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 (card@masi.ibp.fr)
   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 (davem@caip.rutgers.edu), 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
  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 (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 void *ext4_kvmalloc(size_t size, gfp_t flags)
 208 {
 209         void *ret;
 210
 211         ret = kmalloc(size, flags | __GFP_NOWARN);
 212         if (!ret)
 213                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 214         return ret;
 215 }
 216
 217 void *ext4_kvzalloc(size_t size, gfp_t flags)
 218 {
 219         void *ret;
 220
 221         ret = kzalloc(size, flags | __GFP_NOWARN);
 222         if (!ret)
 223                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 224         return ret;
 225 }
 226
 227 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 228                                struct ext4_group_desc *bg)
 229 {
 230         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 231                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 232                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 233 }
 234
 235 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 236                                struct ext4_group_desc *bg)
 237 {
 238         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 239                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 240                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 241 }
 242
 243 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 244                               struct ext4_group_desc *bg)
 245 {
 246         return le32_to_cpu(bg->bg_inode_table_lo) |
 247                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 248                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 249 }
 250
 251 __u32 ext4_free_group_clusters(struct super_block *sb,
 252                                struct ext4_group_desc *bg)
 253 {
 254         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 255                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 256                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 257 }
 258
 259 __u32 ext4_free_inodes_count(struct super_block *sb,
 260                               struct ext4_group_desc *bg)
 261 {
 262         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 263                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 264                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 265 }
 266
 267 __u32 ext4_used_dirs_count(struct super_block *sb,
 268                               struct ext4_group_desc *bg)
 269 {
 270         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 271                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 272                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 273 }
 274
 275 __u32 ext4_itable_unused_count(struct super_block *sb,
 276                               struct ext4_group_desc *bg)
 277 {
 278         return le16_to_cpu(bg->bg_itable_unused_lo) |
 279                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 280                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 281 }
 282
 283 void ext4_block_bitmap_set(struct super_block *sb,
 284                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 285 {
 286         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 287         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 288                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 289 }
 290
 291 void ext4_inode_bitmap_set(struct super_block *sb,
 292                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 293 {
 294         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 295         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 296                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 297 }
 298
 299 void ext4_inode_table_set(struct super_block *sb,
 300                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 301 {
 302         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 303         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 304                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 305 }
 306
 307 void ext4_free_group_clusters_set(struct super_block *sb,
 308                                   struct ext4_group_desc *bg, __u32 count)
 309 {
 310         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 311         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 312                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 313 }
 314
 315 void ext4_free_inodes_set(struct super_block *sb,
 316                           struct ext4_group_desc *bg, __u32 count)
 317 {
 318         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 319         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 320                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 321 }
 322
 323 void ext4_used_dirs_set(struct super_block *sb,
 324                           struct ext4_group_desc *bg, __u32 count)
 325 {
 326         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 327         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 328                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 329 }
 330
 331 void ext4_itable_unused_set(struct super_block *sb,
 332                           struct ext4_group_desc *bg, __u32 count)
 333 {
 334         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 335         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 336                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 337 }
 338
 339 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi)
 340 {
 341         time64_t now = ktime_get_real_seconds();
 342
 343         now = clamp_val(now, 0, (1ull << 40) - 1);
 344
 345         *lo = cpu_to_le32(lower_32_bits(now));
 346         *hi = upper_32_bits(now);
 347 }
 348
 349 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi)
 350 {
 351         return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo);
 352 }
 353 #define ext4_update_tstamp(es, tstamp) \
 354         __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
 355 #define ext4_get_tstamp(es, tstamp) \
 356         __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
 357
 358 static void __save_error_info(struct super_block *sb, const char *func,
 359                             unsigned int line)
 360 {
 361         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 362
 363         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 364         if (bdev_read_only(sb->s_bdev))
 365                 return;
 366         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 367         ext4_update_tstamp(es, s_last_error_time);
 368         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 369         es->s_last_error_line = cpu_to_le32(line);
 370         if (!es->s_first_error_time) {
 371                 es->s_first_error_time = es->s_last_error_time;
 372                 es->s_first_error_time_hi = es->s_last_error_time_hi;
 373                 strncpy(es->s_first_error_func, func,
 374                         sizeof(es->s_first_error_func));
 375                 es->s_first_error_line = cpu_to_le32(line);
 376                 es->s_first_error_ino = es->s_last_error_ino;
 377                 es->s_first_error_block = es->s_last_error_block;
 378         }
 379         /*
 380          * Start the daily error reporting function if it hasn't been
 381          * started already
 382          */
 383         if (!es->s_error_count)
 384                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 385         le32_add_cpu(&es->s_error_count, 1);
 386 }
 387
 388 static void save_error_info(struct super_block *sb, const char *func,
 389                             unsigned int line)
 390 {
 391         __save_error_info(sb, func, line);
 392         ext4_commit_super(sb, 1);
 393 }
 394
 395 /*
 396  * The del_gendisk() function uninitializes the disk-specific data
 397  * structures, including the bdi structure, without telling anyone
 398  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 399  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 400  * This is a kludge to prevent these oops until we can put in a proper
 401  * hook in del_gendisk() to inform the VFS and file system layers.
 402  */
 403 static int block_device_ejected(struct super_block *sb)
 404 {
 405         struct inode *bd_inode = sb->s_bdev->bd_inode;
 406         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 407
 408         return bdi->dev == NULL;
 409 }
 410
 411 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 412 {
 413         struct super_block              *sb = journal->j_private;
 414         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 415         int                             error = is_journal_aborted(journal);
 416         struct ext4_journal_cb_entry    *jce;
 417
 418         BUG_ON(txn->t_state == T_FINISHED);
 419
 420         ext4_process_freed_data(sb, txn->t_tid);
 421
 422         spin_lock(&sbi->s_md_lock);
 423         while (!list_empty(&txn->t_private_list)) {
 424                 jce = list_entry(txn->t_private_list.next,
 425                                  struct ext4_journal_cb_entry, jce_list);
 426                 list_del_init(&jce->jce_list);
 427                 spin_unlock(&sbi->s_md_lock);
 428                 jce->jce_func(sb, jce, error);
 429                 spin_lock(&sbi->s_md_lock);
 430         }
 431         spin_unlock(&sbi->s_md_lock);
 432 }
 433
 434 static bool system_going_down(void)
 435 {
 436         return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF
 437                 || system_state == SYSTEM_RESTART;
 438 }
 439
 440 /* Deal with the reporting of failure conditions on a filesystem such as
 441  * inconsistencies detected or read IO failures.
 442  *
 443  * On ext2, we can store the error state of the filesystem in the
 444  * superblock.  That is not possible on ext4, because we may have other
 445  * write ordering constraints on the superblock which prevent us from
 446  * writing it out straight away; and given that the journal is about to
 447  * be aborted, we can't rely on the current, or future, transactions to
 448  * write out the superblock safely.
 449  *
 450  * We'll just use the jbd2_journal_abort() error code to record an error in
 451  * the journal instead.  On recovery, the journal will complain about
 452  * that error until we've noted it down and cleared it.
 453  */
 454
 455 static void ext4_handle_error(struct super_block *sb)
 456 {
 457         if (test_opt(sb, WARN_ON_ERROR))
 458                 WARN_ON_ONCE(1);
 459
 460         if (sb_rdonly(sb))
 461                 return;
 462
 463         if (!test_opt(sb, ERRORS_CONT)) {
 464                 journal_t *journal = EXT4_SB(sb)->s_journal;
 465
 466                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 467                 if (journal)
 468                         jbd2_journal_abort(journal, -EIO);
 469         }
 470         /*
 471          * We force ERRORS_RO behavior when system is rebooting. Otherwise we
 472          * could panic during 'reboot -f' as the underlying device got already
 473          * disabled.
 474          */
 475         if (test_opt(sb, ERRORS_RO) || system_going_down()) {
 476                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 477                 /*
 478                  * Make sure updated value of ->s_mount_flags will be visible
 479                  * before ->s_flags update
 480                  */
 481                 smp_wmb();
 482                 sb->s_flags |= SB_RDONLY;
 483         } else if (test_opt(sb, ERRORS_PANIC)) {
 484                 if (EXT4_SB(sb)->s_journal &&
 485                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 486                         return;
 487                 panic("EXT4-fs (device %s): panic forced after error\n",
 488                         sb->s_id);
 489         }
 490 }
 491
 492 #define ext4_error_ratelimit(sb)                                        \
 493                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 494                              "EXT4-fs error")
 495
 496 void __ext4_error(struct super_block *sb, const char *function,
 497                   unsigned int line, const char *fmt, ...)
 498 {
 499         struct va_format vaf;
 500         va_list args;
 501
 502         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 503                 return;
 504
 505         trace_ext4_error(sb, function, line);
 506         if (ext4_error_ratelimit(sb)) {
 507                 va_start(args, fmt);
 508                 vaf.fmt = fmt;
 509                 vaf.va = &args;
 510                 printk(KERN_CRIT
 511                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 512                        sb->s_id, function, line, current->comm, &vaf);
 513                 va_end(args);
 514         }
 515         save_error_info(sb, function, line);
 516         ext4_handle_error(sb);
 517 }
 518
 519 void __ext4_error_inode(struct inode *inode, const char *function,
 520                         unsigned int line, ext4_fsblk_t block,
 521                         const char *fmt, ...)
 522 {
 523         va_list args;
 524         struct va_format vaf;
 525         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 526
 527         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 528                 return;
 529
 530         trace_ext4_error(inode->i_sb, function, line);
 531         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 532         es->s_last_error_block = cpu_to_le64(block);
 533         if (ext4_error_ratelimit(inode->i_sb)) {
 534                 va_start(args, fmt);
 535                 vaf.fmt = fmt;
 536                 vaf.va = &args;
 537                 if (block)
 538                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 539                                "inode #%lu: block %llu: comm %s: %pV\n",
 540                                inode->i_sb->s_id, function, line, inode->i_ino,
 541                                block, current->comm, &vaf);
 542                 else
 543                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 544                                "inode #%lu: comm %s: %pV\n",
 545                                inode->i_sb->s_id, function, line, inode->i_ino,
 546                                current->comm, &vaf);
 547                 va_end(args);
 548         }
 549         save_error_info(inode->i_sb, function, line);
 550         ext4_handle_error(inode->i_sb);
 551 }
 552
 553 void __ext4_error_file(struct file *file, const char *function,
 554                        unsigned int line, ext4_fsblk_t block,
 555                        const char *fmt, ...)
 556 {
 557         va_list args;
 558         struct va_format vaf;
 559         struct ext4_super_block *es;
 560         struct inode *inode = file_inode(file);
 561         char pathname[80], *path;
 562
 563         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 564                 return;
 565
 566         trace_ext4_error(inode->i_sb, function, line);
 567         es = EXT4_SB(inode->i_sb)->s_es;
 568         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 569         if (ext4_error_ratelimit(inode->i_sb)) {
 570                 path = file_path(file, pathname, sizeof(pathname));
 571                 if (IS_ERR(path))
 572                         path = "(unknown)";
 573                 va_start(args, fmt);
 574                 vaf.fmt = fmt;
 575                 vaf.va = &args;
 576                 if (block)
 577                         printk(KERN_CRIT
 578                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 579                                "block %llu: comm %s: path %s: %pV\n",
 580                                inode->i_sb->s_id, function, line, inode->i_ino,
 581                                block, current->comm, path, &vaf);
 582                 else
 583                         printk(KERN_CRIT
 584                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 585                                "comm %s: path %s: %pV\n",
 586                                inode->i_sb->s_id, function, line, inode->i_ino,
 587                                current->comm, path, &vaf);
 588                 va_end(args);
 589         }
 590         save_error_info(inode->i_sb, function, line);
 591         ext4_handle_error(inode->i_sb);
 592 }
 593
 594 const char *ext4_decode_error(struct super_block *sb, int errno,
 595                               char nbuf[16])
 596 {
 597         char *errstr = NULL;
 598
 599         switch (errno) {
 600         case -EFSCORRUPTED:
 601                 errstr = "Corrupt filesystem";
 602                 break;
 603         case -EFSBADCRC:
 604                 errstr = "Filesystem failed CRC";
 605                 break;
 606         case -EIO:
 607                 errstr = "IO failure";
 608                 break;
 609         case -ENOMEM:
 610                 errstr = "Out of memory";
 611                 break;
 612         case -EROFS:
 613                 if (!sb || (EXT4_SB(sb)->s_journal &&
 614                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 615                         errstr = "Journal has aborted";
 616                 else
 617                         errstr = "Readonly filesystem";
 618                 break;
 619         default:
 620                 /* If the caller passed in an extra buffer for unknown
 621                  * errors, textualise them now.  Else we just return
 622                  * NULL. */
 623                 if (nbuf) {
 624                         /* Check for truncated error codes... */
 625                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 626                                 errstr = nbuf;
 627                 }
 628                 break;
 629         }
 630
 631         return errstr;
 632 }
 633
 634 /* __ext4_std_error decodes expected errors from journaling functions
 635  * automatically and invokes the appropriate error response.  */
 636
 637 void __ext4_std_error(struct super_block *sb, const char *function,
 638                       unsigned int line, int errno)
 639 {
 640         char nbuf[16];
 641         const char *errstr;
 642
 643         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 644                 return;
 645
 646         /* Special case: if the error is EROFS, and we're not already
 647          * inside a transaction, then there's really no point in logging
 648          * an error. */
 649         if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb))
 650                 return;
 651
 652         if (ext4_error_ratelimit(sb)) {
 653                 errstr = ext4_decode_error(sb, errno, nbuf);
 654                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 655                        sb->s_id, function, line, errstr);
 656         }
 657
 658         save_error_info(sb, function, line);
 659         ext4_handle_error(sb);
 660 }
 661
 662 /*
 663  * ext4_abort is a much stronger failure handler than ext4_error.  The
 664  * abort function may be used to deal with unrecoverable failures such
 665  * as journal IO errors or ENOMEM at a critical moment in log management.
 666  *
 667  * We unconditionally force the filesystem into an ABORT|READONLY state,
 668  * unless the error response on the fs has been set to panic in which
 669  * case we take the easy way out and panic immediately.
 670  */
 671
 672 void __ext4_abort(struct super_block *sb, const char *function,
 673                 unsigned int line, const char *fmt, ...)
 674 {
 675         struct va_format vaf;
 676         va_list args;
 677
 678         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 679                 return;
 680
 681         save_error_info(sb, function, line);
 682         va_start(args, fmt);
 683         vaf.fmt = fmt;
 684         vaf.va = &args;
 685         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: %pV\n",
 686                sb->s_id, function, line, &vaf);
 687         va_end(args);
 688
 689         if (sb_rdonly(sb) == 0) {
 690                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 691                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 692                 /*
 693                  * Make sure updated value of ->s_mount_flags will be visible
 694                  * before ->s_flags update
 695                  */
 696                 smp_wmb();
 697                 sb->s_flags |= SB_RDONLY;
 698                 if (EXT4_SB(sb)->s_journal)
 699                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 700                 save_error_info(sb, function, line);
 701         }
 702         if (test_opt(sb, ERRORS_PANIC) && !system_going_down()) {
 703                 if (EXT4_SB(sb)->s_journal &&
 704                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 705                         return;
 706                 panic("EXT4-fs panic from previous error\n");
 707         }
 708 }
 709
 710 void __ext4_msg(struct super_block *sb,
 711                 const char *prefix, const char *fmt, ...)
 712 {
 713         struct va_format vaf;
 714         va_list args;
 715
 716         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 717                 return;
 718
 719         va_start(args, fmt);
 720         vaf.fmt = fmt;
 721         vaf.va = &args;
 722         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 723         va_end(args);
 724 }
 725
 726 #define ext4_warning_ratelimit(sb)                                      \
 727                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
 728                              "EXT4-fs warning")
 729
 730 void __ext4_warning(struct super_block *sb, const char *function,
 731                     unsigned int line, const char *fmt, ...)
 732 {
 733         struct va_format vaf;
 734         va_list args;
 735
 736         if (!ext4_warning_ratelimit(sb))
 737                 return;
 738
 739         va_start(args, fmt);
 740         vaf.fmt = fmt;
 741         vaf.va = &args;
 742         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 743                sb->s_id, function, line, &vaf);
 744         va_end(args);
 745 }
 746
 747 void __ext4_warning_inode(const struct inode *inode, const char *function,
 748                           unsigned int line, const char *fmt, ...)
 749 {
 750         struct va_format vaf;
 751         va_list args;
 752
 753         if (!ext4_warning_ratelimit(inode->i_sb))
 754                 return;
 755
 756         va_start(args, fmt);
 757         vaf.fmt = fmt;
 758         vaf.va = &args;
 759         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 760                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 761                function, line, inode->i_ino, current->comm, &vaf);
 762         va_end(args);
 763 }
 764
 765 void __ext4_grp_locked_error(const char *function, unsigned int line,
 766                              struct super_block *sb, ext4_group_t grp,
 767                              unsigned long ino, ext4_fsblk_t block,
 768                              const char *fmt, ...)
 769 __releases(bitlock)
 770 __acquires(bitlock)
 771 {
 772         struct va_format vaf;
 773         va_list args;
 774         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 775
 776         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 777                 return;
 778
 779         trace_ext4_error(sb, function, line);
 780         es->s_last_error_ino = cpu_to_le32(ino);
 781         es->s_last_error_block = cpu_to_le64(block);
 782         __save_error_info(sb, function, line);
 783
 784         if (ext4_error_ratelimit(sb)) {
 785                 va_start(args, fmt);
 786                 vaf.fmt = fmt;
 787                 vaf.va = &args;
 788                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 789                        sb->s_id, function, line, grp);
 790                 if (ino)
 791                         printk(KERN_CONT "inode %lu: ", ino);
 792                 if (block)
 793                         printk(KERN_CONT "block %llu:",
 794                                (unsigned long long) block);
 795                 printk(KERN_CONT "%pV\n", &vaf);
 796                 va_end(args);
 797         }
 798
 799         if (test_opt(sb, WARN_ON_ERROR))
 800                 WARN_ON_ONCE(1);
 801
 802         if (test_opt(sb, ERRORS_CONT)) {
 803                 ext4_commit_super(sb, 0);
 804                 return;
 805         }
 806
 807         ext4_unlock_group(sb, grp);
 808         ext4_commit_super(sb, 1);
 809         ext4_handle_error(sb);
 810         /*
 811          * We only get here in the ERRORS_RO case; relocking the group
 812          * may be dangerous, but nothing bad will happen since the
 813          * filesystem will have already been marked read/only and the
 814          * journal has been aborted.  We return 1 as a hint to callers
 815          * who might what to use the return value from
 816          * ext4_grp_locked_error() to distinguish between the
 817          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 818          * aggressively from the ext4 function in question, with a
 819          * more appropriate error code.
 820          */
 821         ext4_lock_group(sb, grp);
 822         return;
 823 }
 824
 825 void ext4_mark_group_bitmap_corrupted(struct super_block *sb,
 826                                      ext4_group_t group,
 827                                      unsigned int flags)
 828 {
 829         struct ext4_sb_info *sbi = EXT4_SB(sb);
 830         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
 831         struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
 832         int ret;
 833
 834         if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) {
 835                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT,
 836                                             &grp->bb_state);
 837                 if (!ret)
 838                         percpu_counter_sub(&sbi->s_freeclusters_counter,
 839                                            grp->bb_free);
 840         }
 841
 842         if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) {
 843                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT,
 844                                             &grp->bb_state);
 845                 if (!ret && gdp) {
 846                         int count;
 847
 848                         count = ext4_free_inodes_count(sb, gdp);
 849                         percpu_counter_sub(&sbi->s_freeinodes_counter,
 850                                            count);
 851                 }
 852         }
 853 }
 854
 855 void ext4_update_dynamic_rev(struct super_block *sb)
 856 {
 857         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 858
 859         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 860                 return;
 861
 862         ext4_warning(sb,
 863                      "updating to rev %d because of new feature flag, "
 864                      "running e2fsck is recommended",
 865                      EXT4_DYNAMIC_REV);
 866
 867         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 868         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 869         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 870         /* leave es->s_feature_*compat flags alone */
 871         /* es->s_uuid will be set by e2fsck if empty */
 872
 873         /*
 874          * The rest of the superblock fields should be zero, and if not it
 875          * means they are likely already in use, so leave them alone.  We
 876          * can leave it up to e2fsck to clean up any inconsistencies there.
 877          */
 878 }
 879
 880 /*
 881  * Open the external journal device
 882  */
 883 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 884 {
 885         struct block_device *bdev;
 886         char b[BDEVNAME_SIZE];
 887
 888         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 889         if (IS_ERR(bdev))
 890                 goto fail;
 891         return bdev;
 892
 893 fail:
 894         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 895                         __bdevname(dev, b), PTR_ERR(bdev));
 896         return NULL;
 897 }
 898
 899 /*
 900  * Release the journal device
 901  */
 902 static void ext4_blkdev_put(struct block_device *bdev)
 903 {
 904         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 905 }
 906
 907 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 908 {
 909         struct block_device *bdev;
 910         bdev = sbi->journal_bdev;
 911         if (bdev) {
 912                 ext4_blkdev_put(bdev);
 913                 sbi->journal_bdev = NULL;
 914         }
 915 }
 916
 917 static inline struct inode *orphan_list_entry(struct list_head *l)
 918 {
 919         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 920 }
 921
 922 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 923 {
 924         struct list_head *l;
 925
 926         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 927                  le32_to_cpu(sbi->s_es->s_last_orphan));
 928
 929         printk(KERN_ERR "sb_info orphan list:\n");
 930         list_for_each(l, &sbi->s_orphan) {
 931                 struct inode *inode = orphan_list_entry(l);
 932                 printk(KERN_ERR "  "
 933                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 934                        inode->i_sb->s_id, inode->i_ino, inode,
 935                        inode->i_mode, inode->i_nlink,
 936                        NEXT_ORPHAN(inode));
 937         }
 938 }
 939
 940 #ifdef CONFIG_QUOTA
 941 static int ext4_quota_off(struct super_block *sb, int type);
 942
 943 static inline void ext4_quota_off_umount(struct super_block *sb)
 944 {
 945         int type;
 946
 947         /* Use our quota_off function to clear inode flags etc. */
 948         for (type = 0; type < EXT4_MAXQUOTAS; type++)
 949                 ext4_quota_off(sb, type);
 950 }
 951
 952 /*
 953  * This is a helper function which is used in the mount/remount
 954  * codepaths (which holds s_umount) to fetch the quota file name.
 955  */
 956 static inline char *get_qf_name(struct super_block *sb,
 957                                 struct ext4_sb_info *sbi,
 958                                 int type)
 959 {
 960         return rcu_dereference_protected(sbi->s_qf_names[type],
 961                                          lockdep_is_held(&sb->s_umount));
 962 }
 963 #else
 964 static inline void ext4_quota_off_umount(struct super_block *sb)
 965 {
 966 }
 967 #endif
 968
 969 static void ext4_put_super(struct super_block *sb)
 970 {
 971         struct ext4_sb_info *sbi = EXT4_SB(sb);
 972         struct ext4_super_block *es = sbi->s_es;
 973         int aborted = 0;
 974         int i, err;
 975
 976         ext4_unregister_li_request(sb);
 977         ext4_quota_off_umount(sb);
 978
 979         destroy_workqueue(sbi->rsv_conversion_wq);
 980
 981         if (sbi->s_journal) {
 982                 aborted = is_journal_aborted(sbi->s_journal);
 983                 err = jbd2_journal_destroy(sbi->s_journal);
 984                 sbi->s_journal = NULL;
 985                 if ((err < 0) && !aborted)
 986                         ext4_abort(sb, "Couldn't clean up the journal");
 987         }
 988
 989         ext4_unregister_sysfs(sb);
 990         ext4_es_unregister_shrinker(sbi);
 991         del_timer_sync(&sbi->s_err_report);
 992         ext4_release_system_zone(sb);
 993         ext4_mb_release(sb);
 994         ext4_ext_release(sb);
 995
 996         if (!sb_rdonly(sb) && !aborted) {
 997                 ext4_clear_feature_journal_needs_recovery(sb);
 998                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 999         }
1000         if (!sb_rdonly(sb))
1001                 ext4_commit_super(sb, 1);
1002
1003         for (i = 0; i < sbi->s_gdb_count; i++)
1004                 brelse(sbi->s_group_desc[i]);
1005         kvfree(sbi->s_group_desc);
1006         kvfree(sbi->s_flex_groups);
1007         percpu_counter_destroy(&sbi->s_freeclusters_counter);
1008         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1009         percpu_counter_destroy(&sbi->s_dirs_counter);
1010         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
1011         percpu_free_rwsem(&sbi->s_journal_flag_rwsem);
1012 #ifdef CONFIG_QUOTA
1013         for (i = 0; i < EXT4_MAXQUOTAS; i++)
1014                 kfree(get_qf_name(sb, sbi, i));
1015 #endif
1016
1017         /* Debugging code just in case the in-memory inode orphan list
1018          * isn't empty.  The on-disk one can be non-empty if we've
1019          * detected an error and taken the fs readonly, but the
1020          * in-memory list had better be clean by this point. */
1021         if (!list_empty(&sbi->s_orphan))
1022                 dump_orphan_list(sb, sbi);
1023         J_ASSERT(list_empty(&sbi->s_orphan));
1024
1025         sync_blockdev(sb->s_bdev);
1026         invalidate_bdev(sb->s_bdev);
1027         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
1028                 /*
1029                  * Invalidate the journal device's buffers.  We don't want them
1030                  * floating about in memory - the physical journal device may
1031                  * hotswapped, and it breaks the `ro-after' testing code.
