesystem? */
5828         if (path->dentry->d_sb != sb)
5829                 return -EXDEV;
5830         /* Journaling quota? */
5831         if (EXT4_SB(sb)->s_qf_names[type]) {
5832                 /* Quotafile not in fs root? */
5833                 if (path->dentry->d_parent != sb->s_root)
5834                         ext4_msg(sb, KERN_WARNING,
5835                                 "Quota file not on filesystem root. "
5836                                 "Journaled quota will not work");
5837                 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
5838         } else {
5839                 /*
5840                  * Clear the flag just in case mount options changed since
5841                  * last time.
5842                  */
5843                 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
5844         }
5845
5846         /*
5847          * When we journal data on quota file, we have to flush journal to see
5848          * all updates to the file when we bypass pagecache...
5849          */
5850         if (EXT4_SB(sb)->s_journal &&
5851             ext4_should_journal_data(d_inode(path->dentry))) {
5852                 /*
5853                  * We don't need to lock updates but journal_flush() could
5854                  * otherwise be livelocked...
5855                  */
5856                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5857                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5858                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5859                 if (err)
5860                         return err;
5861         }
5862
5863         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5864         err = dquot_quota_on(sb, type, format_id, path);
5865         if (err) {
5866                 lockdep_set_quota_inode(path->dentry->d_inode,
5867                                              I_DATA_SEM_NORMAL);
5868         } else {
5869                 struct inode *inode = d_inode(path->dentry);
5870                 handle_t *handle;
5871
5872                 /*
5873                  * Set inode flags to prevent userspace from messing with quota
5874                  * files. If this fails, we return success anyway since quotas
5875                  * are already enabled and this is not a hard failure.
5876                  */
5877                 inode_lock(inode);
5878                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5879                 if (IS_ERR(handle))
5880                         goto unlock_inode;
5881                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
5882                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
5883                                 S_NOATIME | S_IMMUTABLE);
5884                 ext4_mark_inode_dirty(handle, inode);
5885                 ext4_journal_stop(handle);
5886         unlock_inode:
5887                 inode_unlock(inode);
5888         }
5889         return err;
5890 }
5891
5892 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5893                              unsigned int flags)
5894 {
5895         int err;
5896         struct inode *qf_inode;
5897         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5898                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5899                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5900                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5901         };
5902
5903         BUG_ON(!ext4_has_feature_quota(sb));
5904
5905         if (!qf_inums[type])
5906                 return -EPERM;
5907
5908         qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL);
5909         if (IS_ERR(qf_inode)) {
5910                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5911                 return PTR_ERR(qf_inode);
5912         }
5913
5914         /* Don't account quota for quota files to avoid recursion */
5915         qf_inode->i_flags |= S_NOQUOTA;
5916         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
5917         err = dquot_load_quota_inode(qf_inode, type, format_id, flags);
5918         if (err)
5919                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
5920         iput(qf_inode);
5921
5922         return err;
5923 }
5924
5925 /* Enable usage tracking for all quota types. */
5926 static int ext4_enable_quotas(struct super_block *sb)
5927 {
5928         int type, err = 0;
5929         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5930                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5931                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5932                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5933         };
5934         bool quota_mopt[EXT4_MAXQUOTAS] = {
5935                 test_opt(sb, USRQUOTA),
5936                 test_opt(sb, GRPQUOTA),
5937                 test_opt(sb, PRJQUOTA),
5938         };
5939
5940         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
5941         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5942                 if (qf_inums[type]) {
5943                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5944                                 DQUOT_USAGE_ENABLED |
5945                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
5946                         if (err) {
5947                                 ext4_warning(sb,
5948                                         "Failed to enable quota tracking "
5949                                         "(type=%d, err=%d). Please run "
5950                                         "e2fsck to fix.", type, err);
5951                                 for (type--; type >= 0; type--)
5952                                         dquot_quota_off(sb, type);
5953
5954                                 return err;
5955                         }
5956                 }
5957         }
5958         return 0;
5959 }
5960
5961 static int ext4_quota_off(struct super_block *sb, int type)
5962 {
5963         struct inode *inode = sb_dqopt(sb)->files[type];
5964         handle_t *handle;
5965         int err;
5966
5967         /* Force all delayed allocation blocks to be allocated.
5968          * Caller already holds s_umount sem */
5969         if (test_opt(sb, DELALLOC))
5970                 sync_filesystem(sb);
5971
5972         if (!inode || !igrab(inode))
5973                 goto out;
5974
5975         err = dquot_quota_off(sb, type);
5976         if (err || ext4_has_feature_quota(sb))
5977                 goto out_put;
5978
5979         inode_lock(inode);
5980         /*
5981          * Update modification times of quota files when userspace can
5982          * start looking at them. If we fail, we return success anyway since
5983          * this is not a hard failure and quotas are already disabled.