1032                  */
1033                 sync_blockdev(sbi->journal_bdev);
1034                 invalidate_bdev(sbi->journal_bdev);
1035                 ext4_blkdev_remove(sbi);
1036         }
1037
1038         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
1039         sbi->s_ea_inode_cache = NULL;
1040
1041         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
1042         sbi->s_ea_block_cache = NULL;
1043
1044         if (sbi->s_mmp_tsk)
1045                 kthread_stop(sbi->s_mmp_tsk);
1046         brelse(sbi->s_sbh);
1047         sb->s_fs_info = NULL;
1048         /*
1049          * Now that we are completely done shutting down the
1050          * superblock, we need to actually destroy the kobject.
1051          */
1052         kobject_put(&sbi->s_kobj);
1053         wait_for_completion(&sbi->s_kobj_unregister);
1054         if (sbi->s_chksum_driver)
1055                 crypto_free_shash(sbi->s_chksum_driver);
1056         kfree(sbi->s_blockgroup_lock);
1057         fs_put_dax(sbi->s_daxdev);
1058 #ifdef CONFIG_UNICODE
1059         utf8_unload(sbi->s_encoding);
1060 #endif
1061         kfree(sbi);
1062 }
1063
1064 static struct kmem_cache *ext4_inode_cachep;
1065
1066 /*
1067  * Called inside transaction, so use GFP_NOFS
1068  */
1069 static struct inode *ext4_alloc_inode(struct super_block *sb)
1070 {
1071         struct ext4_inode_info *ei;
1072
1073         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
1074         if (!ei)
1075                 return NULL;
1076
1077         inode_set_iversion(&ei->vfs_inode, 1);
1078         spin_lock_init(&ei->i_raw_lock);
1079         INIT_LIST_HEAD(&ei->i_prealloc_list);
1080         spin_lock_init(&ei->i_prealloc_lock);
1081         ext4_es_init_tree(&ei->i_es_tree);
1082         rwlock_init(&ei->i_es_lock);
1083         INIT_LIST_HEAD(&ei->i_es_list);
1084         ei->i_es_all_nr = 0;
1085         ei->i_es_shk_nr = 0;
1086         ei->i_es_shrink_lblk = 0;
1087         ei->i_reserved_data_blocks = 0;
1088         ei->i_da_metadata_calc_len = 0;
1089         ei->i_da_metadata_calc_last_lblock = 0;
1090         spin_lock_init(&(ei->i_block_reservation_lock));
1091         ext4_init_pending_tree(&ei->i_pending_tree);
1092 #ifdef CONFIG_QUOTA
1093         ei->i_reserved_quota = 0;
1094         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
1095 #endif
1096         ei->jinode = NULL;
1097         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
1098         spin_lock_init(&ei->i_completed_io_lock);
1099         ei->i_sync_tid = 0;
1100         ei->i_datasync_tid = 0;
1101         atomic_set(&ei->i_unwritten, 0);
1102         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
1103         return &ei->vfs_inode;
1104 }
1105
1106 static int ext4_drop_inode(struct inode *inode)
1107 {
1108         int drop = generic_drop_inode(inode);
1109
1110         trace_ext4_drop_inode(inode, drop);
1111         return drop;
1112 }
1113
1114 static void ext4_free_in_core_inode(struct inode *inode)
1115 {
1116         fscrypt_free_inode(inode);
1117         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1118 }
1119
1120 static void ext4_destroy_inode(struct inode *inode)
1121 {
1122         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1123                 ext4_msg(inode->i_sb, KERN_ERR,
1124                          "Inode %lu (%p): orphan list check failed!",
1125                          inode->i_ino, EXT4_I(inode));
1126                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1127                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1128                                 true);
1129                 dump_stack();
1130         }
1131 }
1132
1133 static void init_once(void *foo)
1134 {
1135         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1136
1137         INIT_LIST_HEAD(&ei->i_orphan);
1138         init_rwsem(&ei->xattr_sem);
1139         init_rwsem(&ei->i_data_sem);
1140         init_rwsem(&ei->i_mmap_sem);
1141         inode_init_once(&ei->vfs_inode);
1142 }
1143
1144 static int __init init_inodecache(void)
1145 {
1146         ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache",
1147                                 sizeof(struct ext4_inode_info), 0,
1148                                 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
1149                                         SLAB_ACCOUNT),
1150                                 offsetof(struct ext4_inode_info, i_data),
1151                                 sizeof_field(struct ext4_inode_info, i_data),
1152                                 init_once);
1153         if (ext4_inode_cachep == NULL)
1154                 return -ENOMEM;
1155         return 0;
1156 }
1157
1158 static void destroy_inodecache(void)
1159 {
1160         /*
1161          * Make sure all delayed rcu free inodes are flushed before we
1162          * destroy cache.
1163          */
1164         rcu_barrier();
1165         kmem_cache_destroy(ext4_inode_cachep);
1166 }
1167
1168 void ext4_clear_inode(struct inode *inode)
1169 {
1170         invalidate_inode_buffers(inode);
1171         clear_inode(inode);
1172         dquot_drop(inode);
1173         ext4_discard_preallocations(inode);
1174         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1175         if (EXT4_I(inode)->jinode) {
1176                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1177                                                EXT4_I(inode)->jinode);
1178                 jbd2_free_inode(EXT4_I(inode)->jinode);
1179                 EXT4_I(inode)->jinode = NULL;
1180         }
1181         fscrypt_put_encryption_info(inode);
1182         fsverity_cleanup_inode(inode);
1183 }
1184
1185 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1186                                         u64 ino, u32 generation)
1187 {
1188         struct inode *inode;
1189
1190         /*
1191          * Currently we don't know the generation for parent directory, so
1192          * a generation of 0 means "accept any"
1193          */
1194         inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE);
1195         if (IS_ERR(inode))
1196                 return ERR_CAST(inode);
1197         if (generation && inode->i_generation != generation) {
1198                 iput(inode);
1199                 return ERR_PTR(-ESTALE);
1200         }
1201
1202         return inode;
1203 }
1204
1205 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1206                                         int fh_len, int fh_type)
1207 {
1208         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1209                                     ext4_nfs_get_inode);
1210 }
1211
1212 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1213                                         int fh_len, int fh_type)
1214 {
1215         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1216                                     ext4_nfs_get_inode);
1217 }
1218
1219 static int ext4_nfs_commit_metadata(struct inode *inode)
1220 {
1221         struct writeback_control wbc = {
1222                 .sync_mode = WB_SYNC_ALL
1223         };
1224
1225         trace_ext4_nfs_commit_metadata(inode);
1226         return ext4_write_inode(inode, &wbc);
1227 }
1228
1229 /*
1230  * Try to release metadata pages (indirect blocks, directories) which are
1231  * mapped via the block device.  Since these pages could have journal heads
1232  * which would prevent try_to_free_buffers() from freeing them, we must use
1233  * jbd2 layer's try_to_free_buffers() function to release them.
1234  */
1235 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1236                                  gfp_t wait)
1237 {
1238         journal_t *journal = EXT4_SB(sb)->s_journal;
1239
1240         WARN_ON(PageChecked(page));
1241         if (!page_has_buffers(page))
1242                 return 0;
1243         if (journal)
1244                 return jbd2_journal_try_to_free_buffers(journal, page,
1245                                                 wait & ~__GFP_DIRECT_RECLAIM);
1246         return try_to_free_buffers(page);
1247 }
1248
1249 #ifdef CONFIG_FS_ENCRYPTION
1250 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1251 {
1252         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1253                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1254 }
1255
1256 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1257                                                         void *fs_data)
1258 {
1259         handle_t *handle = fs_data;
1260         int res, res2, credits, retries = 0;
1261
1262         /*
1263          * Encrypting the root directory is not allowed because e2fsck expects
1264          * lost+found to exist and be unencrypted, and encrypting the root
1265          * directory would imply encrypting the lost+found directory as well as
1266          * the filename "lost+found" itself.
1267          */
1268         if (inode->i_ino == EXT4_ROOT_INO)
1269                 return -EPERM;
1270
1271         if (WARN_ON_ONCE(IS_DAX(inode) && i_size_read(inode)))
1272                 return -EINVAL;
1273
1274         res = ext4_convert_inline_data(inode);
1275         if (res)
1276                 return res;
1277
1278         /*
1279          * If a journal handle was specified, then the encryption context is
1280          * being set on a new inode via inheritance and is part of a larger
1281          * transaction to create the inode.  Otherwise the encryption context is
1282          * being set on an existing inode in its own transaction.  Only in the
1283          * latter case should the "retry on ENOSPC" logic be used.
1284          */
1285
1286         if (handle) {
1287                 res = ext4_xattr_set_handle(handle, inode,
1288                                             EXT4_XATTR_INDEX_ENCRYPTION,
1289                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1290                                             ctx, len, 0);
1291                 if (!res) {
1292                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1293                         ext4_clear_inode_state(inode,
1294                                         EXT4_STATE_MAY_INLINE_DATA);
1295                         /*
1296                          * Update inode->i_flags - S_ENCRYPTED will be enabled,
1297                          * S_DAX may be disabled
1298                          */
1299                         ext4_set_inode_flags(inode);
1300                 }
1301                 return res;
1302         }
1303
1304         res = dquot_initialize(inode);
1305         if (res)
1306                 return res;
1307 retry:
1308         res = ext4_xattr_set_credits(inode, len, false /* is_create */,
1309                                      &credits);
1310         if (res)
1311                 return res;
1312
1313         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
1314         if (IS_ERR(handle))
1315                 return PTR_ERR(handle);
1316
1317         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1318                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1319                                     ctx, len, 0);
1320         if (!res) {
1321                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1322                 /*
1323                  * Update inode->i_flags - S_ENCRYPTED will be enabled,
1324                  * S_DAX may be disabled
1325                  */
1326                 ext4_set_inode_flags(inode);
1327                 res = ext4_mark_inode_dirty(handle, inode);
1328                 if (res)
1329                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1330         }
1331         res2 = ext4_journal_stop(handle);
1332
1333         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1334                 goto retry;
1335         if (!res)
1336                 res = res2;
1337         return res;
1338 }
1339
1340 static bool ext4_dummy_context(struct inode *inode)
1341 {
1342         return DUMMY_ENCRYPTION_ENABLED(EXT4_SB(inode->i_sb));
1343 }
1344
1345 static const struct fscrypt_operations ext4_cryptops = {
1346         .key_prefix             = "ext4:",
1347         .get_context            = ext4_get_context,
1348         .set_context            = ext4_set_context,
1349         .dummy_context          = ext4_dummy_context,
1350         .empty_dir              = ext4_empty_dir,
1351         .max_namelen            = EXT4_NAME_LEN,
1352 };
1353 #endif
1354
1355 #ifdef CONFIG_QUOTA
1356 static const char * const quotatypes[] = INITQFNAMES;
1357 #define QTYPE2NAME(t) (quotatypes[t])
1358
1359 static int ext4_write_dquot(struct dquot *dquot);
1360 static int ext4_acquire_dquot(struct dquot *dquot);
1361 static int ext4_release_dquot(struct dquot *dquot);
1362 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1363 static int ext4_write_info(struct super_block *sb, int type);
1364 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1365                          const struct path *path);
1366 static int ext4_quota_on_mount(struct super_block *sb, int type);
1367 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1368                                size_t len, loff_t off);
1369 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1370                                 const char *data, size_t len, loff_t off);
1371 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1372                              unsigned int flags);
1373 static int ext4_enable_quotas(struct super_block *sb);
1374 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid);
1375
1376 static struct dquot **ext4_get_dquots(struct inode *inode)
1377 {
1378         return EXT4_I(inode)->i_dquot;
1379 }
1380
1381 static const struct dquot_operations ext4_quota_operations = {
1382         .get_reserved_space     = ext4_get_reserved_space,
1383         .write_dquot            = ext4_write_dquot,
1384         .acquire_dquot          = ext4_acquire_dquot,
1385         .release_dquot          = ext4_release_dquot,
1386         .mark_dirty             = ext4_mark_dquot_dirty,
1387         .write_info             = ext4_write_info,
1388         .alloc_dquot            = dquot_alloc,
1389         .destroy_dquot          = dquot_destroy,
1390         .get_projid             = ext4_get_projid,
1391         .get_inode_usage        = ext4_get_inode_usage,
1392         .get_next_id            = ext4_get_next_id,
1393 };
1394
1395 static const struct quotactl_ops ext4_qctl_operations = {
1396         .quota_on       = ext4_quota_on,
1397         .quota_off      = ext4_quota_off,
1398         .quota_sync     = dquot_quota_sync,
1399         .get_state      = dquot_get_state,
1400         .set_info       = dquot_set_dqinfo,
1401         .get_dqblk      = dquot_get_dqblk,
1402         .set_dqblk      = dquot_set_dqblk,
1403         .get_nextdqblk  = dquot_get_next_dqblk,
1404 };
1405 #endif
1406
1407 static const struct super_operations ext4_sops = {
1408         .alloc_inode    = ext4_alloc_inode,
1409         .free_inode     = ext4_free_in_core_inode,
1410         .destroy_inode  = ext4_destroy_inode,
1411         .write_inode    = ext4_write_inode,
1412         .dirty_inode    = ext4_dirty_inode,
1413         .drop_inode     = ext4_drop_inode,
1414         .evict_inode    = ext4_evict_inode,
1415         .put_super      = ext4_put_super,
1416         .sync_fs        = ext4_sync_fs,
1417         .freeze_fs      = ext4_freeze,
1418         .unfreeze_fs    = ext4_unfreeze,
1419         .statfs         = ext4_statfs,
1420         .remount_fs     = ext4_remount,
1421         .show_options   = ext4_show_options,
1422 #ifdef CONFIG_QUOTA
1423         .quota_read     = ext4_quota_read,
1424         .quota_write    = ext4_quota_write,
1425         .get_dquots     = ext4_get_dquots,
1426 #endif
1427         .bdev_try_to_free_page = bdev_try_to_free_page,
1428 };
1429
1430 static const struct export_operations ext4_export_ops = {
1431         .fh_to_dentry = ext4_fh_to_dentry,
1432         .fh_to_parent = ext4_fh_to_parent,
1433         .get_parent = ext4_get_parent,
1434         .commit_metadata = ext4_nfs_commit_metadata,
1435 };
1436
1437 enum {
1438         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1439         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1440         Opt_nouid32, Opt_debug, Opt_removed,
1441         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1442         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1443         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1444         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1445         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1446         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1447         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1448         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1449         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1450         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, Opt_dax,
1451         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
1452         Opt_nowarn_on_error, Opt_mblk_io_submit,
1453         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1454         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1455         Opt_inode_readahead_blks, Opt_journal_ioprio,
1456         Opt_dioread_nolock, Opt_dioread_lock,
1457         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1458         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1459 };
1460
1461 static const match_table_t tokens = {
1462         {Opt_bsd_df, "bsddf"},
1463         {Opt_minix_df, "minixdf"},
1464         {Opt_grpid, "grpid"},
1465         {Opt_grpid, "bsdgroups"},
1466         {Opt_nogrpid, "nogrpid"},
1467         {Opt_nogrpid, "sysvgroups"},
1468         {Opt_resgid, "resgid=%u"},
1469         {Opt_resuid, "resuid=%u"},
1470         {Opt_sb, "sb=%u"},
1471         {Opt_err_cont, "errors=continue"},
1472         {Opt_err_panic, "errors=panic"},
1473         {Opt_err_ro, "errors=remount-ro"},
1474         {Opt_nouid32, "nouid32"},
1475         {Opt_debug, "debug"},
1476         {Opt_removed, "oldalloc"},
1477         {Opt_removed, "orlov"},
1478         {Opt_user_xattr, "user_xattr"},
1479         {Opt_nouser_xattr, "nouser_xattr"},
1480         {Opt_acl, "acl"},
1481         {Opt_noacl, "noacl"},
1482         {Opt_noload, "norecovery"},
1483         {Opt_noload, "noload"},
1484         {Opt_removed, "nobh"},
1485         {Opt_removed, "bh"},
1486         {Opt_commit, "commit=%u"},
1487         {Opt_min_batch_time, "min_batch_time=%u"},
1488         {Opt_max_batch_time, "max_batch_time=%u"},
1489         {Opt_journal_dev, "journal_dev=%u"},
1490         {Opt_journal_path, "journal_path=%s"},
1491         {Opt_journal_checksum, "journal_checksum"},
1492         {Opt_nojournal_checksum, "nojournal_checksum"},
1493         {Opt_journal_async_commit, "journal_async_commit"},
1494         {Opt_abort, "abort"},
1495         {Opt_data_journal, "data=journal"},
1496         {Opt_data_ordered, "data=ordered"},
1497         {Opt_data_writeback, "data=writeback"},
1498         {Opt_data_err_abort, "data_err=abort"},
1499         {Opt_data_err_ignore, "data_err=ignore"},
1500         {Opt_offusrjquota, "usrjquota="},
1501         {Opt_usrjquota, "usrjquota=%s"},
1502         {Opt_offgrpjquota, "grpjquota="},
1503         {Opt_grpjquota, "grpjquota=%s"},
1504         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1505         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1506         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1507         {Opt_grpquota, "grpquota"},
1508         {Opt_noquota, "noquota"},
1509         {Opt_quota, "quota"},
1510         {Opt_usrquota, "usrquota"},
1511         {Opt_prjquota, "prjquota"},
1512         {Opt_barrier, "barrier=%u"},
1513         {Opt_barrier, "barrier"},
1514         {Opt_nobarrier, "nobarrier"},
1515         {Opt_i_version, "i_version"},
1516         {Opt_dax, "dax"},
1517         {Opt_stripe, "stripe=%u"},
1518         {Opt_delalloc, "delalloc"},
1519         {Opt_warn_on_error, "warn_on_error"},
1520         {Opt_nowarn_on_error, "nowarn_on_error"},
1521         {Opt_lazytime, "lazytime"},
1522         {Opt_nolazytime, "nolazytime"},
1523         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1524         {Opt_nodelalloc, "nodelalloc"},
1525         {Opt_removed, "mblk_io_submit"},
1526         {Opt_removed, "nomblk_io_submit"},
1527         {Opt_block_validity, "block_validity"},
1528         {Opt_noblock_validity, "noblock_validity"},
1529         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1530         {Opt_journal_ioprio, "journal_ioprio=%u"},
1531         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1532         {Opt_auto_da_alloc, "auto_da_alloc"},
1533         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1534         {Opt_dioread_nolock, "dioread_nolock"},
1535         {Opt_dioread_lock, "dioread_lock"},
1536         {Opt_discard, "discard"},
1537         {Opt_nodiscard, "nodiscard"},
1538         {Opt_init_itable, "init_itable=%u"},
1539         {Opt_init_itable, "init_itable"},
1540         {Opt_noinit_itable, "noinit_itable"},
1541         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1542         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1543         {Opt_nombcache, "nombcache"},
1544         {Opt_nombcache, "no_mbcache"},  /* for backward compatibility */
1545         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1546         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1547         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1548         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1549         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1550         {Opt_err, NULL},
1551 };
1552
1553 static ext4_fsblk_t get_sb_block(void **data)
1554 {
1555         ext4_fsblk_t    sb_block;
1556         char            *options = (char *) *data;
1557
1558         if (!options || strncmp(options, "sb=", 3) != 0)
1559                 return 1;       /* Default location */
1560
1561         options += 3;
1562         /* TODO: use simple_strtoll with >32bit ext4 */
1563         sb_block = simple_strtoul(options, &options, 0);
1564         if (*options && *options != ',') {
1565                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1566                        (char *) *data);
1567                 return 1;
1568         }
1569         if (*options == ',')
1570                 options++;
1571         *data = (void *) options;
1572
1573         return sb_block;
1574 }
1575
1576 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1577 static const char deprecated_msg[] =
1578         "Mount option \"%s\" will be removed by %s\n"
1579         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1580
1581 #ifdef CONFIG_QUOTA
1582 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1583 {
1584         struct ext4_sb_info *sbi = EXT4_SB(sb);
1585         char *qname, *old_qname = get_qf_name(sb, sbi, qtype);
1586         int ret = -1;
1587
1588         if (sb_any_quota_loaded(sb) && !old_qname) {
1589                 ext4_msg(sb, KERN_ERR,
1590                         "Cannot change journaled "
1591                         "quota options when quota turned on");
1592                 return -1;
1593         }
1594         if (ext4_has_feature_quota(sb)) {
1595                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1596                          "ignored when QUOTA feature is enabled");
1597                 return 1;
1598         }
1599         qname = match_strdup(args);
1600         if (!qname) {
1601                 ext4_msg(sb, KERN_ERR,
1602                         "Not enough memory for storing quotafile name");
1603                 return -1;
1604         }
1605         if (old_qname) {
1606                 if (strcmp(old_qname, qname) == 0)
1607                         ret = 1;
1608                 else
1609                         ext4_msg(sb, KERN_ERR,
1610                                  "%s quota file already specified",
1611                                  QTYPE2NAME(qtype));
1612                 goto errout;
1613         }
1614         if (strchr(qname, '/')) {
1615                 ext4_msg(sb, KERN_ERR,
1616                         "quotafile must be on filesystem root");
1617                 goto errout;
1618         }
1619         rcu_assign_pointer(sbi->s_qf_names[qtype], qname);
1620         set_opt(sb, QUOTA);
1621         return 1;
1622 errout:
1623         kfree(qname);
1624         return ret;
1625 }
1626
1627 static int clear_qf_name(struct super_block *sb, int qtype)
1628 {
1629
1630         struct ext4_sb_info *sbi = EXT4_SB(sb);
1631         char *old_qname = get_qf_name(sb, sbi, qtype);
1632
1633         if (sb_any_quota_loaded(sb) && old_qname) {
1634                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1635                         " when quota turned on");
1636                 return -1;
1637         }
1638         rcu_assign_pointer(sbi->s_qf_names[qtype], NULL);
1639         synchronize_rcu();
1640         kfree(old_qname);
1641         return 1;
1642 }
1643 #endif
1644
1645 #define MOPT_SET        0x0001
1646 #define MOPT_CLEAR      0x0002
1647 #define MOPT_NOSUPPORT  0x0004
1648 #define MOPT_EXPLICIT   0x0008
1649 #define MOPT_CLEAR_ERR  0x0010
1650 #define MOPT_GTE0       0x0020
1651 #ifdef CONFIG_QUOTA
1652 #define MOPT_Q          0
1653 #define MOPT_QFMT       0x0040
1654 #else
1655 #define MOPT_Q          MOPT_NOSUPPORT
1656 #define MOPT_QFMT       MOPT_NOSUPPORT
1657 #endif
1658 #define MOPT_DATAJ      0x0080
1659 #define MOPT_NO_EXT2    0x0100
1660 #define MOPT_NO_EXT3    0x0200
1661 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1662 #define MOPT_STRING     0x0400
1663
1664 static const struct mount_opts {
1665         int     token;
1666         int     mount_opt;
1667         int     flags;
1668 } ext4_mount_opts[] = {
1669         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1670         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1671         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1672         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1673         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1674         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1675         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1676          MOPT_EXT4_ONLY | MOPT_SET},
1677         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1678          MOPT_EXT4_ONLY | MOPT_CLEAR},
1679         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1680         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1681         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1682          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1683         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1684          MOPT_EXT4_ONLY | MOPT_CLEAR},
1685         {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET},
1686         {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR},
1687         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1688          MOPT_EXT4_ONLY | MOPT_CLEAR},
1689         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1690          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1691         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1692                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1693          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1694         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1695         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1696         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1697         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1698         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1699          MOPT_NO_EXT2},
1700         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1701          MOPT_NO_EXT2},
1702         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1703         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1704         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1705         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1706         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1707         {Opt_commit, 0, MOPT_GTE0},
1708         {Opt_max_batch_time, 0, MOPT_GTE0},
1709         {Opt_min_batch_time, 0, MOPT_GTE0},
1710         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1711         {Opt_init_itable, 0, MOPT_GTE0},
1712         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1713         {Opt_stripe, 0, MOPT_GTE0},
1714         {Opt_resuid, 0, MOPT_GTE0},
1715         {Opt_resgid, 0, MOPT_GTE0},
1716         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1717         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1718         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1719         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1720         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1721         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1722          MOPT_NO_EXT2 | MOPT_DATAJ},
1723         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1724         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1725 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1726         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1727         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1728 #else
1729         {Opt_acl, 0, MOPT_NOSUPPORT},
1730         {Opt_noacl, 0, MOPT_NOSUPPORT},
1731 #endif
1732         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1733         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1734         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1735         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1736         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1737                                                         MOPT_SET | MOPT_Q},
1738         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1739                                                         MOPT_SET | MOPT_Q},
1740         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1741                                                         MOPT_SET | MOPT_Q},
1742         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1743                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1744                                                         MOPT_CLEAR | MOPT_Q},
1745         {Opt_usrjquota, 0, MOPT_Q},
1746         {Opt_grpjquota, 0, MOPT_Q},
1747         {Opt_offusrjquota, 0, MOPT_Q},
1748         {Opt_offgrpjquota, 0, MOPT_Q},
1749         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1750         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1751         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1752         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1753         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1754         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1755         {Opt_err, 0, 0}
1756 };
1757
1758 #ifdef CONFIG_UNICODE
1759 static const struct ext4_sb_encodings {
1760         __u16 magic;
1761         char *name;
1762         char *version;
1763 } ext4_sb_encoding_map[] = {
1764         {EXT4_ENC_UTF8_12_1, "utf8", "12.1.0"},
1765 };
1766
1767 static int ext4_sb_read_encoding(const struct ext4_super_block *es,
1768                                  const struct ext4_sb_encodings **encoding,
1769                                  __u16 *flags)
1770 {
1771         __u16 magic = le16_to_cpu(es->s_encoding);
1772         int i;
1773
1774         for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++)
1775                 if (magic == ext4_sb_encoding_map[i].magic)
1776                         break;
1777
1778         if (i >= ARRAY_SIZE(ext4_sb_encoding_map))
1779                 return -EINVAL;
1780
1781         *encoding = &ext4_sb_encoding_map[i];
1782         *flags = le16_to_cpu(es->s_encoding_flags);
1783
1784         return 0;
1785 }
1786 #endif
1787
1788 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1789                             substring_t *args, unsigned long *journal_devnum,
1790                             unsigned int *journal_ioprio, int is_remount)
1791 {
1792         struct ext4_sb_info *sbi = EXT4_SB(sb);
1793         const struct mount_opts *m;
1794         kuid_t uid;
1795         kgid_t gid;
1796         int arg = 0;
1797
1798 #ifdef CONFIG_QUOTA
1799         if (token == Opt_usrjquota)
1800                 return set_qf_name(sb, USRQUOTA, &args[0]);
1801         else if (token == Opt_grpjquota)
1802                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1803         else if (token == Opt_offusrjquota)
1804                 return clear_qf_name(sb, USRQUOTA);
1805         else if (token == Opt_offgrpjquota)
1806                 return clear_qf_name(sb, GRPQUOTA);
1807 #endif
1808         switch (token) {
1809         case Opt_noacl:
1810         case Opt_nouser_xattr:
1811                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1812                 break;
1813         case Opt_sb:
1814                 return 1;       /* handled by get_sb_block() */
1815         case Opt_removed:
1816                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1817                 return 1;
1818         case Opt_abort:
1819                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1820                 return 1;
1821         case Opt_i_version:
1822                 sb->s_flags |= SB_I_VERSION;
1823                 return 1;
1824         case Opt_lazytime:
1825                 sb->s_flags |= SB_LAZYTIME;
1826                 return 1;
1827         case Opt_nolazytime:
1828                 sb->s_flags &= ~SB_LAZYTIME;
1829                 return 1;
1830         }
1831
1832         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1833                 if (token == m->token)
1834                         break;
1835
1836         if (m->token == Opt_err) {
1837                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1838                          "or missing value", opt);
1839                 return -1;
1840         }
1841
1842         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1843                 ext4_msg(sb, KERN_ERR,
1844                          "Mount option \"%s\" incompatible with ext2", opt);
1845                 return -1;
1846         }
1847         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1848                 ext4_msg(sb, KERN_ERR,
1849                          "Mount option \"%s\" incompatible with ext3", opt);
1850                 return -1;
1851         }
1852
1853         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1854                 return -1;
1855         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1856                 return -1;
1857         if (m->flags & MOPT_EXPLICIT) {
1858                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1859                         set_opt2(sb, EXPLICIT_DELALLOC);
1860                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1861                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1862                 } else
1863                         return -1;
1864         }
1865         if (m->flags & MOPT_CLEAR_ERR)
1866                 clear_opt(sb, ERRORS_MASK);
1867         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1868                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1869                          "options when quota turned on");
1870                 return -1;
1871         }
1872
1873         if (m->flags & MOPT_NOSUPPORT) {
1874                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1875         } else if (token == Opt_commit) {
1876                 if (arg == 0)
1877                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1878                 sbi->s_commit_interval = HZ * arg;
1879         } else if (token == Opt_debug_want_extra_isize) {
1880                 sbi->s_want_extra_isize = arg;
1881         } else if (token == Opt_max_batch_time) {
1882                 sbi->s_max_batch_time = arg;
1883         } else if (token == Opt_min_batch_time) {
1884                 sbi->s_min_batch_time = arg;
1885         } else if (token == Opt_inode_readahead_blks) {
1886                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1887                         ext4_msg(sb, KERN_ERR,
1888                                  "EXT4-fs: inode_readahead_blks must be "
1889                                  "0 or a power of 2 smaller than 2^31");
1890                         return -1;
1891                 }
1892                 sbi->s_inode_readahead_blks = arg;
1893         } else if (token == Opt_init_itable) {
1894                 set_opt(sb, INIT_INODE_TABLE);
1895                 if (!args->from)
1896                         arg = EXT4_DEF_LI_WAIT_MULT;
1897                 sbi->s_li_wait_mult = arg;
1898         } else if (token == Opt_max_dir_size_kb) {
1899                 sbi->s_max_dir_size_kb = arg;
1900         } else if (token == Opt_stripe) {
1901                 sbi->s_stripe = arg;
1902         } else if (token == Opt_resuid) {
1903                 uid = make_kuid(current_user_ns(), arg);
1904                 if (!uid_valid(uid)) {
1905                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1906                         return -1;
1907                 }
1908                 sbi->s_resuid = uid;
1909         } else if (token == Opt_resgid) {
1910                 gid = make_kgid(current_user_ns(), arg);
1911                 if (!gid_valid(gid)) {
1912                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1913                         return -1;
1914                 }
1915                 sbi->s_resgid = gid;
1916         } else if (token == Opt_journal_dev) {
1917                 if (is_remount) {
1918                         ext4_msg(sb, KERN_ERR,
1919                                  "Cannot specify journal on remount");
1920                         return -1;
1921                 }
1922                 *journal_devnum = arg;
1923         } else if (token == Opt_journal_path) {
1924                 char *journal_path;
1925                 struct inode *journal_inode;
1926                 struct path path;
1927                 int error;
1928
1929                 if (is_remount) {
1930                         ext4_msg(sb, KERN_ERR,
1931                                  "Cannot specify journal on remount");
1932                         return -1;
1933                 }
1934                 journal_path = match_strdup(&args[0]);
1935                 if (!journal_path) {
1936                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1937                                 "journal device string");
1938                         return -1;
1939                 }
1940
1941                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1942                 if (error) {
1943                         ext4_msg(sb, KERN_ERR, "error: could not find "
1944                                 "journal device path: error %d", error);
1945                         kfree(journal_path);
1946                         return -1;
1947                 }
1948
1949                 journal_inode = d_inode(path.dentry);
1950                 if (!S_ISBLK(journal_inode->i_mode)) {
1951                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1952                                 "is not a block device", journal_path);
1953                         path_put(&path);
1954                         kfree(journal_path);
1955                         return -1;
1956                 }
1957
1958                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1959                 path_put(&path);
1960                 kfree(journal_path);
1961         } else if (token == Opt_journal_ioprio) {
1962                 if (arg > 7) {
1963                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1964                                  " (must be 0-7)");
1965                         return -1;
1966                 }
1967                 *journal_ioprio =
1968                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1969         } else if (token == Opt_test_dummy_encryption) {
1970 #ifdef CONFIG_FS_ENCRYPTION
1971                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1972                 ext4_msg(sb, KERN_WARNING,
1973                          "Test dummy encryption mode enabled");
1974 #else
1975                 ext4_msg(sb, KERN_WARNING,
1976                          "Test dummy encryption mount option ignored");
1977 #endif
1978         } else if (m->flags & MOPT_DATAJ) {
1979                 if (is_remount) {
1980                         if (!sbi->s_journal)
1981                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1982                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1983                                 ext4_msg(sb, KERN_ERR,
1984                                          "Cannot change data mode on remount");
1985                                 return -1;
1986                         }
1987                 } else {
1988                         clear_opt(sb, DATA_FLAGS);
1989                         sbi->s_mount_opt |= m->mount_opt;
1990                 }
1991 #ifdef CONFIG_QUOTA
1992         } else if (m->flags & MOPT_QFMT) {
1993                 if (sb_any_quota_loaded(sb) &&
1994                     sbi->s_jquota_fmt != m->mount_opt) {
1995                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1996                                  "quota options when quota turned on");
1997                         return -1;
1998                 }
1999                 if (ext4_has_feature_quota(sb)) {
2000                         ext4_msg(sb, KERN_INFO,
2001                                  "Quota format mount options ignored "
2002                                  "when QUOTA feature is enabled");
2003                         return 1;
2004                 }
2005                 sbi->s_jquota_fmt = m->mount_opt;
2006 #endif
2007         } else if (token == Opt_dax) {
2008 #ifdef CONFIG_FS_DAX
2009                 ext4_msg(sb, KERN_WARNING,
2010                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
2011                 sbi->s_mount_opt |= m->mount_opt;
2012 #else
2013                 ext4_msg(sb, KERN_INFO, "dax option not supported");
2014                 return -1;
2015 #endif
2016         } else if (token == Opt_data_err_abort) {
2017                 sbi->s_mount_opt |= m->mount_opt;
2018         } else if (token == Opt_data_err_ignore) {
2019                 sbi->s_mount_opt &= ~m->mount_opt;
2020         } else {
2021                 if (!args->from)
2022                         arg = 1;
2023                 if (m->flags & MOPT_CLEAR)
2024                         arg = !arg;
2025                 else if (unlikely(!(m->flags & MOPT_SET))) {
2026                         ext4_msg(sb, KERN_WARNING,
2027                                  "buggy handling of option %s", opt);
2028                         WARN_ON(1);
2029                         return -1;
2030                 }
2031                 if (arg != 0)
2032                         sbi->s_mount_opt |= m->mount_opt;
2033                 else
2034                         sbi->s_mount_opt &= ~m->mount_opt;
2035         }
2036         return 1;
2037 }
2038
2039 static int parse_options(char *options, struct super_block *sb,
2040                          unsigned long *journal_devnum,
2041                          unsigned int *journal_ioprio,
2042                          int is_remount)
2043 {
2044         struct ext4_sb_info *sbi = EXT4_SB(sb);
2045         char *p, __maybe_unused *usr_qf_name, __maybe_unused *grp_qf_name;
2046         substring_t args[MAX_OPT_ARGS];
2047         int token;
2048
2049         if (!options)
2050                 return 1;
2051
2052         while ((p = strsep(&options, ",")) != NULL) {
2053                 if (!*p)
2054                         continue;
2055                 /*
2056                  * Initialize args struct so we know whether arg was
2057                  * found; some options take optional arguments.
2058                  */
2059                 args[0].to = args[0].from = NULL;
2060                 token = match_token(p, tokens, args);
2061                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
2062                                      journal_ioprio, is_remount) < 0)
2063                         return 0;
2064         }
2065 #ifdef CONFIG_QUOTA
2066         /*
2067          * We do the test below only for project quotas. 'usrquota' and
2068          * 'grpquota' mount options are allowed even without quota feature
2069          * to support legacy quotas in quota files.
2070          */
2071         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
2072                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
2073                          "Cannot enable project quota enforcement.");
2074                 return 0;
2075         }
2076         usr_qf_name = get_qf_name(sb, sbi, USRQUOTA);
2077         grp_qf_name = get_qf_name(sb, sbi, GRPQUOTA);
2078         if (usr_qf_name || grp_qf_name) {
2079                 if (test_opt(sb, USRQUOTA) && usr_qf_name)
2080                         clear_opt(sb, USRQUOTA);
2081
2082                 if (test_opt(sb, GRPQUOTA) && grp_qf_name)
2083                         clear_opt(sb, GRPQUOTA);
2084
2085                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
2086                         ext4_msg(sb, KERN_ERR, "old and new quota "
2087                                         "format mixing");
2088                         return 0;
2089                 }
2090
2091                 if (!sbi->s_jquota_fmt) {
2092                         ext4_msg(sb, KERN_ERR, "journaled quota format "
2093                                         "not specified");
2094                         return 0;
2095                 }
2096         }
2097 #endif
2098         if (test_opt(sb, DIOREAD_NOLOCK)) {
2099                 int blocksize =
2100                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
2101
2102                 if (blocksize < PAGE_SIZE) {
2103                         ext4_msg(sb, KERN_ERR, "can't mount with "
2104                                  "dioread_nolock if block size != PAGE_SIZE");
2105                         return 0;
2106                 }
2107         }
2108         return 1;
2109 }
2110
2111 static inline void ext4_show_quota_options(struct seq_file *seq,
2112                                            struct super_block *sb)
2113 {
2114 #if defined(CONFIG_QUOTA)
2115         struct ext4_sb_info *sbi = EXT4_SB(sb);
2116         char *usr_qf_name, *grp_qf_name;
2117
2118         if (sbi->s_jquota_fmt) {
2119                 char *fmtname = "";
2120
2121                 switch (sbi->s_jquota_fmt) {
2122                 case QFMT_VFS_OLD:
2123                         fmtname = "vfsold";
2124                         break;
2125                 case QFMT_VFS_V0:
2126                         fmtname = "vfsv0";
2127                         break;
2128                 case QFMT_VFS_V1:
2129                         fmtname = "vfsv1";
2130                         break;
2131                 }
2132                 seq_printf(seq, ",jqfmt=%s", fmtname);
2133         }
2134
2135         rcu_read_lock();
2136         usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]);
2137         grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]);
2138         if (usr_qf_name)
2139                 seq_show_option(seq, "usrjquota", usr_qf_name);
2140         if (grp_qf_name)
2141                 seq_show_option(seq, "grpjquota", grp_qf_name);
2142         rcu_read_unlock();
2143 #endif
2144 }
2145
2146 static const char *token2str(int token)
2147 {
2148         const struct match_token *t;
2149
2150         for (t = tokens; t->token != Opt_err; t++)
2151                 if (t->token == token && !strchr(t->pattern, '='))
2152                         break;
2153         return t->pattern;
2154 }
2155
2156 /*
2157  * Show an option if
2158  *  - it's set to a non-default value OR
2159  *  - if the per-sb default is different from the global default
2160  */
2161 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2162                               int nodefs)
2163 {
2164         struct ext4_sb_info *sbi = EXT4_SB(sb);
2165         struct ext4_super_block *es = sbi->s_es;
2166         int def_errors, def_mount_opt = sbi->s_def_mount_opt;
2167         const struct mount_opts *m;
2168         char sep = nodefs ? '\n' : ',';
2169
2170 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2171 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2172
2173         if (sbi->s_sb_block != 1)
2174                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2175
2176         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2177                 int want_set = m->flags & MOPT_SET;
2178                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2179                     (m->flags & MOPT_CLEAR_ERR))
2180                         continue;
2181                 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2182                         continue; /* skip if same as the default */
2183                 if ((want_set &&
2184                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2185                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2186                         continue; /* select Opt_noFoo vs Opt_Foo */
2187                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2188         }
2189
2190         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2191             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2192                 SEQ_OPTS_PRINT("resuid=%u",
2193                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2194         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2195             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2196                 SEQ_OPTS_PRINT("resgid=%u",
2197                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2198         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2199         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2200                 SEQ_OPTS_PUTS("errors=remount-ro");
2201         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2202                 SEQ_OPTS_PUTS("errors=continue");
2203         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2204                 SEQ_OPTS_PUTS("errors=panic");
2205         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2206                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2207         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2208                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2209         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2210                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2211         if (sb->s_flags & SB_I_VERSION)
2212                 SEQ_OPTS_PUTS("i_version");
2213         if (nodefs || sbi->s_stripe)
2214                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2215         if (nodefs || EXT4_MOUNT_DATA_FLAGS &
2216                         (sbi->s_mount_opt ^ def_mount_opt)) {
2217                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2218                         SEQ_OPTS_PUTS("data=journal");
2219                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2220                         SEQ_OPTS_PUTS("data=ordered");
2221                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2222                         SEQ_OPTS_PUTS("data=writeback");
2223         }
2224         if (nodefs ||
2225             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2226                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2227                                sbi->s_inode_readahead_blks);
2228
2229         if (test_opt(sb, INIT_INODE_TABLE) && (nodefs ||
2230                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2231                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2232         if (nodefs || sbi->s_max_dir_size_kb)
2233                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2234         if (test_opt(sb, DATA_ERR_ABORT))
2235                 SEQ_OPTS_PUTS("data_err=abort");
2236         if (DUMMY_ENCRYPTION_ENABLED(sbi))
2237                 SEQ_OPTS_PUTS("test_dummy_encryption");
2238
2239         ext4_show_quota_options(seq, sb);
2240         return 0;
2241 }
2242
2243 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2244 {
2245         return _ext4_show_options(seq, root->d_sb, 0);
2246 }
2247
2248 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2249 {
2250         struct super_block *sb = seq->private;
2251         int rc;
2252
2253         seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
2254         rc = _ext4_show_options(seq, sb, 1);
2255         seq_puts(seq, "\n");
2256         return rc;
2257 }
2258
2259 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2260                             int read_only)
2261 {
2262         struct ext4_sb_info *sbi = EXT4_SB(sb);
2263         int err = 0;
2264
2265         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2266                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2267                          "forcing read-only mode");
2268                 err = -EROFS;
2269         }
2270         if (read_only)
2271                 goto done;
2272         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2273                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2274                          "running e2fsck is recommended");
2275         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2276                 ext4_msg(sb, KERN_WARNING,
2277                          "warning: mounting fs with errors, "
2278                          "running e2fsck is recommended");
2279         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2280                  le16_to_cpu(es->s_mnt_count) >=
2281                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2282                 ext4_msg(sb, KERN_WARNING,
2283                          "warning: maximal mount count reached, "
2284                          "running e2fsck is recommended");
2285         else if (le32_to_cpu(es->s_checkinterval) &&
2286                  (ext4_get_tstamp(es, s_lastcheck) +
2287                   le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds()))
2288                 ext4_msg(sb, KERN_WARNING,
2289                          "warning: checktime reached, "
2290                          "running e2fsck is recommended");
2291         if (!sbi->s_journal)
2292                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2293         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2294                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2295         le16_add_cpu(&es->s_mnt_count, 1);
2296         ext4_update_tstamp(es, s_mtime);
2297         if (sbi->s_journal)
2298                 ext4_set_feature_journal_needs_recovery(sb);
2299
2300         err = ext4_commit_super(sb, 1);
2301 done:
2302         if (test_opt(sb, DEBUG))
2303                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2304                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2305                         sb->s_blocksize,
2306                         sbi->s_groups_count,
2307                         EXT4_BLOCKS_PER_GROUP(sb),
2308                         EXT4_INODES_PER_GROUP(sb),
2309                         sbi->s_mount_opt, sbi->s_mount_opt2);
2310
2311         cleancache_init_fs(sb);
2312         return err;
2313 }
2314
2315 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2316 {
2317         struct ext4_sb_info *sbi = EXT4_SB(sb);
2318         struct flex_groups *new_groups;
2319         int size;
2320
2321         if (!sbi->s_log_groups_per_flex)
2322                 return 0;
2323
2324         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2325         if (size <= sbi->s_flex_groups_allocated)
2326                 return 0;
2327
2328         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
2329         new_groups = kvzalloc(size, GFP_KERNEL);
2330         if (!new_groups) {
2331                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
2332                          size / (int) sizeof(struct flex_groups));
2333                 return -ENOMEM;
2334         }
2335
2336         if (sbi->s_flex_groups) {
2337                 memcpy(new_groups, sbi->s_flex_groups,
2338                        (sbi->s_flex_groups_allocated *
2339                         sizeof(struct flex_groups)));
2340                 kvfree(sbi->s_flex_groups);
2341         }
2342         sbi->s_flex_groups = new_groups;
2343         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
2344         return 0;
2345 }
2346
2347 static int ext4_fill_flex_info(struct super_block *sb)
2348 {
2349         struct ext4_sb_info *sbi = EXT4_SB(sb);
2350         struct ext4_group_desc *gdp = NULL;
2351         ext4_group_t flex_group;
2352         int i, err;
2353
2354         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2355         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2356                 sbi->s_log_groups_per_flex = 0;
2357                 return 1;
2358         }
2359
2360         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2361         if (err)
2362                 goto failed;
2363
2364         for (i = 0; i < sbi->s_groups_count; i++) {
2365                 gdp = ext4_get_group_desc(sb, i, NULL);
2366
2367                 flex_group = ext4_flex_group(sbi, i);
2368                 atomic_add(ext4_free_inodes_count(sb, gdp),
2369                            &sbi->s_flex_groups[flex_group].free_inodes);
2370                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2371                              &sbi->s_flex_groups[flex_group].free_clusters);
2372                 atomic_add(ext4_used_dirs_count(sb, gdp),
2373                            &sbi->s_flex_groups[flex_group].used_dirs);
2374         }
2375
2376         return 1;
2377 failed:
2378         return 0;
2379 }
2380
2381 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2382                                    struct ext4_group_desc *gdp)
2383 {
2384         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2385         __u16 crc = 0;
2386         __le32 le_group = cpu_to_le32(block_group);
2387         struct ext4_sb_info *sbi = EXT4_SB(sb);
2388
2389         if (ext4_has_metadata_csum(sbi->s_sb)) {
2390                 /* Use new metadata_csum algorithm */
2391                 __u32 csum32;
2392                 __u16 dummy_csum = 0;
2393
2394                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2395                                      sizeof(le_group));
2396                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2397                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2398                                      sizeof(dummy_csum));
2399                 offset += sizeof(dummy_csum);
2400                 if (offset < sbi->s_desc_size)
2401                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2402                                              sbi->s_desc_size - offset);
2403
2404                 crc = csum32 & 0xFFFF;
2405                 goto out;
2406         }
2407
2408         /* old crc16 code */
2409         if (!ext4_has_feature_gdt_csum(sb))
2410                 return 0;
2411
2412         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2413         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2414         crc = crc16(crc, (__u8 *)gdp, offset);
2415         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2416         /* for checksum of struct ext4_group_desc do the rest...*/
2417         if (ext4_has_feature_64bit(sb) &&
2418             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2419                 crc = crc16(crc, (__u8 *)gdp + offset,
2420                             le16_to_cpu(sbi->s_es->s_desc_size) -
2421                                 offset);
2422
2423 out:
2424         return cpu_to_le16(crc);
2425 }
2426
2427 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2428                                 struct ext4_group_desc *gdp)
2429 {
2430         if (ext4_has_group_desc_csum(sb) &&
2431             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2432                 return 0;
2433
2434         return 1;
2435 }
2436
2437 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2438                               struct ext4_group_desc *gdp)
2439 {
2440         if (!ext4_has_group_desc_csum(sb))
2441                 return;
2442         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2443 }
2444
2445 /* Called at mount-time, super-block is locked */
2446 static int ext4_check_descriptors(struct super_block *sb,
2447                                   ext4_fsblk_t sb_block,
2448                                   ext4_group_t *first_not_zeroed)
2449 {
2450         struct ext4_sb_info *sbi = EXT4_SB(sb);
2451         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2452         ext4_fsblk_t last_block;
2453         ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0);
2454         ext4_fsblk_t block_bitmap;
2455         ext4_fsblk_t inode_bitmap;
2456         ext4_fsblk_t inode_table;
2457         int flexbg_flag = 0;
2458         ext4_group_t i, grp = sbi->s_groups_count;
2459
2460         if (ext4_has_feature_flex_bg(sb))
2461                 flexbg_flag = 1;
2462
2463         ext4_debug("Checking group descriptors");
2464
2465         for (i = 0; i < sbi->s_groups_count; i++) {
2466                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2467
2468                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2469                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2470                 else
2471                         last_block = first_block +
2472                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2473
2474                 if ((grp == sbi->s_groups_count) &&
2475                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2476                         grp = i;
2477
2478                 block_bitmap = ext4_block_bitmap(sb, gdp);
2479                 if (block_bitmap == sb_block) {
2480                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2481                                  "Block bitmap for group %u overlaps "
2482                                  "superblock", i);
2483                         if (!sb_rdonly(sb))
2484                                 return 0;
2485                 }
2486                 if (block_bitmap >= sb_block + 1 &&
2487                     block_bitmap <= last_bg_block) {
2488                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2489                                  "Block bitmap for group %u overlaps "
2490                                  "block group descriptors", i);
2491                         if (!sb_rdonly(sb))
2492                                 return 0;
2493                 }
2494                 if (block_bitmap < first_block || block_bitmap > last_block) {
2495                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2496                                "Block bitmap for group %u not in group "
2497                                "(block %llu)!", i, block_bitmap);
2498                         return 0;
2499                 }
2500                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2501                 if (inode_bitmap == sb_block) {
2502                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2503                                  "Inode bitmap for group %u overlaps "
2504                                  "superblock", i);
2505                         if (!sb_rdonly(sb))
2506                                 return 0;
2507                 }
2508                 if (inode_bitmap >= sb_block + 1 &&
2509                     inode_bitmap <= last_bg_block) {
2510                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2511                                  "Inode bitmap for group %u overlaps "
2512                                  "block group descriptors", i);
2513                         if (!sb_rdonly(sb))
2514                                 return 0;
2515                 }
2516                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2517                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2518                                "Inode bitmap for group %u not in group "
2519                                "(block %llu)!", i, inode_bitmap);
2520                         return 0;
2521                 }
2522                 inode_table = ext4_inode_table(sb, gdp);
2523                 if (inode_table == sb_block) {
2524                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2525                                  "Inode table for group %u overlaps "
2526                                  "superblock", i);
2527                         if (!sb_rdonly(sb))
2528                                 return 0;
2529                 }
2530                 if (inode_table >= sb_block + 1 &&
2531                     inode_table <= last_bg_block) {
2532                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2533                                  "Inode table for group %u overlaps "
2534                                  "block group descriptors", i);
2535                         if (!sb_rdonly(sb))
2536                                 return 0;
2537                 }
2538                 if (inode_table < first_block ||
2539                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2540                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2541                                "Inode table for group %u not in group "
2542                                "(block %llu)!", i, inode_table);
2543                         return 0;
2544                 }
2545                 ext4_lock_group(sb, i);
2546                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2547                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2548                                  "Checksum for group %u failed (%u!=%u)",
2549                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2550                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2551                         if (!sb_rdonly(sb)) {
2552                                 ext4_unlock_group(sb, i);
2553                                 return 0;
2554                         }
2555                 }
2556                 ext4_unlock_group(sb, i);
2557                 if (!flexbg_flag)
2558                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2559         }
2560         if (NULL != first_not_zeroed)
2561                 *first_not_zeroed = grp;
2562         return 1;
2563 }
2564
2565 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2566  * the superblock) which were deleted from all directories, but held open by
2567  * a process at the time of a crash.  We walk the list and try to delete these
2568  * inodes at recovery time (only with a read-write filesystem).