5984          */
5985         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5986         if (IS_ERR(handle))
5987                 goto out_unlock;
5988         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
5989         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
5990         inode->i_mtime = inode->i_ctime = current_time(inode);
5991         ext4_mark_inode_dirty(handle, inode);
5992         ext4_journal_stop(handle);
5993 out_unlock:
5994         inode_unlock(inode);
5995 out_put:
5996         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
5997         iput(inode);
5998         return err;
5999 out:
6000         return dquot_quota_off(sb, type);
6001 }
6002
6003 /* Read data from quotafile - avoid pagecache and such because we cannot afford
6004  * acquiring the locks... As quota files are never truncated and quota code
6005  * itself serializes the operations (and no one else should touch the files)
6006  * we don't have to be afraid of races */
6007 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
6008                                size_t len, loff_t off)
6009 {
6010         struct inode *inode = sb_dqopt(sb)->files[type];
6011         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
6012         int offset = off & (sb->s_blocksize - 1);
6013         int tocopy;
6014         size_t toread;
6015         struct buffer_head *bh;
6016         loff_t i_size = i_size_read(inode);
6017
6018         if (off > i_size)
6019                 return 0;
6020         if (off+len > i_size)
6021                 len = i_size-off;
6022         toread = len;
6023         while (toread > 0) {
6024                 tocopy = sb->s_blocksize - offset < toread ?
6025                                 sb->s_blocksize - offset : toread;
6026                 bh = ext4_bread(NULL, inode, blk, 0);
6027                 if (IS_ERR(bh))
6028                         return PTR_ERR(bh);
6029                 if (!bh)        /* A hole? */
6030                         memset(data, 0, tocopy);
6031                 else
6032                         memcpy(data, bh->b_data+offset, tocopy);
6033                 brelse(bh);
6034                 offset = 0;
6035                 toread -= tocopy;
6036                 data += tocopy;
6037                 blk++;
6038         }
6039         return len;
6040 }
6041
6042 /* Write to quotafile (we know the transaction is already started and has
6043  * enough credits) */
6044 static ssize_t ext4_quota_write(struct super_block *sb, int type,
6045                                 const char *data, size_t len, loff_t off)
6046 {
6047         struct inode *inode = sb_dqopt(sb)->files[type];
6048         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
6049         int err, offset = off & (sb->s_blocksize - 1);
6050         int retries = 0;
6051         struct buffer_head *bh;
6052         handle_t *handle = journal_current_handle();
6053
6054         if (EXT4_SB(sb)->s_journal && !handle) {
6055                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
6056                         " cancelled because transaction is not started",
6057                         (unsigned long long)off, (unsigned long long)len);
6058                 return -EIO;
6059         }
6060         /*
6061          * Since we account only one data block in transaction credits,
6062          * then it is impossible to cross a block boundary.
6063          */
6064         if (sb->s_blocksize - offset < len) {
6065                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
6066                         " cancelled because not block aligned",
6067                         (unsigned long long)off, (unsigned long long)len);
6068                 return -EIO;
6069         }
6070
6071         do {
6072                 bh = ext4_bread(handle, inode, blk,
6073                                 EXT4_GET_BLOCKS_CREATE |
6074                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
6075         } while (PTR_ERR(bh) == -ENOSPC &&
6076                  ext4_should_retry_alloc(inode->i_sb, &retries));
6077         if (IS_ERR(bh))
6078                 return PTR_ERR(bh);
6079         if (!bh)
6080                 goto out;
6081         BUFFER_TRACE(bh, "get write access");
6082         err = ext4_journal_get_write_access(handle, bh);
6083         if (err) {
6084                 brelse(bh);
6085                 return err;
6086         }
6087         lock_buffer(bh);
6088         memcpy(bh->b_data+offset, data, len);
6089         flush_dcache_page(bh->b_page);
6090         unlock_buffer(bh);
6091         err = ext4_handle_dirty_metadata(handle, NULL, bh);
6092         brelse(bh);
6093 out:
6094         if (inode->i_size < off + len) {
6095                 i_size_write(inode, off + len);
6096                 EXT4_I(inode)->i_disksize = inode->i_size;
6097                 ext4_mark_inode_dirty(handle, inode);
6098         }
6099         return len;
6100 }
6101 #endif
6102
6103 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