2569  *
2570  * In order to keep the orphan inode chain consistent during traversal (in
2571  * case of crash during recovery), we link each inode into the superblock
2572  * orphan list_head and handle it the same way as an inode deletion during
2573  * normal operation (which journals the operations for us).
2574  *
2575  * We only do an iget() and an iput() on each inode, which is very safe if we
2576  * accidentally point at an in-use or already deleted inode.  The worst that
2577  * can happen in this case is that we get a "bit already cleared" message from
2578  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2579  * e2fsck was run on this filesystem, and it must have already done the orphan
2580  * inode cleanup for us, so we can safely abort without any further action.
2581  */
2582 static void ext4_orphan_cleanup(struct super_block *sb,
2583                                 struct ext4_super_block *es)
2584 {
2585         unsigned int s_flags = sb->s_flags;
2586         int ret, nr_orphans = 0, nr_truncates = 0;
2587 #ifdef CONFIG_QUOTA
2588         int quota_update = 0;
2589         int i;
2590 #endif
2591         if (!es->s_last_orphan) {
2592                 jbd_debug(4, "no orphan inodes to clean up\n");
2593                 return;
2594         }
2595
2596         if (bdev_read_only(sb->s_bdev)) {
2597                 ext4_msg(sb, KERN_ERR, "write access "
2598                         "unavailable, skipping orphan cleanup");
2599                 return;
2600         }
2601
2602         /* Check if feature set would not allow a r/w mount */
2603         if (!ext4_feature_set_ok(sb, 0)) {
2604                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2605                          "unknown ROCOMPAT features");
2606                 return;
2607         }
2608
2609         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2610                 /* don't clear list on RO mount w/ errors */
2611                 if (es->s_last_orphan && !(s_flags & SB_RDONLY)) {
2612                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2613                                   "clearing orphan list.\n");
2614                         es->s_last_orphan = 0;
2615                 }
2616                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2617                 return;
2618         }
2619
2620         if (s_flags & SB_RDONLY) {
2621                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2622                 sb->s_flags &= ~SB_RDONLY;
2623         }
2624 #ifdef CONFIG_QUOTA
2625         /* Needed for iput() to work correctly and not trash data */
2626         sb->s_flags |= SB_ACTIVE;
2627
2628         /*
2629          * Turn on quotas which were not enabled for read-only mounts if
2630          * filesystem has quota feature, so that they are updated correctly.
2631          */
2632         if (ext4_has_feature_quota(sb) && (s_flags & SB_RDONLY)) {
2633                 int ret = ext4_enable_quotas(sb);
2634
2635                 if (!ret)
2636                         quota_update = 1;
2637                 else
2638                         ext4_msg(sb, KERN_ERR,
2639                                 "Cannot turn on quotas: error %d", ret);
2640         }
2641
2642         /* Turn on journaled quotas used for old sytle */
2643         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2644                 if (EXT4_SB(sb)->s_qf_names[i]) {
2645                         int ret = ext4_quota_on_mount(sb, i);
2646
2647                         if (!ret)
2648                                 quota_update = 1;
2649                         else
2650                                 ext4_msg(sb, KERN_ERR,
2651                                         "Cannot turn on journaled "
2652                                         "quota: type %d: error %d", i, ret);
2653                 }
2654         }
2655 #endif
2656
2657         while (es->s_last_orphan) {
2658                 struct inode *inode;
2659
2660                 /*
2661                  * We may have encountered an error during cleanup; if
2662                  * so, skip the rest.
2663                  */
2664                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2665                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2666                         es->s_last_orphan = 0;
2667                         break;
2668                 }
2669
2670                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2671                 if (IS_ERR(inode)) {
2672                         es->s_last_orphan = 0;
2673                         break;
2674                 }
2675
2676                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2677                 dquot_initialize(inode);
2678                 if (inode->i_nlink) {
2679                         if (test_opt(sb, DEBUG))
2680                                 ext4_msg(sb, KERN_DEBUG,
2681                                         "%s: truncating inode %lu to %lld bytes",
2682                                         __func__, inode->i_ino, inode->i_size);
2683                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2684                                   inode->i_ino, inode->i_size);
2685                         inode_lock(inode);
2686                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2687                         ret = ext4_truncate(inode);
2688                         if (ret)
2689                                 ext4_std_error(inode->i_sb, ret);
2690                         inode_unlock(inode);
2691                         nr_truncates++;
2692                 } else {
2693                         if (test_opt(sb, DEBUG))
2694                                 ext4_msg(sb, KERN_DEBUG,
2695                                         "%s: deleting unreferenced inode %lu",
2696                                         __func__, inode->i_ino);
2697                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2698                                   inode->i_ino);
2699                         nr_orphans++;
2700                 }
2701                 iput(inode);  /* The delete magic happens here! */
2702         }
2703
2704 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2705
2706         if (nr_orphans)
2707                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2708                        PLURAL(nr_orphans));
2709         if (nr_truncates)
2710                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2711                        PLURAL(nr_truncates));
2712 #ifdef CONFIG_QUOTA
2713         /* Turn off quotas if they were enabled for orphan cleanup */
2714         if (quota_update) {
2715                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2716                         if (sb_dqopt(sb)->files[i])
2717                                 dquot_quota_off(sb, i);
2718                 }
2719         }
2720 #endif
2721         sb->s_flags = s_flags; /* Restore SB_RDONLY status */
2722 }
2723
2724 /*
2725  * Maximal extent format file size.
2726  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2727  * extent format containers, within a sector_t, and within i_blocks
2728  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2729  * so that won't be a limiting factor.
2730  *
2731  * However there is other limiting factor. We do store extents in the form
2732  * of starting block and length, hence the resulting length of the extent
2733  * covering maximum file size must fit into on-disk format containers as
2734  * well. Given that length is always by 1 unit bigger than max unit (because
2735  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2736  *
2737  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2738  */
2739 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2740 {
2741         loff_t res;
2742         loff_t upper_limit = MAX_LFS_FILESIZE;
2743
2744         BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64));
2745
2746         if (!has_huge_files) {
2747                 upper_limit = (1LL << 32) - 1;
2748
2749                 /* total blocks in file system block size */
2750                 upper_limit >>= (blkbits - 9);
2751                 upper_limit <<= blkbits;
2752         }
2753
2754         /*
2755          * 32-bit extent-start container, ee_block. We lower the maxbytes
2756          * by one fs block, so ee_len can cover the extent of maximum file
2757          * size
2758          */
2759         res = (1LL << 32) - 1;
2760         res <<= blkbits;
2761
2762         /* Sanity check against vm- & vfs- imposed limits */
2763         if (res > upper_limit)
2764                 res = upper_limit;
2765
2766         return res;
2767 }
2768
2769 /*
2770  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2771  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2772  * We need to be 1 filesystem block less than the 2^48 sector limit.
2773  */
2774 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2775 {
2776         loff_t res = EXT4_NDIR_BLOCKS;
2777         int meta_blocks;
2778         loff_t upper_limit;
2779         /* This is calculated to be the largest file size for a dense, block
2780          * mapped file such that the file's total number of 512-byte sectors,
2781          * including data and all indirect blocks, does not exceed (2^48 - 1).
2782          *
2783          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2784          * number of 512-byte sectors of the file.
2785          */
2786
2787         if (!has_huge_files) {
2788                 /*
2789                  * !has_huge_files or implies that the inode i_block field
2790                  * represents total file blocks in 2^32 512-byte sectors ==
2791                  * size of vfs inode i_blocks * 8
2792                  */
2793                 upper_limit = (1LL << 32) - 1;
2794
2795                 /* total blocks in file system block size */
2796                 upper_limit >>= (bits - 9);
2797
2798         } else {
2799                 /*
2800                  * We use 48 bit ext4_inode i_blocks
2801                  * With EXT4_HUGE_FILE_FL set the i_blocks
2802                  * represent total number of blocks in
2803                  * file system block size
2804                  */
2805                 upper_limit = (1LL << 48) - 1;
2806
2807         }
2808
2809         /* indirect blocks */
2810         meta_blocks = 1;
2811         /* double indirect blocks */
2812         meta_blocks += 1 + (1LL << (bits-2));
2813         /* tripple indirect blocks */
2814         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2815
2816         upper_limit -= meta_blocks;
2817         upper_limit <<= bits;
2818
2819         res += 1LL << (bits-2);
2820         res += 1LL << (2*(bits-2));
2821         res += 1LL << (3*(bits-2));
2822         res <<= bits;
2823         if (res > upper_limit)
2824                 res = upper_limit;
2825
2826         if (res > MAX_LFS_FILESIZE)
2827                 res = MAX_LFS_FILESIZE;
2828
2829         return res;
2830 }
2831
2832 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2833                                    ext4_fsblk_t logical_sb_block, int nr)
2834 {
2835         struct ext4_sb_info *sbi = EXT4_SB(sb);
2836         ext4_group_t bg, first_meta_bg;
2837         int has_super = 0;
2838
2839         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2840
2841         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2842                 return logical_sb_block + nr + 1;
2843         bg = sbi->s_desc_per_block * nr;
2844         if (ext4_bg_has_super(sb, bg))
2845                 has_super = 1;
2846
2847         /*
2848          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2849          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2850          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2851          * compensate.
2852          */
2853         if (sb->s_blocksize == 1024 && nr == 0 &&
2854             le32_to_cpu(sbi->s_es->s_first_data_block) == 0)
2855                 has_super++;
2856
2857         return (has_super + ext4_group_first_block_no(sb, bg));
2858 }
2859
2860 /**
2861  * ext4_get_stripe_size: Get the stripe size.
2862  * @sbi: In memory super block info
2863  *
2864  * If we have specified it via mount option, then
2865  * use the mount option value. If the value specified at mount time is
2866  * greater than the blocks per group use the super block value.
2867  * If the super block value is greater than blocks per group return 0.
2868  * Allocator needs it be less than blocks per group.
2869  *
2870  */
2871 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2872 {
2873         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2874         unsigned long stripe_width =
2875                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2876         int ret;
2877
2878         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2879                 ret = sbi->s_stripe;
2880         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
2881                 ret = stripe_width;
2882         else if (stride && stride <= sbi->s_blocks_per_group)
2883                 ret = stride;
2884         else
2885                 ret = 0;
2886
2887         /*
2888          * If the stripe width is 1, this makes no sense and
2889          * we set it to 0 to turn off stripe handling code.
2890          */
2891         if (ret <= 1)
2892                 ret = 0;
2893
2894         return ret;
2895 }
2896
2897 /*
2898  * Check whether this filesystem can be mounted based on
2899  * the features present and the RDONLY/RDWR mount requested.
2900  * Returns 1 if this filesystem can be mounted as requested,
2901  * 0 if it cannot be.
2902  */
2903 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2904 {
2905         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2906                 ext4_msg(sb, KERN_ERR,
2907                         "Couldn't mount because of "
2908                         "unsupported optional features (%x)",
2909                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2910                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2911                 return 0;
2912         }
2913
2914 #ifndef CONFIG_UNICODE
2915         if (ext4_has_feature_casefold(sb)) {
2916                 ext4_msg(sb, KERN_ERR,
2917                          "Filesystem with casefold feature cannot be "
2918                          "mounted without CONFIG_UNICODE");
2919                 return 0;
2920         }
2921 #endif
2922
2923         if (readonly)
2924                 return 1;
2925
2926         if (ext4_has_feature_readonly(sb)) {
2927                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2928                 sb->s_flags |= SB_RDONLY;
2929                 return 1;
2930         }
2931
2932         /* Check that feature set is OK for a read-write mount */
2933         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2934                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2935                          "unsupported optional features (%x)",
2936                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2937                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2938                 return 0;
2939         }
2940         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2941                 ext4_msg(sb, KERN_ERR,
2942                          "Can't support bigalloc feature without "
2943                          "extents feature\n");
2944                 return 0;
2945         }
2946
2947 #ifndef CONFIG_QUOTA
2948         if (ext4_has_feature_quota(sb) && !readonly) {
2949                 ext4_msg(sb, KERN_ERR,
2950                          "Filesystem with quota feature cannot be mounted RDWR "
2951                          "without CONFIG_QUOTA");
2952                 return 0;
2953         }
2954         if (ext4_has_feature_project(sb) && !readonly) {
2955                 ext4_msg(sb, KERN_ERR,
2956                          "Filesystem with project quota feature cannot be mounted RDWR "
2957                          "without CONFIG_QUOTA");
2958                 return 0;
2959         }
2960 #endif  /* CONFIG_QUOTA */
2961         return 1;
2962 }
2963
2964 /*
2965  * This function is called once a day if we have errors logged
2966  * on the file system
2967  */
2968 static void print_daily_error_info(struct timer_list *t)
2969 {
2970         struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report);
2971         struct super_block *sb = sbi->s_sb;
2972         struct ext4_super_block *es = sbi->s_es;
2973
2974         if (es->s_error_count)
2975                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2976                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2977                          le32_to_cpu(es->s_error_count));
2978         if (es->s_first_error_time) {
2979                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d",
2980                        sb->s_id,
2981                        ext4_get_tstamp(es, s_first_error_time),
2982                        (int) sizeof(es->s_first_error_func),
2983                        es->s_first_error_func,
2984                        le32_to_cpu(es->s_first_error_line));
2985                 if (es->s_first_error_ino)
2986                         printk(KERN_CONT ": inode %u",
2987                                le32_to_cpu(es->s_first_error_ino));
2988                 if (es->s_first_error_block)
2989                         printk(KERN_CONT ": block %llu", (unsigned long long)
2990                                le64_to_cpu(es->s_first_error_block));
2991                 printk(KERN_CONT "\n");
2992         }
2993         if (es->s_last_error_time) {
2994                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d",
2995                        sb->s_id,
2996                        ext4_get_tstamp(es, s_last_error_time),
2997                        (int) sizeof(es->s_last_error_func),
2998                        es->s_last_error_func,
2999                        le32_to_cpu(es->s_last_error_line));
3000                 if (es->s_last_error_ino)
3001                         printk(KERN_CONT ": inode %u",
3002                                le32_to_cpu(es->s_last_error_ino));
3003                 if (es->s_last_error_block)
3004                         printk(KERN_CONT ": block %llu", (unsigned long long)
3005                                le64_to_cpu(es->s_last_error_block));
3006                 printk(KERN_CONT "\n");
3007         }
3008         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
3009 }
3010
3011 /* Find next suitable group and run ext4_init_inode_table */
3012 static int ext4_run_li_request(struct ext4_li_request *elr)
3013 {
3014         struct ext4_group_desc *gdp = NULL;
3015         ext4_group_t group, ngroups;
3016         struct super_block *sb;
3017         unsigned long timeout = 0;
3018         int ret = 0;
3019
3020         sb = elr->lr_super;
3021         ngroups = EXT4_SB(sb)->s_groups_count;
3022
3023         for (group = elr->lr_next_group; group < ngroups; group++) {
3024                 gdp = ext4_get_group_desc(sb, group, NULL);
3025                 if (!gdp) {
3026                         ret = 1;
3027                         break;
3028                 }
3029
3030                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3031                         break;
3032         }
3033
3034         if (group >= ngroups)
3035                 ret = 1;
3036
3037         if (!ret) {
3038                 timeout = jiffies;
3039                 ret = ext4_init_inode_table(sb, group,
3040                                             elr->lr_timeout ? 0 : 1);
3041                 if (elr->lr_timeout == 0) {
3042                         timeout = (jiffies - timeout) *
3043                                   elr->lr_sbi->s_li_wait_mult;
3044                         elr->lr_timeout = timeout;
3045                 }
3046                 elr->lr_next_sched = jiffies + elr->lr_timeout;
3047                 elr->lr_next_group = group + 1;
3048         }
3049         return ret;
3050 }
3051
3052 /*
3053  * Remove lr_request from the list_request and free the
3054  * request structure. Should be called with li_list_mtx held
3055  */
3056 static void ext4_remove_li_request(struct ext4_li_request *elr)
3057 {
3058         struct ext4_sb_info *sbi;
3059
3060         if (!elr)
3061                 return;
3062
3063         sbi = elr->lr_sbi;
3064
3065         list_del(&elr->lr_request);
3066         sbi->s_li_request = NULL;
3067         kfree(elr);
3068 }
3069
3070 static void ext4_unregister_li_request(struct super_block *sb)
3071 {
3072         mutex_lock(&ext4_li_mtx);
3073         if (!ext4_li_info) {
3074                 mutex_unlock(&ext4_li_mtx);
3075                 return;
3076         }
3077
3078         mutex_lock(&ext4_li_info->li_list_mtx);
3079         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
3080         mutex_unlock(&ext4_li_info->li_list_mtx);
3081         mutex_unlock(&ext4_li_mtx);
3082 }
3083
3084 static struct task_struct *ext4_lazyinit_task;
3085
3086 /*
3087  * This is the function where ext4lazyinit thread lives. It walks
3088  * through the request list searching for next scheduled filesystem.
3089  * When such a fs is found, run the lazy initialization request
3090  * (ext4_rn_li_request) and keep track of the time spend in this
3091  * function. Based on that time we compute next schedule time of
3092  * the request. When walking through the list is complete, compute
3093  * next waking time and put itself into sleep.
3094  */
3095 static int ext4_lazyinit_thread(void *arg)
3096 {
3097         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
3098         struct list_head *pos, *n;
3099         struct ext4_li_request *elr;
3100         unsigned long next_wakeup, cur;
3101
3102         BUG_ON(NULL == eli);
3103
3104 cont_thread:
3105         while (true) {
3106                 next_wakeup = MAX_JIFFY_OFFSET;
3107
3108                 mutex_lock(&eli->li_list_mtx);
3109                 if (list_empty(&eli->li_request_list)) {
3110                         mutex_unlock(&eli->li_list_mtx);
3111                         goto exit_thread;
3112                 }
3113                 list_for_each_safe(pos, n, &eli->li_request_list) {
3114                         int err = 0;
3115                         int progress = 0;
3116                         elr = list_entry(pos, struct ext4_li_request,
3117                                          lr_request);
3118
3119                         if (time_before(jiffies, elr->lr_next_sched)) {
3120                                 if (time_before(elr->lr_next_sched, next_wakeup))
3121                                         next_wakeup = elr->lr_next_sched;
3122                                 continue;
3123                         }
3124                         if (down_read_trylock(&elr->lr_super->s_umount)) {
3125                                 if (sb_start_write_trylock(elr->lr_super)) {
3126                                         progress = 1;
3127                                         /*
3128                                          * We hold sb->s_umount, sb can not
3129                                          * be removed from the list, it is
3130                                          * now safe to drop li_list_mtx
3131                                          */
3132                                         mutex_unlock(&eli->li_list_mtx);
3133                                         err = ext4_run_li_request(elr);
3134                                         sb_end_write(elr->lr_super);
3135                                         mutex_lock(&eli->li_list_mtx);
3136                                         n = pos->next;
3137                                 }
3138                                 up_read((&elr->lr_super->s_umount));
3139                         }
3140                         /* error, remove the lazy_init job */
3141                         if (err) {
3142                                 ext4_remove_li_request(elr);
3143                                 continue;
3144                         }
3145                         if (!progress) {
3146                                 elr->lr_next_sched = jiffies +
3147                                         (prandom_u32()
3148                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3149                         }
3150                         if (time_before(elr->lr_next_sched, next_wakeup))
3151                                 next_wakeup = elr->lr_next_sched;
3152                 }
3153                 mutex_unlock(&eli->li_list_mtx);
3154
3155                 try_to_freeze();
3156
3157                 cur = jiffies;
3158                 if ((time_after_eq(cur, next_wakeup)) ||
3159                     (MAX_JIFFY_OFFSET == next_wakeup)) {
3160                         cond_resched();
3161                         continue;
3162                 }
3163
3164                 schedule_timeout_interruptible(next_wakeup - cur);
3165
3166                 if (kthread_should_stop()) {
3167                         ext4_clear_request_list();
3168                         goto exit_thread;
3169                 }
3170         }
3171
3172 exit_thread:
3173         /*
3174          * It looks like the request list is empty, but we need
3175          * to check it under the li_list_mtx lock, to prevent any
3176          * additions into it, and of course we should lock ext4_li_mtx
3177          * to atomically free the list and ext4_li_info, because at
3178          * this point another ext4 filesystem could be registering
3179          * new one.
3180          */
3181         mutex_lock(&ext4_li_mtx);
3182         mutex_lock(&eli->li_list_mtx);
3183         if (!list_empty(&eli->li_request_list)) {
3184                 mutex_unlock(&eli->li_list_mtx);
3185                 mutex_unlock(&ext4_li_mtx);
3186                 goto cont_thread;
3187         }
3188         mutex_unlock(&eli->li_list_mtx);
3189         kfree(ext4_li_info);
3190         ext4_li_info = NULL;
3191         mutex_unlock(&ext4_li_mtx);
3192
3193         return 0;
3194 }
3195
3196 static void ext4_clear_request_list(void)
3197 {
3198         struct list_head *pos, *n;
3199         struct ext4_li_request *elr;
3200
3201         mutex_lock(&ext4_li_info->li_list_mtx);
3202         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3203                 elr = list_entry(pos, struct ext4_li_request,
3204                                  lr_request);
3205                 ext4_remove_li_request(elr);
3206         }
3207         mutex_unlock(&ext4_li_info->li_list_mtx);
3208 }
3209
3210 static int ext4_run_lazyinit_thread(void)
3211 {
3212         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3213                                          ext4_li_info, "ext4lazyinit");
3214         if (IS_ERR(ext4_lazyinit_task)) {
3215                 int err = PTR_ERR(ext4_lazyinit_task);
3216                 ext4_clear_request_list();
3217                 kfree(ext4_li_info);
3218                 ext4_li_info = NULL;
3219                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3220                                  "initialization thread\n",
3221                                  err);
3222                 return err;
3223         }
3224         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3225         return 0;
3226 }
3227
3228 /*
3229  * Check whether it make sense to run itable init. thread or not.
3230  * If there is at least one uninitialized inode table, return
3231  * corresponding group number, else the loop goes through all
3232  * groups and return total number of groups.
3233  */
3234 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3235 {
3236         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3237         struct ext4_group_desc *gdp = NULL;
3238
3239         if (!ext4_has_group_desc_csum(sb))
3240                 return ngroups;
3241
3242         for (group = 0; group < ngroups; group++) {
3243                 gdp = ext4_get_group_desc(sb, group, NULL);
3244                 if (!gdp)
3245                         continue;
3246
3247                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3248                         break;
3249         }
3250
3251         return group;
3252 }
3253
3254 static int ext4_li_info_new(void)
3255 {
3256         struct ext4_lazy_init *eli = NULL;
3257
3258         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3259         if (!eli)
3260                 return -ENOMEM;
3261
3262         INIT_LIST_HEAD(&eli->li_request_list);
3263         mutex_init(&eli->li_list_mtx);
3264
3265         eli->li_state |= EXT4_LAZYINIT_QUIT;
3266
3267         ext4_li_info = eli;
3268
3269         return 0;
3270 }
3271
3272 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3273                                             ext4_group_t start)
3274 {
3275         struct ext4_sb_info *sbi = EXT4_SB(sb);
3276         struct ext4_li_request *elr;
3277
3278         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3279         if (!elr)
3280                 return NULL;
3281
3282         elr->lr_super = sb;
3283         elr->lr_sbi = sbi;
3284         elr->lr_next_group = start;
3285
3286         /*
3287          * Randomize first schedule time of the request to
3288          * spread the inode table initialization requests
3289          * better.
3290          */
3291         elr->lr_next_sched = jiffies + (prandom_u32() %
3292                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3293         return elr;
3294 }
3295
3296 int ext4_register_li_request(struct super_block *sb,
3297                              ext4_group_t first_not_zeroed)
3298 {
3299         struct ext4_sb_info *sbi = EXT4_SB(sb);
3300         struct ext4_li_request *elr = NULL;
3301         ext4_group_t ngroups = sbi->s_groups_count;
3302         int ret = 0;
3303
3304         mutex_lock(&ext4_li_mtx);
3305         if (sbi->s_li_request != NULL) {
3306                 /*
3307                  * Reset timeout so it can be computed again, because
3308                  * s_li_wait_mult might have changed.
3309                  */
3310                 sbi->s_li_request->lr_timeout = 0;
3311                 goto out;
3312         }
3313
3314         if (first_not_zeroed == ngroups || sb_rdonly(sb) ||
3315             !test_opt(sb, INIT_INODE_TABLE))
3316                 goto out;
3317
3318         elr = ext4_li_request_new(sb, first_not_zeroed);
3319         if (!elr) {
3320                 ret = -ENOMEM;
3321                 goto out;
3322         }
3323
3324         if (NULL == ext4_li_info) {
3325                 ret = ext4_li_info_new();
3326                 if (ret)
3327                         goto out;
3328         }
3329
3330         mutex_lock(&ext4_li_info->li_list_mtx);
3331         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3332         mutex_unlock(&ext4_li_info->li_list_mtx);
3333
3334         sbi->s_li_request = elr;
3335         /*
3336          * set elr to NULL here since it has been inserted to
3337          * the request_list and the removal and free of it is
3338          * handled by ext4_clear_request_list from now on.
3339          */
3340         elr = NULL;
3341
3342         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3343                 ret = ext4_run_lazyinit_thread();
3344                 if (ret)
3345                         goto out;
3346         }
3347 out:
3348         mutex_unlock(&ext4_li_mtx);
3349         if (ret)
3350                 kfree(elr);
3351         return ret;
3352 }
3353
3354 /*
3355  * We do not need to lock anything since this is called on
3356  * module unload.
3357  */
3358 static void ext4_destroy_lazyinit_thread(void)
3359 {
3360         /*
3361          * If thread exited earlier
3362          * there's nothing to be done.
3363          */
3364         if (!ext4_li_info || !ext4_lazyinit_task)
3365                 return;
3366
3367         kthread_stop(ext4_lazyinit_task);
3368 }
3369
3370 static int set_journal_csum_feature_set(struct super_block *sb)
3371 {
3372         int ret = 1;
3373         int compat, incompat;
3374         struct ext4_sb_info *sbi = EXT4_SB(sb);
3375
3376         if (ext4_has_metadata_csum(sb)) {
3377                 /* journal checksum v3 */
3378                 compat = 0;
3379                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3380         } else {
3381                 /* journal checksum v1 */
3382                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3383                 incompat = 0;
3384         }
3385
3386         jbd2_journal_clear_features(sbi->s_journal,
3387                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3388                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3389                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3390         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3391                 ret = jbd2_journal_set_features(sbi->s_journal,
3392                                 compat, 0,
3393                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3394                                 incompat);
3395         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3396                 ret = jbd2_journal_set_features(sbi->s_journal,
3397                                 compat, 0,
3398                                 incompat);
3399                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3400                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3401         } else {
3402                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3403                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3404         }
3405
3406         return ret;
3407 }
3408
3409 /*
3410  * Note: calculating the overhead so we can be compatible with
3411  * historical BSD practice is quite difficult in the face of
3412  * clusters/bigalloc.  This is because multiple metadata blocks from
3413  * different block group can end up in the same allocation cluster.
3414  * Calculating the exact overhead in the face of clustered allocation
3415  * requires either O(all block bitmaps) in memory or O(number of block
3416  * groups**2) in time.  We will still calculate the superblock for
3417  * older file systems --- and if we come across with a bigalloc file
3418  * system with zero in s_overhead_clusters the estimate will be close to
3419  * correct especially for very large cluster sizes --- but for newer
3420  * file systems, it's better to calculate this figure once at mkfs
3421  * time, and store it in the superblock.  If the superblock value is
3422  * present (even for non-bigalloc file systems), we will use it.
3423  */
3424 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3425                           char *buf)
3426 {
3427         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3428         struct ext4_group_desc  *gdp;
3429         ext4_fsblk_t            first_block, last_block, b;
3430         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3431         int                     s, j, count = 0;
3432
3433         if (!ext4_has_feature_bigalloc(sb))
3434                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3435                         sbi->s_itb_per_group + 2);
3436
3437         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3438                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3439         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3440         for (i = 0; i < ngroups; i++) {
3441                 gdp = ext4_get_group_desc(sb, i, NULL);
3442                 b = ext4_block_bitmap(sb, gdp);
3443                 if (b >= first_block && b <= last_block) {
3444                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3445                         count++;
3446                 }
3447                 b = ext4_inode_bitmap(sb, gdp);
3448                 if (b >= first_block && b <= last_block) {
3449                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3450                         count++;
3451                 }
3452                 b = ext4_inode_table(sb, gdp);
3453                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3454                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3455                                 int c = EXT4_B2C(sbi, b - first_block);
3456                                 ext4_set_bit(c, buf);
3457                                 count++;
3458                         }
3459                 if (i != grp)
3460                         continue;
3461                 s = 0;
3462                 if (ext4_bg_has_super(sb, grp)) {
3463                         ext4_set_bit(s++, buf);
3464                         count++;
3465                 }
3466                 j = ext4_bg_num_gdb(sb, grp);
3467                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3468                         ext4_error(sb, "Invalid number of block group "
3469                                    "descriptor blocks: %d", j);
3470                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3471                 }
3472                 count += j;
3473                 for (; j > 0; j--)
3474                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3475         }
3476         if (!count)
3477                 return 0;
3478         return EXT4_CLUSTERS_PER_GROUP(sb) -
3479                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3480 }
3481
3482 /*
3483  * Compute the overhead and stash it in sbi->s_overhead
3484  */
3485 int ext4_calculate_overhead(struct super_block *sb)
3486 {
3487         struct ext4_sb_info *sbi = EXT4_SB(sb);
3488         struct ext4_super_block *es = sbi->s_es;
3489         struct inode *j_inode;
3490         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3491         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3492         ext4_fsblk_t overhead = 0;
3493         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3494
3495         if (!buf)
3496                 return -ENOMEM;
3497
3498         /*
3499          * Compute the overhead (FS structures).  This is constant
3500          * for a given filesystem unless the number of block groups
3501          * changes so we cache the previous value until it does.
3502          */
3503
3504         /*
3505          * All of the blocks before first_data_block are overhead
3506          */
3507         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3508
3509         /*
3510          * Add the overhead found in each block group
3511          */
3512         for (i = 0; i < ngroups; i++) {
3513                 int blks;
3514
3515                 blks = count_overhead(sb, i, buf);
3516                 overhead += blks;
3517                 if (blks)
3518                         memset(buf, 0, PAGE_SIZE);
3519                 cond_resched();
3520         }
3521
3522         /*
3523          * Add the internal journal blocks whether the journal has been
3524          * loaded or not
3525          */
3526         if (sbi->s_journal && !sbi->journal_bdev)
3527                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3528         else if (ext4_has_feature_journal(sb) && !sbi->s_journal) {
3529                 j_inode = ext4_get_journal_inode(sb, j_inum);
3530                 if (j_inode) {
3531                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3532                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
3533                         iput(j_inode);
3534                 } else {
3535                         ext4_msg(sb, KERN_ERR, "can't get journal size");
3536                 }
3537         }
3538         sbi->s_overhead = overhead;
3539         smp_wmb();
3540         free_page((unsigned long) buf);
3541         return 0;
3542 }
3543
3544 static void ext4_clamp_want_extra_isize(struct super_block *sb)
3545 {
3546         struct ext4_sb_info *sbi = EXT4_SB(sb);
3547         struct ext4_super_block *es = sbi->s_es;
3548
3549         /* determine the minimum size of new large inodes, if present */
3550         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE &&
3551             sbi->s_want_extra_isize == 0) {
3552                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3553                                                      EXT4_GOOD_OLD_INODE_SIZE;
3554                 if (ext4_has_feature_extra_isize(sb)) {
3555                         if (sbi->s_want_extra_isize <
3556                             le16_to_cpu(es->s_want_extra_isize))
3557                                 sbi->s_want_extra_isize =
3558                                         le16_to_cpu(es->s_want_extra_isize);
3559                         if (sbi->s_want_extra_isize <
3560                             le16_to_cpu(es->s_min_extra_isize))
3561                                 sbi->s_want_extra_isize =
3562                                         le16_to_cpu(es->s_min_extra_isize);
3563                 }
3564         }
3565         /* Check if enough inode space is available */
3566         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3567                                                         sbi->s_inode_size) {
3568                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3569                                                        EXT4_GOOD_OLD_INODE_SIZE;
3570                 ext4_msg(sb, KERN_INFO,
3571                          "required extra inode space not available");
3572         }
3573 }
3574
3575 static void ext4_set_resv_clusters(struct super_block *sb)
3576 {
3577         ext4_fsblk_t resv_clusters;
3578         struct ext4_sb_info *sbi = EXT4_SB(sb);
3579
3580         /*
3581          * There's no need to reserve anything when we aren't using extents.
3582          * The space estimates are exact, there are no unwritten extents,
3583          * hole punching doesn't need new metadata... This is needed especially
3584          * to keep ext2/3 backward compatibility.
3585          */
3586         if (!ext4_has_feature_extents(sb))
3587                 return;
3588         /*
3589          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3590          * This should cover the situations where we can not afford to run
3591          * out of space like for example punch hole, or converting
3592          * unwritten extents in delalloc path. In most cases such
3593          * allocation would require 1, or 2 blocks, higher numbers are
3594          * very rare.
3595          */
3596         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3597                          sbi->s_cluster_bits);
3598
3599         do_div(resv_clusters, 50);
3600         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3601
3602         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3603 }
3604
3605 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3606 {
3607         struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev);
3608         char *orig_data = kstrdup(data, GFP_KERNEL);
3609         struct buffer_head *bh;
3610         struct ext4_super_block *es = NULL;
3611         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3612         ext4_fsblk_t block;
3613         ext4_fsblk_t sb_block = get_sb_block(&data);
3614         ext4_fsblk_t logical_sb_block;
3615         unsigned long offset = 0;
3616         unsigned long journal_devnum = 0;
3617         unsigned long def_mount_opts;
3618         struct inode *root;
3619         const char *descr;
3620         int ret = -ENOMEM;
3621         int blocksize, clustersize;
3622         unsigned int db_count;
3623         unsigned int i;
3624         int needs_recovery, has_huge_files, has_bigalloc;
3625         __u64 blocks_count;
3626         int err = 0;
3627         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3628         ext4_group_t first_not_zeroed;
3629
3630         if ((data && !orig_data) || !sbi)
3631                 goto out_free_base;
3632
3633         sbi->s_daxdev = dax_dev;
3634         sbi->s_blockgroup_lock =
3635                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3636         if (!sbi->s_blockgroup_lock)
3637                 goto out_free_base;
3638
3639         sb->s_fs_info = sbi;
3640         sbi->s_sb = sb;
3641         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3642         sbi->s_sb_block = sb_block;
3643         if (sb->s_bdev->bd_part)
3644                 sbi->s_sectors_written_start =
3645                         part_stat_read(sb->s_bdev->bd_part, sectors[STAT_WRITE]);
3646
3647         /* Cleanup superblock name */
3648         strreplace(sb->s_id, '/', '!');
3649
3650         /* -EINVAL is default */
3651         ret = -EINVAL;
3652         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3653         if (!blocksize) {
3654                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3655                 goto out_fail;
3656         }
3657
3658         /*
3659          * The ext4 superblock will not be buffer aligned for other than 1kB
3660          * block sizes.  We need to calculate the offset from buffer start.
3661          */
3662         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3663                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3664                 offset = do_div(logical_sb_block, blocksize);
3665         } else {
3666                 logical_sb_block = sb_block;
3667         }
3668
3669         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3670                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3671                 goto out_fail;
3672         }
3673         /*
3674          * Note: s_es must be initialized as soon as possible because
3675          *       some ext4 macro-instructions depend on its value
3676          */
3677         es = (struct ext4_super_block *) (bh->b_data + offset);
3678         sbi->s_es = es;
3679         sb->s_magic = le16_to_cpu(es->s_magic);
3680         if (sb->s_magic != EXT4_SUPER_MAGIC)
3681                 goto cantfind_ext4;
3682         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3683
3684         /* Warn if metadata_csum and gdt_csum are both set. */
3685         if (ext4_has_feature_metadata_csum(sb) &&
3686             ext4_has_feature_gdt_csum(sb))
3687                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3688                              "redundant flags; please run fsck.");
3689
3690         /* Check for a known checksum algorithm */
3691         if (!ext4_verify_csum_type(sb, es)) {
3692                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3693                          "unknown checksum algorithm.");
3694                 silent = 1;
3695                 goto cantfind_ext4;
3696         }
3697
3698         /* Load the checksum driver */
3699         sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3700         if (IS_ERR(sbi->s_chksum_driver)) {
3701                 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3702                 ret = PTR_ERR(sbi->s_chksum_driver);
3703                 sbi->s_chksum_driver = NULL;
3704                 goto failed_mount;
3705         }
3706
3707         /* Check superblock checksum */
3708         if (!ext4_superblock_csum_verify(sb, es)) {
3709                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3710                          "invalid superblock checksum.  Run e2fsck?");
3711                 silent = 1;
3712                 ret = -EFSBADCRC;
3713                 goto cantfind_ext4;
3714         }
3715
3716         /* Precompute checksum seed for all metadata */
3717         if (ext4_has_feature_csum_seed(sb))
3718                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3719         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
3720                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3721                                                sizeof(es->s_uuid));
3722
3723         /* Set defaults before we parse the mount options */
3724         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3725         set_opt(sb, INIT_INODE_TABLE);
3726         if (def_mount_opts & EXT4_DEFM_DEBUG)
3727                 set_opt(sb, DEBUG);
3728         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3729                 set_opt(sb, GRPID);
3730         if (def_mount_opts & EXT4_DEFM_UID16)
3731                 set_opt(sb, NO_UID32);
3732         /* xattr user namespace & acls are now defaulted on */
3733         set_opt(sb, XATTR_USER);
3734 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3735         set_opt(sb, POSIX_ACL);
3736 #endif
3737         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3738         if (ext4_has_metadata_csum(sb))
3739                 set_opt(sb, JOURNAL_CHECKSUM);
3740
3741         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3742                 set_opt(sb, JOURNAL_DATA);
3743         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3744                 set_opt(sb, ORDERED_DATA);
3745         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3746                 set_opt(sb, WRITEBACK_DATA);
3747
3748         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3749                 set_opt(sb, ERRORS_PANIC);
3750         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3751                 set_opt(sb, ERRORS_CONT);
3752         else
3753                 set_opt(sb, ERRORS_RO);
3754         /* block_validity enabled by default; disable with noblock_validity */
3755         set_opt(sb, BLOCK_VALIDITY);
3756         if (def_mount_opts & EXT4_DEFM_DISCARD)
3757                 set_opt(sb, DISCARD);
3758
3759         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3760         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3761         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3762         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3763         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3764
3765         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3766                 set_opt(sb, BARRIER);
3767
3768         /*
3769          * enable delayed allocation by default
3770          * Use -o nodelalloc to turn it off
3771          */
3772         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3773             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3774                 set_opt(sb, DELALLOC);
3775
3776         /*
3777          * set default s_li_wait_mult for lazyinit, for the case there is
3778          * no mount option specified.
3779          */
3780         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3781
3782         if (sbi->s_es->s_mount_opts[0]) {
3783                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3784                                               sizeof(sbi->s_es->s_mount_opts),
3785                                               GFP_KERNEL);
3786                 if (!s_mount_opts)
3787                         goto failed_mount;
3788                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3789                                    &journal_ioprio, 0)) {
3790                         ext4_msg(sb, KERN_WARNING,
3791                                  "failed to parse options in superblock: %s",
3792                                  s_mount_opts);
3793                 }
3794                 kfree(s_mount_opts);
3795         }
3796         sbi->s_def_mount_opt = sbi->s_mount_opt;
3797         if (!parse_options((char *) data, sb, &journal_devnum,
3798                            &journal_ioprio, 0))
3799                 goto failed_mount;
3800
3801 #ifdef CONFIG_UNICODE
3802         if (ext4_has_feature_casefold(sb) && !sbi->s_encoding) {
3803                 const struct ext4_sb_encodings *encoding_info;
3804                 struct unicode_map *encoding;
3805                 __u16 encoding_flags;
3806
3807                 if (ext4_has_feature_encrypt(sb)) {
3808                         ext4_msg(sb, KERN_ERR,
3809                                  "Can't mount with encoding and encryption");
3810                         goto failed_mount;
3811                 }
3812
3813                 if (ext4_sb_read_encoding(es, &encoding_info,
3814                                           &encoding_flags)) {
3815                         ext4_msg(sb, KERN_ERR,
3816                                  "Encoding requested by superblock is unknown");
3817                         goto failed_mount;
3818                 }
3819
3820                 encoding = utf8_load(encoding_info->version);
3821                 if (IS_ERR(encoding)) {
3822                         ext4_msg(sb, KERN_ERR,
3823                                  "can't mount with superblock charset: %s-%s "
3824                                  "not supported by the kernel. flags: 0x%x.",
3825                                  encoding_info->name, encoding_info->version,
3826                                  encoding_flags);
3827                         goto failed_mount;
3828                 }
3829                 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: "
3830                          "%s-%s with flags 0x%hx", encoding_info->name,
3831                          encoding_info->version?:"\b", encoding_flags);
3832
3833                 sbi->s_encoding = encoding;
3834                 sbi->s_encoding_flags = encoding_flags;
3835         }
3836 #endif
3837
3838         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3839                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3840                             "with data=journal disables delayed "
3841                             "allocation and O_DIRECT support!\n");
3842                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3843                         ext4_msg(sb, KERN_ERR, "can't mount with "
3844                                  "both data=journal and delalloc");
3845                         goto failed_mount;
3846                 }
3847                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3848                         ext4_msg(sb, KERN_ERR, "can't mount with "
3849                                  "both data=journal and dioread_nolock");
3850                         goto failed_mount;
3851                 }
3852                 if (test_opt(sb, DAX)) {
3853                         ext4_msg(sb, KERN_ERR, "can't mount with "
3854                                  "both data=journal and dax");
3855                         goto failed_mount;
3856                 }
3857                 if (ext4_has_feature_encrypt(sb)) {
3858                         ext4_msg(sb, KERN_WARNING,
3859                                  "encrypted files will use data=ordered "
3860                                  "instead of data journaling mode");
3861                 }
3862                 if (test_opt(sb, DELALLOC))
3863                         clear_opt(sb, DELALLOC);
3864         } else {
3865                 sb->s_iflags |= SB_I_CGROUPWB;
3866         }
3867
3868         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
3869                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
3870
3871         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3872             (ext4_has_compat_features(sb) ||
3873              ext4_has_ro_compat_features(sb) ||
3874              ext4_has_incompat_features(sb)))
3875                 ext4_msg(sb, KERN_WARNING,
3876                        "feature flags set on rev 0 fs, "
3877                        "running e2fsck is recommended");
3878
3879         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3880                 set_opt2(sb, HURD_COMPAT);
3881                 if (ext4_has_feature_64bit(sb)) {
3882                         ext4_msg(sb, KERN_ERR,
3883                                  "The Hurd can't support 64-bit file systems");
3884                         goto failed_mount;
3885                 }
3886
3887                 /*
3888                  * ea_inode feature uses l_i_version field which is not
3889                  * available in HURD_COMPAT mode.
3890                  */
3891                 if (ext4_has_feature_ea_inode(sb)) {
3892                         ext4_msg(sb, KERN_ERR,
3893                                  "ea_inode feature is not supported for Hurd");
3894                         goto failed_mount;
3895                 }
3896         }
3897
3898         if (IS_EXT2_SB(sb)) {
3899                 if (ext2_feature_set_ok(sb))
3900                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3901                                  "using the ext4 subsystem");
3902                 else {
3903                         /*
3904                          * If we're probing be silent, if this looks like
3905                          * it's actually an ext[34] filesystem.