6104                        const char *dev_name, void *data)
6105 {
6106         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
6107 }
6108
6109 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
6110 static inline void register_as_ext2(void)
6111 {
6112         int err = register_filesystem(&ext2_fs_type);
6113         if (err)
6114                 printk(KERN_WARNING
6115                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
6116 }
6117
6118 static inline void unregister_as_ext2(void)
6119 {
6120         unregister_filesystem(&ext2_fs_type);
6121 }
6122
6123 static inline int ext2_feature_set_ok(struct super_block *sb)
6124 {
6125         if (ext4_has_unknown_ext2_incompat_features(sb))
6126                 return 0;
6127         if (sb_rdonly(sb))
6128                 return 1;
6129         if (ext4_has_unknown_ext2_ro_compat_features(sb))
6130                 return 0;
6131         return 1;
6132 }
6133 #else
6134 static inline void register_as_ext2(void) { }
6135 static inline void unregister_as_ext2(void) { }
6136 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
6137 #endif
6138
6139 static inline void register_as_ext3(void)
6140 {
6141         int err = register_filesystem(&ext3_fs_type);
6142         if (err)
6143                 printk(KERN_WARNING
6144                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
6145 }
6146
6147 static inline void unregister_as_ext3(void)
6148 {
6149         unregister_filesystem(&ext3_fs_type);
6150 }
6151
6152 static inline int ext3_feature_set_ok(struct super_block *sb)
6153 {
6154         if (ext4_has_unknown_ext3_incompat_features(sb))
6155                 return 0;
6156         if (!ext4_has_feature_journal(sb))
6157                 return 0;
6158         if (sb_rdonly(sb))
6159                 return 1;
6160         if (ext4_has_unknown_ext3_ro_compat_features(sb))
6161                 return 0;
6162         return 1;
6163 }
6164
6165 static struct file_system_type ext4_fs_type = {
6166         .owner          = THIS_MODULE,
6167         .name           = "ext4",
6168         .mount          = ext4_mount,
6169         .kill_sb        = kill_block_super,
6170         .fs_flags       = FS_REQUIRES_DEV,
6171 };
6172 MODULE_ALIAS_FS("ext4");
6173
6174 /* Shared across all ext4 file systems */
6175 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
6176
6177 static int __init ext4_init_fs(void)
6178 {
6179         int i, err;
6180
6181         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
6182         ext4_li_info = NULL;
6183         mutex_init(&ext4_li_mtx);
6184
6185         /* Build-time check for flags consistency */
6186         ext4_check_flag_values();
6187
6188         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
6189                 init_waitqueue_head(&ext4__ioend_wq[i]);
6190
6191         err = ext4_init_es();
6192         if (err)
6193                 return err;
6194
6195         err = ext4_init_pending();
6196         if (err)
6197                 goto out7;
6198
6199         err = ext4_init_post_read_processing();
6200         if (err)
6201                 goto out6;
6202
6203         err = ext4_init_pageio();
6204         if (err)
6205                 goto out5;
6206
6207         err = ext4_init_system_zone();
6208         if (err)
6209                 goto out4;
6210
6211         err = ext4_init_sysfs();
6212         if (err)
6213                 goto out3;
6214
6215         err = ext4_init_mballoc();
6216         if (err)
6217                 goto out2;
6218         err = init_inodecache();
6219         if (err)
6220                 goto out1;
6221         register_as_ext3();
6222         register_as_ext2();
6223         err = register_filesystem(&ext4_fs_type);
6224         if (err)
6225                 goto out;
6226
6227         return 0;
6228 out:
6229         unregister_as_ext2();
6230         unregister_as_ext3();
6231         destroy_inodecache();
6232 out1:
6233         ext4_exit_mballoc();
6234 out2:
6235         ext4_exit_sysfs();
6236 out3:
6237         ext4_exit_system_zone();
6238 out4:
6239         ext4_exit_pageio();
6240 out5:
6241         ext4_exit_post_read_processing();
6242 out6:
6243         ext4_exit_pending();
6244 out7:
6245         ext4_exit_es();
6246
6247         return err;
6248 }
6249
6250 static void __exit ext4_exit_fs(void)
6251 {
6252         ext4_destroy_lazyinit_thread();
6253         unregister_as_ext2();
6254         unregister_as_ext3();
6255         unregister_filesystem(&ext4_fs_type);
6256         destroy_inodecache();
6257         ext4_exit_mballoc();
6258         ext4_exit_sysfs();
6259         ext4_exit_system_zone();
6260         ext4_exit_pageio();
6261         ext4_exit_post_read_processing();
6262         ext4_exit_es();
6263         ext4_exit_pending();
6264 }
6265
6266 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
6267 MODULE_DESCRIPTION("Fourth Extended Filesystem");
6268 MODULE_LICENSE("GPL");
6269 MODULE_SOFTDEP("pre: crc32c");
6270 module_init(ext4_init_fs)
6271 module_exit(ext4_exit_fs)