3906                          */
3907                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
3908                                 goto failed_mount;
3909                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3910                                  "to feature incompatibilities");
3911                         goto failed_mount;
3912                 }
3913         }
3914
3915         if (IS_EXT3_SB(sb)) {
3916                 if (ext3_feature_set_ok(sb))
3917                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3918                                  "using the ext4 subsystem");
3919                 else {
3920                         /*
3921                          * If we're probing be silent, if this looks like
3922                          * it's actually an ext4 filesystem.
3923                          */
3924                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
3925                                 goto failed_mount;
3926                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3927                                  "to feature incompatibilities");
3928                         goto failed_mount;
3929                 }
3930         }
3931
3932         /*
3933          * Check feature flags regardless of the revision level, since we
3934          * previously didn't change the revision level when setting the flags,
3935          * so there is a chance incompat flags are set on a rev 0 filesystem.
3936          */
3937         if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
3938                 goto failed_mount;
3939
3940         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3941         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3942             blocksize > EXT4_MAX_BLOCK_SIZE) {
3943                 ext4_msg(sb, KERN_ERR,
3944                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3945                          blocksize, le32_to_cpu(es->s_log_block_size));
3946                 goto failed_mount;
3947         }
3948         if (le32_to_cpu(es->s_log_block_size) >
3949             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3950                 ext4_msg(sb, KERN_ERR,
3951                          "Invalid log block size: %u",
3952                          le32_to_cpu(es->s_log_block_size));
3953                 goto failed_mount;
3954         }
3955         if (le32_to_cpu(es->s_log_cluster_size) >
3956             (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3957                 ext4_msg(sb, KERN_ERR,
3958                          "Invalid log cluster size: %u",
3959                          le32_to_cpu(es->s_log_cluster_size));
3960                 goto failed_mount;
3961         }
3962
3963         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
3964                 ext4_msg(sb, KERN_ERR,
3965                          "Number of reserved GDT blocks insanely large: %d",
3966                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
3967                 goto failed_mount;
3968         }
3969
3970         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3971                 if (ext4_has_feature_inline_data(sb)) {
3972                         ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
3973                                         " that may contain inline data");
3974                         goto failed_mount;
3975                 }
3976                 if (!bdev_dax_supported(sb->s_bdev, blocksize)) {
3977                         ext4_msg(sb, KERN_ERR,
3978                                 "DAX unsupported by block device.");
3979                         goto failed_mount;
3980                 }
3981         }
3982
3983         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3984                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3985                          es->s_encryption_level);
3986                 goto failed_mount;
3987         }
3988
3989         if (sb->s_blocksize != blocksize) {
3990                 /* Validate the filesystem blocksize */
3991                 if (!sb_set_blocksize(sb, blocksize)) {
3992                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3993                                         blocksize);
3994                         goto failed_mount;
3995                 }
3996
3997                 brelse(bh);
3998                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3999                 offset = do_div(logical_sb_block, blocksize);
4000                 bh = sb_bread_unmovable(sb, logical_sb_block);
4001                 if (!bh) {
4002                         ext4_msg(sb, KERN_ERR,
4003                                "Can't read superblock on 2nd try");
4004                         goto failed_mount;
4005                 }
4006                 es = (struct ext4_super_block *)(bh->b_data + offset);
4007                 sbi->s_es = es;
4008                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
4009                         ext4_msg(sb, KERN_ERR,
4010                                "Magic mismatch, very weird!");
4011                         goto failed_mount;
4012                 }
4013         }
4014
4015         has_huge_files = ext4_has_feature_huge_file(sb);
4016         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
4017                                                       has_huge_files);
4018         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
4019
4020         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
4021                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
4022                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
4023         } else {
4024                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
4025                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
4026                 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) {
4027                         ext4_msg(sb, KERN_ERR, "invalid first ino: %u",
4028                                  sbi->s_first_ino);
4029                         goto failed_mount;
4030                 }
4031                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
4032                     (!is_power_of_2(sbi->s_inode_size)) ||
4033                     (sbi->s_inode_size > blocksize)) {
4034                         ext4_msg(sb, KERN_ERR,
4035                                "unsupported inode size: %d",
4036                                sbi->s_inode_size);
4037                         goto failed_mount;
4038                 }
4039                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
4040                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
4041         }
4042
4043         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
4044         if (ext4_has_feature_64bit(sb)) {
4045                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
4046                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
4047                     !is_power_of_2(sbi->s_desc_size)) {
4048                         ext4_msg(sb, KERN_ERR,
4049                                "unsupported descriptor size %lu",
4050                                sbi->s_desc_size);
4051                         goto failed_mount;
4052                 }
4053         } else
4054                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
4055
4056         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
4057         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
4058
4059         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
4060         if (sbi->s_inodes_per_block == 0)
4061                 goto cantfind_ext4;
4062         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
4063             sbi->s_inodes_per_group > blocksize * 8) {
4064                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
4065                          sbi->s_blocks_per_group);
4066                 goto failed_mount;
4067         }
4068         sbi->s_itb_per_group = sbi->s_inodes_per_group /
4069                                         sbi->s_inodes_per_block;
4070         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
4071         sbi->s_sbh = bh;
4072         sbi->s_mount_state = le16_to_cpu(es->s_state);
4073         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
4074         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
4075
4076         for (i = 0; i < 4; i++)
4077                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
4078         sbi->s_def_hash_version = es->s_def_hash_version;
4079         if (ext4_has_feature_dir_index(sb)) {
4080                 i = le32_to_cpu(es->s_flags);
4081                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
4082                         sbi->s_hash_unsigned = 3;
4083                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
4084 #ifdef __CHAR_UNSIGNED__
4085                         if (!sb_rdonly(sb))
4086                                 es->s_flags |=
4087                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
4088                         sbi->s_hash_unsigned = 3;
4089 #else
4090                         if (!sb_rdonly(sb))
4091                                 es->s_flags |=
4092                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
4093 #endif
4094                 }
4095         }
4096
4097         /* Handle clustersize */
4098         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
4099         has_bigalloc = ext4_has_feature_bigalloc(sb);
4100         if (has_bigalloc) {
4101                 if (clustersize < blocksize) {
4102                         ext4_msg(sb, KERN_ERR,
4103                                  "cluster size (%d) smaller than "
4104                                  "block size (%d)", clustersize, blocksize);
4105                         goto failed_mount;
4106                 }
4107                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
4108                         le32_to_cpu(es->s_log_block_size);
4109                 sbi->s_clusters_per_group =
4110                         le32_to_cpu(es->s_clusters_per_group);
4111                 if (sbi->s_clusters_per_group > blocksize * 8) {
4112                         ext4_msg(sb, KERN_ERR,
4113                                  "#clusters per group too big: %lu",
4114                                  sbi->s_clusters_per_group);
4115                         goto failed_mount;
4116                 }
4117                 if (sbi->s_blocks_per_group !=
4118                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
4119                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
4120                                  "clusters per group (%lu) inconsistent",
4121                                  sbi->s_blocks_per_group,
4122                                  sbi->s_clusters_per_group);
4123                         goto failed_mount;
4124                 }
4125         } else {
4126                 if (clustersize != blocksize) {
4127                         ext4_msg(sb, KERN_ERR,
4128                                  "fragment/cluster size (%d) != "
4129                                  "block size (%d)", clustersize, blocksize);
4130                         goto failed_mount;
4131                 }
4132                 if (sbi->s_blocks_per_group > blocksize * 8) {
4133                         ext4_msg(sb, KERN_ERR,
4134                                  "#blocks per group too big: %lu",
4135                                  sbi->s_blocks_per_group);
4136                         goto failed_mount;
4137                 }
4138                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
4139                 sbi->s_cluster_bits = 0;
4140         }
4141         sbi->s_cluster_ratio = clustersize / blocksize;
4142
4143         /* Do we have standard group size of clustersize * 8 blocks ? */
4144         if (sbi->s_blocks_per_group == clustersize << 3)
4145                 set_opt2(sb, STD_GROUP_SIZE);
4146
4147         /*
4148          * Test whether we have more sectors than will fit in sector_t,
4149          * and whether the max offset is addressable by the page cache.
4150          */
4151         err = generic_check_addressable(sb->s_blocksize_bits,
4152                                         ext4_blocks_count(es));
4153         if (err) {
4154                 ext4_msg(sb, KERN_ERR, "filesystem"
4155                          " too large to mount safely on this system");
4156                 goto failed_mount;
4157         }
4158
4159         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
4160                 goto cantfind_ext4;
4161
4162         /* check blocks count against device size */
4163         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
4164         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
4165                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
4166                        "exceeds size of device (%llu blocks)",
4167                        ext4_blocks_count(es), blocks_count);
4168                 goto failed_mount;
4169         }
4170
4171         /*
4172          * It makes no sense for the first data block to be beyond the end
4173          * of the filesystem.
4174          */
4175         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
4176                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4177                          "block %u is beyond end of filesystem (%llu)",
4178                          le32_to_cpu(es->s_first_data_block),
4179                          ext4_blocks_count(es));
4180                 goto failed_mount;
4181         }
4182         if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) &&
4183             (sbi->s_cluster_ratio == 1)) {
4184                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4185                          "block is 0 with a 1k block and cluster size");
4186                 goto failed_mount;
4187         }
4188
4189         blocks_count = (ext4_blocks_count(es) -
4190                         le32_to_cpu(es->s_first_data_block) +
4191                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
4192         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
4193         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
4194                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
4195                        "(block count %llu, first data block %u, "
4196                        "blocks per group %lu)", sbi->s_groups_count,
4197                        ext4_blocks_count(es),
4198                        le32_to_cpu(es->s_first_data_block),
4199                        EXT4_BLOCKS_PER_GROUP(sb));
4200                 goto failed_mount;
4201         }
4202         sbi->s_groups_count = blocks_count;
4203         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
4204                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
4205         if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) !=
4206             le32_to_cpu(es->s_inodes_count)) {
4207                 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu",
4208                          le32_to_cpu(es->s_inodes_count),
4209                          ((u64)sbi->s_groups_count * sbi->s_inodes_per_group));
4210                 ret = -EINVAL;
4211                 goto failed_mount;
4212         }
4213         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
4214                    EXT4_DESC_PER_BLOCK(sb);
4215         if (ext4_has_feature_meta_bg(sb)) {
4216                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
4217                         ext4_msg(sb, KERN_WARNING,
4218                                  "first meta block group too large: %u "
4219                                  "(group descriptor block count %u)",
4220                                  le32_to_cpu(es->s_first_meta_bg), db_count);
4221                         goto failed_mount;
4222                 }
4223         }
4224         sbi->s_group_desc = kvmalloc_array(db_count,
4225                                            sizeof(struct buffer_head *),
4226                                            GFP_KERNEL);
4227         if (sbi->s_group_desc == NULL) {
4228                 ext4_msg(sb, KERN_ERR, "not enough memory");
4229                 ret = -ENOMEM;
4230                 goto failed_mount;
4231         }
4232
4233         bgl_lock_init(sbi->s_blockgroup_lock);
4234
4235         /* Pre-read the descriptors into the buffer cache */
4236         for (i = 0; i < db_count; i++) {
4237                 block = descriptor_loc(sb, logical_sb_block, i);
4238                 sb_breadahead(sb, block);
4239         }
4240
4241         for (i = 0; i < db_count; i++) {
4242                 block = descriptor_loc(sb, logical_sb_block, i);
4243                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
4244                 if (!sbi->s_group_desc[i]) {
4245                         ext4_msg(sb, KERN_ERR,
4246                                "can't read group descriptor %d", i);
4247                         db_count = i;
4248                         goto failed_mount2;
4249                 }
4250         }
4251         sbi->s_gdb_count = db_count;
4252         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
4253                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
4254                 ret = -EFSCORRUPTED;
4255                 goto failed_mount2;
4256         }
4257
4258         timer_setup(&sbi->s_err_report, print_daily_error_info, 0);
4259
4260         /* Register extent status tree shrinker */
4261         if (ext4_es_register_shrinker(sbi))
4262                 goto failed_mount3;
4263
4264         sbi->s_stripe = ext4_get_stripe_size(sbi);
4265         sbi->s_extent_max_zeroout_kb = 32;
4266
4267         /*
4268          * set up enough so that it can read an inode
4269          */
4270         sb->s_op = &ext4_sops;
4271         sb->s_export_op = &ext4_export_ops;
4272         sb->s_xattr = ext4_xattr_handlers;
4273 #ifdef CONFIG_FS_ENCRYPTION
4274         sb->s_cop = &ext4_cryptops;
4275 #endif
4276 #ifdef CONFIG_FS_VERITY
4277         sb->s_vop = &ext4_verityops;
4278 #endif
4279 #ifdef CONFIG_QUOTA
4280         sb->dq_op = &ext4_quota_operations;
4281         if (ext4_has_feature_quota(sb))
4282                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4283         else
4284                 sb->s_qcop = &ext4_qctl_operations;
4285         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
4286 #endif
4287         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
4288
4289         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
4290         mutex_init(&sbi->s_orphan_lock);
4291
4292         sb->s_root = NULL;
4293
4294         needs_recovery = (es->s_last_orphan != 0 ||
4295                           ext4_has_feature_journal_needs_recovery(sb));
4296
4297         if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb))
4298                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
4299                         goto failed_mount3a;
4300
4301         /*
4302          * The first inode we look at is the journal inode.  Don't try
4303          * root first: it may be modified in the journal!
4304          */
4305         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
4306                 err = ext4_load_journal(sb, es, journal_devnum);
4307                 if (err)
4308                         goto failed_mount3a;
4309         } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
4310                    ext4_has_feature_journal_needs_recovery(sb)) {
4311                 ext4_msg(sb, KERN_ERR, "required journal recovery "
4312                        "suppressed and not mounted read-only");
4313                 goto failed_mount_wq;
4314         } else {
4315                 /* Nojournal mode, all journal mount options are illegal */
4316                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
4317                         ext4_msg(sb, KERN_ERR, "can't mount with "
4318                                  "journal_checksum, fs mounted w/o journal");
4319                         goto failed_mount_wq;
4320                 }
4321                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4322                         ext4_msg(sb, KERN_ERR, "can't mount with "
4323                                  "journal_async_commit, fs mounted w/o journal");
4324                         goto failed_mount_wq;
4325                 }
4326                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
4327                         ext4_msg(sb, KERN_ERR, "can't mount with "
4328                                  "commit=%lu, fs mounted w/o journal",
4329                                  sbi->s_commit_interval / HZ);
4330                         goto failed_mount_wq;
4331                 }
4332                 if (EXT4_MOUNT_DATA_FLAGS &
4333                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
4334                         ext4_msg(sb, KERN_ERR, "can't mount with "
4335                                  "data=, fs mounted w/o journal");
4336                         goto failed_mount_wq;
4337                 }
4338                 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM;
4339                 clear_opt(sb, JOURNAL_CHECKSUM);
4340                 clear_opt(sb, DATA_FLAGS);
4341                 sbi->s_journal = NULL;
4342                 needs_recovery = 0;
4343                 goto no_journal;
4344         }
4345
4346         if (ext4_has_feature_64bit(sb) &&
4347             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4348                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4349                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4350                 goto failed_mount_wq;
4351         }
4352
4353         if (!set_journal_csum_feature_set(sb)) {
4354                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4355                          "feature set");
4356                 goto failed_mount_wq;
4357         }
4358
4359         /* We have now updated the journal if required, so we can
4360          * validate the data journaling mode. */
4361         switch (test_opt(sb, DATA_FLAGS)) {
4362         case 0:
4363                 /* No mode set, assume a default based on the journal
4364                  * capabilities: ORDERED_DATA if the journal can
4365                  * cope, else JOURNAL_DATA
4366                  */
4367                 if (jbd2_journal_check_available_features
4368                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4369                         set_opt(sb, ORDERED_DATA);
4370                         sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
4371                 } else {
4372                         set_opt(sb, JOURNAL_DATA);
4373                         sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
4374                 }
4375                 break;
4376
4377         case EXT4_MOUNT_ORDERED_DATA:
4378         case EXT4_MOUNT_WRITEBACK_DATA:
4379                 if (!jbd2_journal_check_available_features
4380                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4381                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4382                                "requested data journaling mode");
4383                         goto failed_mount_wq;
4384                 }
4385         default:
4386                 break;
4387         }
4388
4389         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4390             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4391                 ext4_msg(sb, KERN_ERR, "can't mount with "
4392                         "journal_async_commit in data=ordered mode");
4393                 goto failed_mount_wq;
4394         }
4395
4396         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4397
4398         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4399
4400 no_journal:
4401         if (!test_opt(sb, NO_MBCACHE)) {
4402                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
4403                 if (!sbi->s_ea_block_cache) {
4404                         ext4_msg(sb, KERN_ERR,
4405                                  "Failed to create ea_block_cache");
4406                         goto failed_mount_wq;
4407                 }
4408
4409                 if (ext4_has_feature_ea_inode(sb)) {
4410                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
4411                         if (!sbi->s_ea_inode_cache) {
4412                                 ext4_msg(sb, KERN_ERR,
4413                                          "Failed to create ea_inode_cache");
4414                                 goto failed_mount_wq;
4415                         }
4416                 }
4417         }
4418
4419         if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
4420             (blocksize != PAGE_SIZE)) {
4421                 ext4_msg(sb, KERN_ERR,
4422                          "Unsupported blocksize for fs encryption");
4423                 goto failed_mount_wq;
4424         }
4425
4426         if (ext4_has_feature_verity(sb) && blocksize != PAGE_SIZE) {
4427                 ext4_msg(sb, KERN_ERR, "Unsupported blocksize for fs-verity");
4428                 goto failed_mount_wq;
4429         }
4430
4431         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) &&
4432             !ext4_has_feature_encrypt(sb)) {
4433                 ext4_set_feature_encrypt(sb);
4434                 ext4_commit_super(sb, 1);
4435         }
4436
4437         /*
4438          * Get the # of file system overhead blocks from the
4439          * superblock if present.
4440          */
4441         if (es->s_overhead_clusters)
4442                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4443         else {
4444                 err = ext4_calculate_overhead(sb);
4445                 if (err)
4446                         goto failed_mount_wq;
4447         }
4448
4449         /*
4450          * The maximum number of concurrent works can be high and
4451          * concurrency isn't really necessary.  Limit it to 1.
4452          */
4453         EXT4_SB(sb)->rsv_conversion_wq =
4454                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4455         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4456                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4457                 ret = -ENOMEM;
4458                 goto failed_mount4;
4459         }
4460
4461         /*
4462          * The jbd2_journal_load will have done any necessary log recovery,
4463          * so we can safely mount the rest of the filesystem now.
4464          */
4465
4466         root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL);
4467         if (IS_ERR(root)) {
4468                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4469                 ret = PTR_ERR(root);
4470                 root = NULL;
4471                 goto failed_mount4;
4472         }
4473         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4474                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4475                 iput(root);
4476                 goto failed_mount4;
4477         }
4478
4479 #ifdef CONFIG_UNICODE
4480         if (sbi->s_encoding)
4481                 sb->s_d_op = &ext4_dentry_ops;
4482 #endif
4483
4484         sb->s_root = d_make_root(root);
4485         if (!sb->s_root) {
4486                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4487                 ret = -ENOMEM;
4488                 goto failed_mount4;
4489         }
4490
4491         ret = ext4_setup_super(sb, es, sb_rdonly(sb));
4492         if (ret == -EROFS) {
4493                 sb->s_flags |= SB_RDONLY;
4494                 ret = 0;
4495         } else if (ret)
4496                 goto failed_mount4a;
4497
4498         ext4_clamp_want_extra_isize(sb);
4499
4500         ext4_set_resv_clusters(sb);
4501
4502         err = ext4_setup_system_zone(sb);
4503         if (err) {
4504                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4505                          "zone (%d)", err);
4506                 goto failed_mount4a;
4507         }
4508
4509         ext4_ext_init(sb);
4510         err = ext4_mb_init(sb);
4511         if (err) {
4512                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4513                          err);
4514                 goto failed_mount5;
4515         }
4516
4517         block = ext4_count_free_clusters(sb);
4518         ext4_free_blocks_count_set(sbi->s_es,
4519                                    EXT4_C2B(sbi, block));
4520         ext4_superblock_csum_set(sb);
4521         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4522                                   GFP_KERNEL);
4523         if (!err) {
4524                 unsigned long freei = ext4_count_free_inodes(sb);
4525                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4526                 ext4_superblock_csum_set(sb);
4527                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4528                                           GFP_KERNEL);
4529         }
4530         if (!err)
4531                 err = percpu_counter_init(&sbi->s_dirs_counter,
4532                                           ext4_count_dirs(sb), GFP_KERNEL);
4533         if (!err)
4534                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4535                                           GFP_KERNEL);
4536         if (!err)
4537                 err = percpu_init_rwsem(&sbi->s_journal_flag_rwsem);
4538
4539         if (err) {
4540                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4541                 goto failed_mount6;
4542         }
4543
4544         if (ext4_has_feature_flex_bg(sb))
4545                 if (!ext4_fill_flex_info(sb)) {
4546                         ext4_msg(sb, KERN_ERR,
4547                                "unable to initialize "
4548                                "flex_bg meta info!");
4549                         goto failed_mount6;
4550                 }
4551
4552         err = ext4_register_li_request(sb, first_not_zeroed);
4553         if (err)
4554                 goto failed_mount6;
4555
4556         err = ext4_register_sysfs(sb);
4557         if (err)
4558                 goto failed_mount7;
4559
4560 #ifdef CONFIG_QUOTA
4561         /* Enable quota usage during mount. */
4562         if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
4563                 err = ext4_enable_quotas(sb);
4564                 if (err)
4565                         goto failed_mount8;
4566         }
4567 #endif  /* CONFIG_QUOTA */
4568
4569         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4570         ext4_orphan_cleanup(sb, es);
4571         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4572         if (needs_recovery) {
4573                 ext4_msg(sb, KERN_INFO, "recovery complete");
4574                 ext4_mark_recovery_complete(sb, es);
4575         }
4576         if (EXT4_SB(sb)->s_journal) {
4577                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4578                         descr = " journalled data mode";
4579                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4580                         descr = " ordered data mode";
4581                 else
4582                         descr = " writeback data mode";
4583         } else
4584                 descr = "out journal";
4585
4586         if (test_opt(sb, DISCARD)) {
4587                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4588                 if (!blk_queue_discard(q))
4589                         ext4_msg(sb, KERN_WARNING,
4590                                  "mounting with \"discard\" option, but "
4591                                  "the device does not support discard");
4592         }
4593
4594         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4595                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4596                          "Opts: %.*s%s%s", descr,
4597                          (int) sizeof(sbi->s_es->s_mount_opts),
4598                          sbi->s_es->s_mount_opts,
4599                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4600
4601         if (es->s_error_count)
4602                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4603
4604         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4605         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4606         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4607         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4608
4609         kfree(orig_data);
4610         return 0;
4611
4612 cantfind_ext4:
4613         if (!silent)
4614                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4615         goto failed_mount;
4616
4617 #ifdef CONFIG_QUOTA
4618 failed_mount8:
4619         ext4_unregister_sysfs(sb);
4620 #endif
4621 failed_mount7:
4622         ext4_unregister_li_request(sb);
4623 failed_mount6:
4624         ext4_mb_release(sb);
4625         if (sbi->s_flex_groups)
4626                 kvfree(sbi->s_flex_groups);
4627         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4628         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4629         percpu_counter_destroy(&sbi->s_dirs_counter);
4630         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4631         percpu_free_rwsem(&sbi->s_journal_flag_rwsem);
4632 failed_mount5:
4633         ext4_ext_release(sb);
4634         ext4_release_system_zone(sb);
4635 failed_mount4a:
4636         dput(sb->s_root);
4637         sb->s_root = NULL;
4638 failed_mount4:
4639         ext4_msg(sb, KERN_ERR, "mount failed");
4640         if (EXT4_SB(sb)->rsv_conversion_wq)
4641                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4642 failed_mount_wq:
4643         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
4644         sbi->s_ea_inode_cache = NULL;
4645
4646         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
4647         sbi->s_ea_block_cache = NULL;
4648
4649         if (sbi->s_journal) {
4650                 jbd2_journal_destroy(sbi->s_journal);
4651                 sbi->s_journal = NULL;
4652         }
4653 failed_mount3a:
4654         ext4_es_unregister_shrinker(sbi);
4655 failed_mount3:
4656         del_timer_sync(&sbi->s_err_report);
4657         if (sbi->s_mmp_tsk)
4658                 kthread_stop(sbi->s_mmp_tsk);
4659 failed_mount2:
4660         for (i = 0; i < db_count; i++)
4661                 brelse(sbi->s_group_desc[i]);
4662         kvfree(sbi->s_group_desc);
4663 failed_mount:
4664         if (sbi->s_chksum_driver)
4665                 crypto_free_shash(sbi->s_chksum_driver);
4666
4667 #ifdef CONFIG_UNICODE
4668         utf8_unload(sbi->s_encoding);
4669 #endif
4670
4671 #ifdef CONFIG_QUOTA
4672         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4673                 kfree(get_qf_name(sb, sbi, i));
4674 #endif
4675         ext4_blkdev_remove(sbi);
4676         brelse(bh);
4677 out_fail:
4678         sb->s_fs_info = NULL;
4679         kfree(sbi->s_blockgroup_lock);
4680 out_free_base:
4681         kfree(sbi);
4682         kfree(orig_data);
4683         fs_put_dax(dax_dev);
4684         return err ? err : ret;
4685 }
4686
4687 /*
4688  * Setup any per-fs journal parameters now.  We'll do this both on
4689  * initial mount, once the journal has been initialised but before we've
4690  * done any recovery; and again on any subsequent remount.
4691  */
4692 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4693 {
4694         struct ext4_sb_info *sbi = EXT4_SB(sb);
4695
4696         journal->j_commit_interval = sbi->s_commit_interval;
4697         journal->j_min_batch_time = sbi->s_min_batch_time;
4698         journal->j_max_batch_time = sbi->s_max_batch_time;
4699
4700         write_lock(&journal->j_state_lock);
4701         if (test_opt(sb, BARRIER))
4702                 journal->j_flags |= JBD2_BARRIER;
4703         else
4704                 journal->j_flags &= ~JBD2_BARRIER;
4705         if (test_opt(sb, DATA_ERR_ABORT))
4706                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4707         else
4708                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4709         write_unlock(&journal->j_state_lock);
4710 }
4711
4712 static struct inode *ext4_get_journal_inode(struct super_block *sb,
4713                                              unsigned int journal_inum)
4714 {
4715         struct inode *journal_inode;
4716
4717         /*
4718          * Test for the existence of a valid inode on disk.  Bad things
4719          * happen if we iget() an unused inode, as the subsequent iput()
4720          * will try to delete it.
4721          */
4722         journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL);
4723         if (IS_ERR(journal_inode)) {
4724                 ext4_msg(sb, KERN_ERR, "no journal found");
4725                 return NULL;
4726         }
4727         if (!journal_inode->i_nlink) {
4728                 make_bad_inode(journal_inode);
4729                 iput(journal_inode);
4730                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4731                 return NULL;
4732         }
4733
4734         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4735                   journal_inode, journal_inode->i_size);
4736         if (!S_ISREG(journal_inode->i_mode)) {
4737                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4738                 iput(journal_inode);
4739                 return NULL;
4740         }
4741         return journal_inode;
4742 }
4743
4744 static journal_t *ext4_get_journal(struct super_block *sb,
4745                                    unsigned int journal_inum)
4746 {
4747         struct inode *journal_inode;
4748         journal_t *journal;
4749
4750         BUG_ON(!ext4_has_feature_journal(sb));
4751
4752         journal_inode = ext4_get_journal_inode(sb, journal_inum);
4753         if (!journal_inode)
4754                 return NULL;
4755
4756         journal = jbd2_journal_init_inode(journal_inode);
4757         if (!journal) {
4758                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4759                 iput(journal_inode);
4760                 return NULL;
4761         }
4762         journal->j_private = sb;
4763         ext4_init_journal_params(sb, journal);
4764         return journal;
4765 }
4766
4767 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4768                                        dev_t j_dev)
4769 {
4770         struct buffer_head *bh;
4771         journal_t *journal;
4772         ext4_fsblk_t start;
4773         ext4_fsblk_t len;
4774         int hblock, blocksize;
4775         ext4_fsblk_t sb_block;
4776         unsigned long offset;
4777         struct ext4_super_block *es;
4778         struct block_device *bdev;
4779
4780         BUG_ON(!ext4_has_feature_journal(sb));
4781
4782         bdev = ext4_blkdev_get(j_dev, sb);
4783         if (bdev == NULL)
4784                 return NULL;
4785
4786         blocksize = sb->s_blocksize;
4787         hblock = bdev_logical_block_size(bdev);
4788         if (blocksize < hblock) {
4789                 ext4_msg(sb, KERN_ERR,
4790                         "blocksize too small for journal device");
4791                 goto out_bdev;
4792         }
4793
4794         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4795         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4796         set_blocksize(bdev, blocksize);
4797         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4798                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4799                        "external journal");
4800                 goto out_bdev;
4801         }
4802
4803         es = (struct ext4_super_block *) (bh->b_data + offset);
4804         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4805             !(le32_to_cpu(es->s_feature_incompat) &
4806               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4807                 ext4_msg(sb, KERN_ERR, "external journal has "
4808                                         "bad superblock");
4809                 brelse(bh);
4810                 goto out_bdev;
4811         }
4812
4813         if ((le32_to_cpu(es->s_feature_ro_compat) &
4814              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4815             es->s_checksum != ext4_superblock_csum(sb, es)) {
4816                 ext4_msg(sb, KERN_ERR, "external journal has "
4817                                        "corrupt superblock");
4818                 brelse(bh);
4819                 goto out_bdev;
4820         }
4821
4822         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4823                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4824                 brelse(bh);
4825                 goto out_bdev;
4826         }
4827
4828         len = ext4_blocks_count(es);
4829         start = sb_block + 1;
4830         brelse(bh);     /* we're done with the superblock */
4831
4832         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4833                                         start, len, blocksize);
4834         if (!journal) {
4835                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4836                 goto out_bdev;
4837         }
4838         journal->j_private = sb;
4839         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4840         wait_on_buffer(journal->j_sb_buffer);
4841         if (!buffer_uptodate(journal->j_sb_buffer)) {
4842                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4843                 goto out_journal;
4844         }
4845         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4846                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4847                                         "user (unsupported) - %d",
4848                         be32_to_cpu(journal->j_superblock->s_nr_users));
4849                 goto out_journal;
4850         }
4851         EXT4_SB(sb)->journal_bdev = bdev;
4852         ext4_init_journal_params(sb, journal);
4853         return journal;
4854
4855 out_journal:
4856         jbd2_journal_destroy(journal);
4857 out_bdev:
4858         ext4_blkdev_put(bdev);
4859         return NULL;
4860 }
4861
4862 static int ext4_load_journal(struct super_block *sb,
4863                              struct ext4_super_block *es,
4864                              unsigned long journal_devnum)
4865 {
4866         journal_t *journal;
4867         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4868         dev_t journal_dev;
4869         int err = 0;
4870         int really_read_only;
4871
4872         BUG_ON(!ext4_has_feature_journal(sb));
4873
4874         if (journal_devnum &&
4875             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4876                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4877                         "numbers have changed");
4878                 journal_dev = new_decode_dev(journal_devnum);
4879         } else
4880                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4881
4882         really_read_only = bdev_read_only(sb->s_bdev);
4883
4884         /*
4885          * Are we loading a blank journal or performing recovery after a
4886          * crash?  For recovery, we need to check in advance whether we
4887          * can get read-write access to the device.
4888          */
4889         if (ext4_has_feature_journal_needs_recovery(sb)) {
4890                 if (sb_rdonly(sb)) {
4891                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4892                                         "required on readonly filesystem");
4893                         if (really_read_only) {
4894                                 ext4_msg(sb, KERN_ERR, "write access "
4895                                         "unavailable, cannot proceed "
4896                                         "(try mounting with noload)");
4897                                 return -EROFS;
4898                         }
4899                         ext4_msg(sb, KERN_INFO, "write access will "
4900                                "be enabled during recovery");
4901                 }
4902         }
4903
4904         if (journal_inum && journal_dev) {
4905                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4906                        "and inode journals!");
4907                 return -EINVAL;
4908         }
4909
4910         if (journal_inum) {
4911                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4912                         return -EINVAL;
4913         } else {
4914                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4915                         return -EINVAL;
4916         }
4917
4918         if (!(journal->j_flags & JBD2_BARRIER))
4919                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4920
4921         if (!ext4_has_feature_journal_needs_recovery(sb))
4922                 err = jbd2_journal_wipe(journal, !really_read_only);
4923         if (!err) {
4924                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4925                 if (save)
4926                         memcpy(save, ((char *) es) +
4927                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4928                 err = jbd2_journal_load(journal);
4929                 if (save)
4930                         memcpy(((char *) es) + EXT4_S_ERR_START,
4931                                save, EXT4_S_ERR_LEN);
4932                 kfree(save);
4933         }
4934
4935         if (err) {
4936                 ext4_msg(sb, KERN_ERR, "error loading journal");
4937                 jbd2_journal_destroy(journal);
4938                 return err;
4939         }
4940
4941         EXT4_SB(sb)->s_journal = journal;
4942         ext4_clear_journal_err(sb, es);
4943
4944         if (!really_read_only && journal_devnum &&
4945             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4946                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4947
4948                 /* Make sure we flush the recovery flag to disk. */
4949                 ext4_commit_super(sb, 1);
4950         }
4951
4952         return 0;
4953 }
4954
4955 static int ext4_commit_super(struct super_block *sb, int sync)
4956 {
4957         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4958         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4959         int error = 0;
4960
4961         if (!sbh || block_device_ejected(sb))
4962                 return error;
4963
4964         /*
4965          * The superblock bh should be mapped, but it might not be if the
4966          * device was hot-removed. Not much we can do but fail the I/O.
4967          */
4968         if (!buffer_mapped(sbh))
4969                 return error;
4970
4971         /*
4972          * If the file system is mounted read-only, don't update the
4973          * superblock write time.  This avoids updating the superblock
4974          * write time when we are mounting the root file system
4975          * read/only but we need to replay the journal; at that point,
4976          * for people who are east of GMT and who make their clock
4977          * tick in localtime for Windows bug-for-bug compatibility,
4978          * the clock is set in the future, and this will cause e2fsck
4979          * to complain and force a full file system check.
4980          */
4981         if (!(sb->s_flags & SB_RDONLY))
4982                 ext4_update_tstamp(es, s_wtime);
4983         if (sb->s_bdev->bd_part)
4984                 es->s_kbytes_written =
4985                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4986                             ((part_stat_read(sb->s_bdev->bd_part,
4987                                              sectors[STAT_WRITE]) -
4988                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4989         else
4990                 es->s_kbytes_written =
4991                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4992         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4993                 ext4_free_blocks_count_set(es,
4994                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4995                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4996         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4997                 es->s_free_inodes_count =
4998                         cpu_to_le32(percpu_counter_sum_positive(
4999                                 &EXT4_SB(sb)->s_freeinodes_counter));
5000         BUFFER_TRACE(sbh, "marking dirty");
5001         ext4_superblock_csum_set(sb);
5002         if (sync)
5003                 lock_buffer(sbh);
5004         if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
5005                 /*
5006                  * Oh, dear.  A previous attempt to write the
5007                  * superblock failed.  This could happen because the
5008                  * USB device was yanked out.  Or it could happen to
5009                  * be a transient write error and maybe the block will
5010                  * be remapped.  Nothing we can do but to retry the
5011                  * write and hope for the best.
5012                  */
5013                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
5014                        "superblock detected");
5015                 clear_buffer_write_io_error(sbh);
5016                 set_buffer_uptodate(sbh);
5017         }
5018         mark_buffer_dirty(sbh);
5019         if (sync) {
5020                 unlock_buffer(sbh);
5021                 error = __sync_dirty_buffer(sbh,
5022                         REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
5023                 if (buffer_write_io_error(sbh)) {
5024                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
5025                                "superblock");
5026                         clear_buffer_write_io_error(sbh);
5027                         set_buffer_uptodate(sbh);
5028                 }
5029         }
5030         return error;
5031 }
5032
5033 /*
5034  * Have we just finished recovery?  If so, and if we are mounting (or
5035  * remounting) the filesystem readonly, then we will end up with a
5036  * consistent fs on disk.  Record that fact.
5037  */
5038 static void ext4_mark_recovery_complete(struct super_block *sb,
5039                                         struct ext4_super_block *es)
5040 {
5041         journal_t *journal = EXT4_SB(sb)->s_journal;
5042
5043         if (!ext4_has_feature_journal(sb)) {
5044                 BUG_ON(journal != NULL);
5045                 return;
5046         }
5047         jbd2_journal_lock_updates(journal);
5048         if (jbd2_journal_flush(journal) < 0)
5049                 goto out;
5050
5051         if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) {
5052                 ext4_clear_feature_journal_needs_recovery(sb);
5053                 ext4_commit_super(sb, 1);
5054         }
5055
5056 out:
5057         jbd2_journal_unlock_updates(journal);
5058 }
5059
5060 /*
5061  * If we are mounting (or read-write remounting) a filesystem whose journal
5062  * has recorded an error from a previous lifetime, move that error to the
5063  * main filesystem now.
5064  */
5065 static void ext4_clear_journal_err(struct super_block *sb,
5066                                    struct ext4_super_block *es)
5067 {
5068         journal_t *journal;
5069         int j_errno;
5070         const char *errstr;
5071
5072         BUG_ON(!ext4_has_feature_journal(sb));
5073
5074         journal = EXT4_SB(sb)->s_journal;
5075
5076         /*
5077          * Now check for any error status which may have been recorded in the
5078          * journal by a prior ext4_error() or ext4_abort()
5079          */
5080
5081         j_errno = jbd2_journal_errno(journal);
5082         if (j_errno) {
5083                 char nbuf[16];
5084
5085                 errstr = ext4_decode_error(sb, j_errno, nbuf);
5086                 ext4_warning(sb, "Filesystem error recorded "
5087                              "from previous mount: %s", errstr);
5088                 ext4_warning(sb, "Marking fs in need of filesystem check.");
5089
5090                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
5091                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
5092                 ext4_commit_super(sb, 1);
5093
5094                 jbd2_journal_clear_err(journal);
5095                 jbd2_journal_update_sb_errno(journal);
5096         }
5097 }
5098
5099 /*
5100  * Force the running and committing transactions to commit,
5101  * and wait on the commit.
5102  */
5103 int ext4_force_commit(struct super_block *sb)
5104 {
5105         journal_t *journal;
5106
5107         if (sb_rdonly(sb))
5108                 return 0;
5109
5110         journal = EXT4_SB(sb)->s_journal;
5111         return ext4_journal_force_commit(journal);
5112 }
5113
5114 static int ext4_sync_fs(struct super_block *sb, int wait)
5115 {
5116         int ret = 0;
5117         tid_t target;
5118         bool needs_barrier = false;
5119         struct ext4_sb_info *sbi = EXT4_SB(sb);
5120
5121         if (unlikely(ext4_forced_shutdown(sbi)))
5122                 return 0;
5123
5124         trace_ext4_sync_fs(sb, wait);
5125         flush_workqueue(sbi->rsv_conversion_wq);
5126         /*
5127          * Writeback quota in non-journalled quota case - journalled quota has
5128          * no dirty dquots
5129          */
5130         dquot_writeback_dquots(sb, -1);
5131         /*
5132          * Data writeback is possible w/o journal transaction, so barrier must
5133          * being sent at the end of the function. But we can skip it if
5134          * transaction_commit will do it for us.
5135          */
5136         if (sbi->s_journal) {
5137                 target = jbd2_get_latest_transaction(sbi->s_journal);
5138                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
5139                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
5140                         needs_barrier = true;
5141
5142                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
5143                         if (wait)
5144                                 ret = jbd2_log_wait_commit(sbi->s_journal,
5145                                                            target);
5146                 }
5147         } else if (wait && test_opt(sb, BARRIER))
5148                 needs_barrier = true;
5149         if (needs_barrier) {
5150                 int err;
5151                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
5152                 if (!ret)
5153                         ret = err;
5154         }
5155
5156         return ret;
5157 }
5158
5159 /*
5160  * LVM calls this function before a (read-only) snapshot is created.  This
5161  * gives us a chance to flush the journal completely and mark the fs clean.
5162  *
5163  * Note that only this function cannot bring a filesystem to be in a clean
5164  * state independently. It relies on upper layer to stop all data & metadata
5165  * modifications.
5166  */
5167 static int ext4_freeze(struct super_block *sb)
5168 {
5169         int error = 0;
5170         journal_t *journal;
5171
5172         if (sb_rdonly(sb))
5173                 return 0;
5174
5175         journal = EXT4_SB(sb)->s_journal;
5176
5177         if (journal) {
5178                 /* Now we set up the journal barrier. */
5179                 jbd2_journal_lock_updates(journal);
5180
5181                 /*
5182                  * Don't clear the needs_recovery flag if we failed to
5183                  * flush the journal.
5184                  */
5185                 error = jbd2_journal_flush(journal);
5186                 if (error < 0)
5187                         goto out;
5188
5189                 /* Journal blocked and flushed, clear needs_recovery flag. */
5190                 ext4_clear_feature_journal_needs_recovery(sb);
5191         }
5192
5193         error = ext4_commit_super(sb, 1);
5194 out:
5195         if (journal)
5196                 /* we rely on upper layer to stop further updates */
5197                 jbd2_journal_unlock_updates(journal);
5198         return error;
5199 }
5200
5201 /*
5202  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
5203  * flag here, even though the filesystem is not technically dirty yet.
5204  */
5205 static int ext4_unfreeze(struct super_block *sb)
5206 {
5207         if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
5208                 return 0;
5209
5210         if (EXT4_SB(sb)->s_journal) {
5211                 /* Reset the needs_recovery flag before the fs is unlocked. */
5212                 ext4_set_feature_journal_needs_recovery(sb);
5213         }
5214
5215         ext4_commit_super(sb, 1);
5216         return 0;
5217 }
5218
5219 /*
5220  * Structure to save mount options for ext4_remount's benefit
5221  */
5222 struct ext4_mount_options {
5223         unsigned long s_mount_opt;
5224         unsigned long s_mount_opt2;
5225         kuid_t s_resuid;
5226         kgid_t s_resgid;
5227         unsigned long s_commit_interval;
5228         u32 s_min_batch_time, s_max_batch_time;
5229 #ifdef CONFIG_QUOTA
5230         int s_jquota_fmt;
5231         char *s_qf_names[EXT4_MAXQUOTAS];
5232 #endif
5233 };
5234
5235 static int ext4_remount(struct super_block *sb, int *flags, char *data)
5236 {
5237         struct ext4_super_block *es;
5238         struct ext4_sb_info *sbi = EXT4_SB(sb);
5239         unsigned long old_sb_flags;
5240         struct ext4_mount_options old_opts;
5241         int enable_quota = 0;
5242         ext4_group_t g;
5243         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
5244         int err = 0;
5245 #ifdef CONFIG_QUOTA
5246         int i, j;
5247         char *to_free[EXT4_MAXQUOTAS];
5248 #endif
5249         char *orig_data = kstrdup(data, GFP_KERNEL);
5250
5251         if (data && !orig_data)
5252                 return -ENOMEM;
5253
5254         /* Store the original options */
5255         old_sb_flags = sb->s_flags;
5256         old_opts.s_mount_opt = sbi->s_mount_opt;
5257         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
5258         old_opts.s_resuid = sbi->s_resuid;
5259         old_opts.s_resgid = sbi->s_resgid;
5260         old_opts.s_commit_interval = sbi->s_commit_interval;
5261         old_opts.s_min_batch_time = sbi->s_min_batch_time;
5262         old_opts.s_max_batch_time = sbi->s_max_batch_time;
5263 #ifdef CONFIG_QUOTA
5264         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
5265         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5266                 if (sbi->s_qf_names[i]) {
5267                         char *qf_name = get_qf_name(sb, sbi, i);
5268
5269                         old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL);
5270                         if (!old_opts.s_qf_names[i]) {
5271                                 for (j = 0; j < i; j++)
5272                                         kfree(old_opts.s_qf_names[j]);
5273                                 kfree(orig_data);
5274                                 return -ENOMEM;
5275                         }
5276                 } else
5277                         old_opts.s_qf_names[i] = NULL;
5278 #endif
5279         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
5280                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
5281
5282         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
5283                 err = -EINVAL;
5284                 goto restore_opts;
5285         }
5286
5287         ext4_clamp_want_extra_isize(sb);
5288
5289         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
5290             test_opt(sb, JOURNAL_CHECKSUM)) {
5291                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
5292                          "during remount not supported; ignoring");
5293                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
5294         }
5295
5296         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
5297                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
5298                         ext4_msg(sb, KERN_ERR, "can't mount with "
5299                                  "both data=journal and delalloc");
5300                         err = -EINVAL;
5301                         goto restore_opts;
5302                 }
5303                 if (test_opt(sb, DIOREAD_NOLOCK)) {
5304                         ext4_msg(sb, KERN_ERR, "can't mount with "
5305                                  "both data=journal and dioread_nolock");
5306                         err = -EINVAL;
5307                         goto restore_opts;
5308                 }
5309                 if (test_opt(sb, DAX)) {
5310                         ext4_msg(sb, KERN_ERR, "can't mount with "
5311                                  "both data=journal and dax");
5312                         err = -EINVAL;
5313                         goto restore_opts;
5314                 }
5315         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
5316                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5317                         ext4_msg(sb, KERN_ERR, "can't mount with "
5318                                 "journal_async_commit in data=ordered mode");
5319                         err = -EINVAL;
5320                         goto restore_opts;
5321                 }
5322         }
5323
5324         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
5325                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
5326                 err = -EINVAL;
5327                 goto restore_opts;
5328         }
5329
5330         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
5331                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
5332                         "dax flag with busy inodes while remounting");
5333                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
5334         }
5335
5336         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
5337                 ext4_abort(sb, "Abort forced by user");
5338
5339         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
5340                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
5341
5342         es = sbi->s_es;
5343
5344         if (sbi->s_journal) {
5345                 ext4_init_journal_params(sb, sbi->s_journal);
5346                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
5347         }
5348
5349         if (*flags & SB_LAZYTIME)
5350                 sb->s_flags |= SB_LAZYTIME;
5351
5352         if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) {
5353                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
5354                         err = -EROFS;
5355                         goto restore_opts;
5356                 }
5357
5358                 if (*flags & SB_RDONLY) {
5359                         err = sync_filesystem(sb);
5360                         if (err < 0)
5361                                 goto restore_opts;
5362                         err = dquot_suspend(sb, -1);
5363                         if (err < 0)
5364                                 goto restore_opts;
5365
5366                         /*
5367                          * First of all, the unconditional stuff we have to do
5368                          * to disable replay of the journal when we next remount
5369                          */
5370                         sb->s_flags |= SB_RDONLY;
5371
5372                         /*
5373                          * OK, test if we are remounting a valid rw partition
5374                          * readonly, and if so set the rdonly flag and then
5375                          * mark the partition as valid again.
5376                          */
5377                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5378                             (sbi->s_mount_state & EXT4_VALID_FS))
5379                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5380
5381                         if (sbi->s_journal)
5382                                 ext4_mark_recovery_complete(sb, es);
5383                         if (sbi->s_mmp_tsk)
5384                                 kthread_stop(sbi->s_mmp_tsk);
5385                 } else {
5386                         /* Make sure we can mount this feature set readwrite */
5387                         if (ext4_has_feature_readonly(sb) ||
5388                             !ext4_feature_set_ok(sb, 0)) {
5389                                 err = -EROFS;
5390                                 goto restore_opts;
5391                         }
5392                         /*
5393                          * Make sure the group descriptor checksums
5394                          * are sane.  If they aren't, refuse to remount r/w.
5395                          */
5396                         for (g = 0; g < sbi->s_groups_count; g++) {
5397                                 struct ext4_group_desc *gdp =
5398                                         ext4_get_group_desc(sb, g, NULL);
5399
5400                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5401                                         ext4_msg(sb, KERN_ERR,
5402                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5403                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
5404                                                le16_to_cpu(gdp->bg_checksum));
5405                                         err = -EFSBADCRC;
5406                                         goto restore_opts;
5407                                 }
5408                         }
5409
5410                         /*
5411                          * If we have an unprocessed orphan list hanging
5412                          * around from a previously readonly bdev mount,
5413                          * require a full umount/remount for now.
5414                          */
5415                         if (es->s_last_orphan) {
5416                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5417                                        "remount RDWR because of unprocessed "
5418                                        "orphan inode list.  Please "
5419                                        "umount/remount instead");
5420                                 err = -EINVAL;
5421                                 goto restore_opts;
5422                         }
5423
5424                         /*
5425                          * Mounting a RDONLY partition read-write, so reread
5426                          * and store the current valid flag.  (It may have
5427                          * been changed by e2fsck since we originally mounted
5428                          * the partition.)
5429                          */
5430                         if (sbi->s_journal)
5431                                 ext4_clear_journal_err(sb, es);
5432                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5433
5434                         err = ext4_setup_super(sb, es, 0);
5435                         if (err)
5436                                 goto restore_opts;
5437
5438                         sb->s_flags &= ~SB_RDONLY;
5439                         if (ext4_has_feature_mmp(sb))
5440                                 if (ext4_multi_mount_protect(sb,
5441                                                 le64_to_cpu(es->s_mmp_block))) {
5442                                         err = -EROFS;
5443                                         goto restore_opts;
5444                                 }
5445                         enable_quota = 1;
5446                 }
5447         }
5448
5449         /*
5450          * Reinitialize lazy itable initialization thread based on
5451          * current settings
5452          */
5453         if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
5454                 ext4_unregister_li_request(sb);
5455         else {
5456                 ext4_group_t first_not_zeroed;
5457                 first_not_zeroed = ext4_has_uninit_itable(sb);
5458                 ext4_register_li_request(sb, first_not_zeroed);
5459         }
5460
5461         ext4_setup_system_zone(sb);
5462         if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) {
5463                 err = ext4_commit_super(sb, 1);
5464                 if (err)
5465                         goto restore_opts;
5466         }
5467
5468 #ifdef CONFIG_QUOTA
5469         /* Release old quota file names */
5470         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5471                 kfree(old_opts.s_qf_names[i]);
5472         if (enable_quota) {
5473                 if (sb_any_quota_suspended(sb))
5474                         dquot_resume(sb, -1);
5475                 else if (ext4_has_feature_quota(sb)) {
5476                         err = ext4_enable_quotas(sb);
5477                         if (err)
5478                                 goto restore_opts;
5479                 }
5480         }
5481 #endif
5482
5483         *flags = (*flags & ~SB_LAZYTIME) | (sb->s_flags & SB_LAZYTIME);
5484         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5485         kfree(orig_data);
5486         return 0;
5487
5488 restore_opts:
5489         sb->s_flags = old_sb_flags;
5490         sbi->s_mount_opt = old_opts.s_mount_opt;
5491         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5492         sbi->s_resuid = old_opts.s_resuid;
5493         sbi->s_resgid = old_opts.s_resgid;
5494         sbi->s_commit_interval = old_opts.s_commit_interval;
5495         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5496         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5497 #ifdef CONFIG_QUOTA
5498         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5499         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5500                 to_free[i] = get_qf_name(sb, sbi, i);
5501                 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]);
5502         }
5503         synchronize_rcu();
5504         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5505                 kfree(to_free[i]);
5506 #endif
5507         kfree(orig_data);
5508         return err;
5509 }
5510
5511 #ifdef CONFIG_QUOTA
5512 static int ext4_statfs_project(struct super_block *sb,
5513                                kprojid_t projid, struct kstatfs *buf)
5514 {
5515         struct kqid qid;
5516         struct dquot *dquot;
5517         u64 limit;
5518         u64 curblock;
5519
5520         qid = make_kqid_projid(projid);
5521         dquot = dqget(sb, qid);
5522         if (IS_ERR(dquot))
5523                 return PTR_ERR(dquot);
5524         spin_lock(&dquot->dq_dqb_lock);
5525
5526         limit = (dquot->dq_dqb.dqb_bsoftlimit ?
5527                  dquot->dq_dqb.dqb_bsoftlimit :
5528                  dquot->dq_dqb.dqb_bhardlimit) >> sb->s_blocksize_bits;
5529         if (limit && buf->f_blocks > limit) {
5530                 curblock = (dquot->dq_dqb.dqb_curspace +
5531                             dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits;
5532                 buf->f_blocks = limit;
5533                 buf->f_bfree = buf->f_bavail =
5534                         (buf->f_blocks > curblock) ?
5535                          (buf->f_blocks - curblock) : 0;
5536         }
5537
5538         limit = dquot->dq_dqb.dqb_isoftlimit ?
5539                 dquot->dq_dqb.dqb_isoftlimit :
5540                 dquot->dq_dqb.dqb_ihardlimit;
5541         if (limit && buf->f_files > limit) {
5542                 buf->f_files = limit;
5543                 buf->f_ffree =
5544                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
5545                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
5546         }
5547
5548         spin_unlock(&dquot->dq_dqb_lock);
5549         dqput(dquot);
5550         return 0;
5551 }
5552 #endif
5553
5554 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5555 {
5556         struct super_block *sb = dentry->d_sb;
5557         struct ext4_sb_info *sbi = EXT4_SB(sb);
5558         struct ext4_super_block *es = sbi->s_es;
5559         ext4_fsblk_t overhead = 0, resv_blocks;
5560         u64 fsid;
5561         s64 bfree;
5562         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5563
5564         if (!test_opt(sb, MINIX_DF))
5565                 overhead = sbi->s_overhead;
5566
5567         buf->f_type = EXT4_SUPER_MAGIC;
5568         buf->f_bsize = sb->s_blocksize;
5569         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5570         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5571                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5572         /* prevent underflow in case that few free space is available */
5573         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5574         buf->f_bavail = buf->f_bfree -
5575                         (ext4_r_blocks_count(es) + resv_blocks);
5576         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5577                 buf->f_bavail = 0;
5578         buf->f_files = le32_to_cpu(es->s_inodes_count);
5579         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5580         buf->f_namelen = EXT4_NAME_LEN;
5581         fsid = le64_to_cpup((void *)es->s_uuid) ^
5582                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5583         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5584         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5585
5586 #ifdef CONFIG_QUOTA
5587         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
5588             sb_has_quota_limits_enabled(sb, PRJQUOTA))
5589                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
5590 #endif
5591         return 0;
5592 }
5593
5594
5595 #ifdef CONFIG_QUOTA
5596
5597 /*
5598  * Helper functions so that transaction is started before we acquire dqio_sem
5599  * to keep correct lock ordering of transaction > dqio_sem
5600  */
5601 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5602 {
5603         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5604 }
5605
5606 static int ext4_write_dquot(struct dquot *dquot)
5607 {
5608         int ret, err;
5609         handle_t *handle;
5610         struct inode *inode;
5611
5612         inode = dquot_to_inode(dquot);
5613         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5614                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5615         if (IS_ERR(handle))
5616                 return PTR_ERR(handle);
5617         ret = dquot_commit(dquot);
5618         err = ext4_journal_stop(handle);
5619         if (!ret)
5620                 ret = err;
5621         return ret;
5622 }
5623
5624 static int ext4_acquire_dquot(struct dquot *dquot)
5625 {
5626         int ret, err;
5627         handle_t *handle;
5628
5629         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5630                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5631         if (IS_ERR(handle))
5632                 return PTR_ERR(handle);
5633         ret = dquot_acquire(dquot);
5634         err = ext4_journal_stop(handle);
5635         if (!ret)
5636                 ret = err;
5637         return ret;
5638 }
5639
5640 static int ext4_release_dquot(struct dquot *dquot)
5641 {
5642         int ret, err;
5643         handle_t *handle;
5644
5645         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5646                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5647         if (IS_ERR(handle)) {
5648                 /* Release dquot anyway to avoid endless cycle in dqput() */
5649                 dquot_release(dquot);
5650                 return PTR_ERR(handle);
5651         }
5652         ret = dquot_release(dquot);
5653         err = ext4_journal_stop(handle);
5654         if (!ret)
5655                 ret = err;
5656         return ret;
5657 }
5658
5659 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5660 {
5661         struct super_block *sb = dquot->dq_sb;
5662         struct ext4_sb_info *sbi = EXT4_SB(sb);
5663
5664         /* Are we journaling quotas? */
5665         if (ext4_has_feature_quota(sb) ||
5666             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5667                 dquot_mark_dquot_dirty(dquot);
5668                 return ext4_write_dquot(dquot);
5669         } else {
5670                 return dquot_mark_dquot_dirty(dquot);
5671         }
5672 }
5673
5674 static int ext4_write_info(struct super_block *sb, int type)
5675 {
5676         int ret, err;
5677         handle_t *handle;
5678
5679         /* Data block + inode block */
5680         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5681         if (IS_ERR(handle))
5682                 return PTR_ERR(handle);
5683         ret = dquot_commit_info(sb, type);
5684         err = ext4_journal_stop(handle);
5685         if (!ret)
5686                 ret = err;
5687         return ret;
5688 }
5689
5690 /*
5691  * Turn on quotas during mount time - we need to find
5692  * the quota file and such...
5693  */
5694 static int ext4_quota_on_mount(struct super_block *sb, int type)
5695 {
5696         return dquot_quota_on_mount(sb, get_qf_name(sb, EXT4_SB(sb), type),
5697                                         EXT4_SB(sb)->s_jquota_fmt, type);
5698 }
5699
5700 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5701 {
5702         struct ext4_inode_info *ei = EXT4_I(inode);
5703
5704         /* The first argument of lockdep_set_subclass has to be
5705          * *exactly* the same as the argument to init_rwsem() --- in
5706          * this case, in init_once() --- or lockdep gets unhappy
5707          * because the name of the lock is set using the
5708          * stringification of the argument to init_rwsem().
5709          */
5710         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5711         lockdep_set_subclass(&ei->i_data_sem, subclass);
5712 }
5713
5714 /*
5715  * Standard function to be called on quota_on
5716  */
5717 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5718                          const struct path *path)
5719 {
5720         int err;
5721
5722         if (!test_opt(sb, QUOTA))
5723                 return -EINVAL;
5724
5725         /* Quotafile not on the same filesystem? */
5726         if (path->dentry->d_sb != sb)
5727                 return -EXDEV;
5728         /* Journaling quota? */
5729         if (EXT4_SB(sb)->s_qf_names[type]) {
5730                 /* Quotafile not in fs root? */
5731                 if (path->dentry->d_parent != sb->s_root)
5732                         ext4_msg(sb, KERN_WARNING,
5733                                 "Quota file not on filesystem root. "
5734                                 "Journaled quota will not work");
5735                 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
5736         } else {
5737                 /*
5738                  * Clear the flag just in case mount options changed since
5739                  * last time.
5740                  */
5741                 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
5742         }
5743
5744         /*
5745          * When we journal data on quota file, we have to flush journal to see
5746          * all updates to the file when we bypass pagecache...
5747          */
5748         if (EXT4_SB(sb)->s_journal &&
5749             ext4_should_journal_data(d_inode(path->dentry))) {
5750                 /*
5751                  * We don't need to lock updates but journal_flush() could
5752                  * otherwise be livelocked...
5753                  */
5754                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5755                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5756                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5757                 if (err)
5758                         return err;
5759         }
5760
5761         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5762         err = dquot_quota_on(sb, type, format_id, path);
5763         if (err) {
5764                 lockdep_set_quota_inode(path->dentry->d_inode,
5765                                              I_DATA_SEM_NORMAL);
5766         } else {
5767                 struct inode *inode = d_inode(path->dentry);
5768                 handle_t *handle;
5769
5770                 /*
5771                  * Set inode flags to prevent userspace from messing with quota
5772                  * files. If this fails, we return success anyway since quotas
5773                  * are already enabled and this is not a hard failure.
5774                  */
5775                 inode_lock(inode);
5776                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5777                 if (IS_ERR(handle))
5778                         goto unlock_inode;
5779                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
5780                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
5781                                 S_NOATIME | S_IMMUTABLE);
5782                 ext4_mark_inode_dirty(handle, inode);
5783                 ext4_journal_stop(handle);
5784         unlock_inode:
5785                 inode_unlock(inode);
5786         }
5787         return err;
5788 }
5789
5790 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5791                              unsigned int flags)
5792 {
5793         int err;
5794         struct inode *qf_inode;
5795         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5796                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5797                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5798                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5799         };
5800
5801         BUG_ON(!ext4_has_feature_quota(sb));
5802
5803         if (!qf_inums[type])
5804                 return -EPERM;
5805
5806         qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL);
5807         if (IS_ERR(qf_inode)) {
5808                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5809                 return PTR_ERR(qf_inode);
5810         }
5811
5812         /* Don't account quota for quota files to avoid recursion */
5813         qf_inode->i_flags |= S_NOQUOTA;
5814         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
5815         err = dquot_enable(qf_inode, type, format_id, flags);
5816         if (err)
5817                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
5818         iput(qf_inode);
5819
5820         return err;
5821 }
5822
5823 /* Enable usage tracking for all quota types. */
5824 static int ext4_enable_quotas(struct super_block *sb)
5825 {
5826         int type, err = 0;
5827         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5828                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5829                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5830                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5831         };
5832         bool quota_mopt[EXT4_MAXQUOTAS] = {
5833                 test_opt(sb, USRQUOTA),
5834                 test_opt(sb, GRPQUOTA),
5835                 test_opt(sb, PRJQUOTA),
5836         };
5837
5838         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
5839         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5840                 if (qf_inums[type]) {
5841                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5842                                 DQUOT_USAGE_ENABLED |
5843                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
5844                         if (err) {
5845                                 ext4_warning(sb,
5846                                         "Failed to enable quota tracking "
5847                                         "(type=%d, err=%d). Please run "
5848                                         "e2fsck to fix.", type, err);
5849                                 for (type--; type >= 0; type--)
5850                                         dquot_quota_off(sb, type);
5851
5852                                 return err;
5853                         }
5854                 }
5855         }
5856         return 0;
5857 }
5858
5859 static int ext4_quota_off(struct super_block *sb, int type)
5860 {
5861         struct inode *inode = sb_dqopt(sb)->files[type];
5862         handle_t *handle;
5863         int err;
5864
5865         /* Force all delayed allocation blocks to be allocated.
5866          * Caller already holds s_umount sem */
5867         if (test_opt(sb, DELALLOC))
5868                 sync_filesystem(sb);
5869
5870         if (!inode || !igrab(inode))
5871                 goto out;
5872
5873         err = dquot_quota_off(sb, type);
5874         if (err || ext4_has_feature_quota(sb))
5875                 goto out_put;
5876
5877         inode_lock(inode);
5878         /*
5879          * Update modification times of quota files when userspace can
5880          * start looking at them. If we fail, we return success anyway since
5881          * this is not a hard failure and quotas are already disabled.
5882          */
5883         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5884         if (IS_ERR(handle))
5885                 goto out_unlock;
5886         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
5887         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
5888         inode->i_mtime = inode->i_ctime = current_time(inode);
5889         ext4_mark_inode_dirty(handle, inode);
5890         ext4_journal_stop(handle);
5891 out_unlock:
5892         inode_unlock(inode);
5893 out_put:
5894         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
5895         iput(inode);
5896         return err;
5897 out:
5898         return dquot_quota_off(sb, type);
5899 }
5900
5901 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5902  * acquiring the locks... As quota files are never truncated and quota code
5903  * itself serializes the operations (and no one else should touch the files)
5904  * we don't have to be afraid of races */
5905 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5906                                size_t len, loff_t off)
5907 {
5908         struct inode *inode = sb_dqopt(sb)->files[type];
5909         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5910         int offset = off & (sb->s_blocksize - 1);
5911         int tocopy;
5912         size_t toread;
5913         struct buffer_head *bh;
5914         loff_t i_size = i_size_read(inode);
5915
5916         if (off > i_size)
5917                 return 0;
5918         if (off+len > i_size)
5919                 len = i_size-off;
5920         toread = len;
5921         while (toread > 0) {
5922                 tocopy = sb->s_blocksize - offset < toread ?
5923                                 sb->s_blocksize - offset : toread;
5924                 bh = ext4_bread(NULL, inode, blk, 0);
5925                 if (IS_ERR(bh))
5926                         return PTR_ERR(bh);
5927                 if (!bh)        /* A hole? */
5928                         memset(data, 0, tocopy);
5929                 else
5930                         memcpy(data, bh->b_data+offset, tocopy);
5931                 brelse(bh);
5932                 offset = 0;
5933                 toread -= tocopy;
5934                 data += tocopy;
5935                 blk++;
5936         }
5937         return len;
5938 }
5939
5940 /* Write to quotafile (we know the transaction is already started and has
5941  * enough credits) */
5942 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5943                                 const char *data, size_t len, loff_t off)
5944 {
5945         struct inode *inode = sb_dqopt(sb)->files[type];
5946         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5947         int err, offset = off & (sb->s_blocksize - 1);
5948         int retries = 0;
5949         struct buffer_head *bh;
5950         handle_t *handle = journal_current_handle();
5951
5952         if (EXT4_SB(sb)->s_journal && !handle) {
5953                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5954                         " cancelled because transaction is not started",
5955                         (unsigned long long)off, (unsigned long long)len);
5956                 return -EIO;
5957         }
5958         /*
5959          * Since we account only one data block in transaction credits,
5960          * then it is impossible to cross a block boundary.
5961          */
5962         if (sb->s_blocksize - offset < len) {
5963                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5964                         " cancelled because not block aligned",
5965                         (unsigned long long)off, (unsigned long long)len);
5966                 return -EIO;
5967         }
5968
5969         do {
5970                 bh = ext4_bread(handle, inode, blk,
5971                                 EXT4_GET_BLOCKS_CREATE |
5972                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5973         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5974                  ext4_should_retry_alloc(inode->i_sb, &retries));
5975         if (IS_ERR(bh))
5976                 return PTR_ERR(bh);
5977         if (!bh)
5978                 goto out;
5979         BUFFER_TRACE(bh, "get write access");
5980         err = ext4_journal_get_write_access(handle, bh);
5981         if (err) {
5982                 brelse(bh);
5983                 return err;
5984         }
5985         lock_buffer(bh);
5986         memcpy(bh->b_data+offset, data, len);
5987         flush_dcache_page(bh->b_page);
5988         unlock_buffer(bh);
5989         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5990         brelse(bh);
5991 out:
5992         if (inode->i_size < off + len) {
5993                 i_size_write(inode, off + len);
5994                 EXT4_I(inode)->i_disksize = inode->i_size;
5995                 ext4_mark_inode_dirty(handle, inode);
5996         }
5997         return len;
5998 }
5999
6000 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid)
6001 {
6002         const struct quota_format_ops   *ops;
6003
6004         if (!sb_has_quota_loaded(sb, qid->type))
6005                 return -ESRCH;
6006         ops = sb_dqopt(sb)->ops[qid->type];
6007         if (!ops || !ops->get_next_id)
6008                 return -ENOSYS;
6009         return dquot_get_next_id(sb, qid);
6010 }
6011 #endif
6012
6013 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
6014                        const char *dev_name, void *data)
6015 {
6016         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
6017 }
6018
6019 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
6020 static inline void register_as_ext2(void)
6021 {
6022         int err = register_filesystem(&ext2_fs_type);
6023         if (err)
6024                 printk(KERN_WARNING
6025                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
6026 }
6027
6028 static inline void unregister_as_ext2(void)
6029 {
6030         unregister_filesystem(&ext2_fs_type);
6031 }
6032
6033 static inline int ext2_feature_set_ok(struct super_block *sb)
6034 {
6035         if (ext4_has_unknown_ext2_incompat_features(sb))
6036                 return 0;
6037         if (sb_rdonly(sb))
6038                 return 1;
6039         if (ext4_has_unknown_ext2_ro_compat_features(sb))
6040                 return 0;
6041         return 1;
6042 }
6043 #else
6044 static inline void register_as_ext2(void) { }
6045 static inline void unregister_as_ext2(void) { }
6046 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
6047 #endif
6048
6049 static inline void register_as_ext3(void)
6050 {
6051         int err = register_filesystem(&ext3_fs_type);
6052         if (err)
6053                 printk(KERN_WARNING
6054                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
6055 }
6056
6057 static inline void unregister_as_ext3(void)
6058 {
6059         unregister_filesystem(&ext3_fs_type);
6060 }
6061
6062 static inline int ext3_feature_set_ok(struct super_block *sb)
6063 {
6064         if (ext4_has_unknown_ext3_incompat_features(sb))
6065                 return 0;
6066         if (!ext4_has_feature_journal(sb))
6067                 return 0;
6068         if (sb_rdonly(sb))
6069                 return 1;
6070         if (ext4_has_unknown_ext3_ro_compat_features(sb))
6071                 return 0;
6072         return 1;
6073 }
6074
6075 static struct file_system_type ext4_fs_type = {
6076         .owner          = THIS_MODULE,
6077         .name           = "ext4",
6078         .mount          = ext4_mount,
6079         .kill_sb        = kill_block_super,
6080         .fs_flags       = FS_REQUIRES_DEV,
6081 };
6082 MODULE_ALIAS_FS("ext4");
6083
6084 /* Shared across all ext4 file systems */
6085 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
6086
6087 static int __init ext4_init_fs(void)
6088 {
6089         int i, err;
6090
6091         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
6092         ext4_li_info = NULL;
6093         mutex_init(&ext4_li_mtx);
6094
6095         /* Build-time check for flags consistency */
6096         ext4_check_flag_values();
6097
6098         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
6099                 init_waitqueue_head(&ext4__ioend_wq[i]);
6100
6101         err = ext4_init_es();
6102         if (err)
6103                 return err;
6104
6105         err = ext4_init_pending();
6106         if (err)
6107                 goto out7;
6108
6109         err = ext4_init_post_read_processing();
6110         if (err)
6111                 goto out6;
6112
6113         err = ext4_init_pageio();
6114         if (err)
6115                 goto out5;
6116
6117         err = ext4_init_system_zone();
6118         if (err)
6119                 goto out4;
6120
6121         err = ext4_init_sysfs();
6122         if (err)
6123                 goto out3;
6124
6125         err = ext4_init_mballoc();
6126         if (err)
6127                 goto out2;
6128         err = init_inodecache();
6129         if (err)
6130                 goto out1;
6131         register_as_ext3();
6132         register_as_ext2();
6133         err = register_filesystem(&ext4_fs_type);
6134         if (err)
6135                 goto out;
6136
6137         return 0;
6138 out:
6139         unregister_as_ext2();
6140         unregister_as_ext3();
6141         destroy_inodecache();
6142 out1:
6143         ext4_exit_mballoc();
6144 out2:
6145         ext4_exit_sysfs();
6146 out3:
6147         ext4_exit_system_zone();
6148 out4:
6149         ext4_exit_pageio();
6150 out5:
6151         ext4_exit_post_read_processing();
6152 out6:
6153         ext4_exit_pending();
6154 out7:
6155         ext4_exit_es();
6156
6157         return err;
6158 }
6159
6160 static void __exit ext4_exit_fs(void)
6161 {
6162         ext4_destroy_lazyinit_thread();
6163         unregister_as_ext2();
6164         unregister_as_ext3();
6165         unregister_filesystem(&ext4_fs_type);
6166         destroy_inodecache();
6167         ext4_exit_mballoc();
6168         ext4_exit_sysfs();
6169         ext4_exit_system_zone();
6170         ext4_exit_pageio();
6171         ext4_exit_post_read_processing();
6172         ext4_exit_es();
6173         ext4_exit_pending();
6174 }
6175
6176 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
6177 MODULE_DESCRIPTION("Fourth Extended Filesystem");
6178 MODULE_LICENSE("GPL");
6179 MODULE_SOFTDEP("pre: crc32c");
6180 module_init(ext4_init_fs)
6181 module_exit(ext4_exit_fs)