fs/ext4/super.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/ext4/super.c
   4  *
   5  * Copyright (C) 1992, 1993, 1994, 1995
   6  * Remy Card ([email protected])
   7  * Laboratoire MASI - Institut Blaise Pascal
   8  * Universite Pierre et Marie Curie (Paris VI)
   9  *
  10  *  from
  11  *
  12  *  linux/fs/minix/inode.c
  13  *
  14  *  Copyright (C) 1991, 1992  Linus Torvalds
  15  *
  16  *  Big-endian to little-endian byte-swapping/bitmaps by
  17  *        David S. Miller ([email protected]), 1995
  18  */
  19
  20 #include <linux/module.h>
  21 #include <linux/string.h>
  22 #include <linux/fs.h>
  23 #include <linux/time.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/backing-dev.h>
  29 #include <linux/parser.h>
  30 #include <linux/buffer_head.h>
  31 #include <linux/exportfs.h>
  32 #include <linux/vfs.h>
  33 #include <linux/random.h>
  34 #include <linux/mount.h>
  35 #include <linux/namei.h>
  36 #include <linux/quotaops.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/dax.h>
  42 #include <linux/cleancache.h>
  43 #include <linux/uaccess.h>
  44 #include <linux/iversion.h>
  45 #include <linux/unicode.h>
  46 #include <linux/part_stat.h>
  47 #include <linux/kthread.h>
  48 #include <linux/freezer.h>
  49
  50 #include "ext4.h"
  51 #include "ext4_extents.h"       /* Needed for trace points definition */
  52 #include "ext4_jbd2.h"
  53 #include "xattr.h"
  54 #include "acl.h"
  55 #include "mballoc.h"
  56 #include "fsmap.h"
  57
  58 #define CREATE_TRACE_POINTS
  59 #include <trace/events/ext4.h>
  60
  61 static struct ext4_lazy_init *ext4_li_info;
  62 static struct mutex ext4_li_mtx;
  63 static struct ratelimit_state ext4_mount_msg_ratelimit;
  64
  65 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  66                              unsigned long journal_devnum);
  67 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  68 static int ext4_commit_super(struct super_block *sb, int sync);
  69 static void ext4_mark_recovery_complete(struct super_block *sb,
  70                                         struct ext4_super_block *es);
  71 static void ext4_clear_journal_err(struct super_block *sb,
  72                                    struct ext4_super_block *es);
  73 static int ext4_sync_fs(struct super_block *sb, int wait);
  74 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  75 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  76 static int ext4_unfreeze(struct super_block *sb);
  77 static int ext4_freeze(struct super_block *sb);
  78 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  79                        const char *dev_name, void *data);
  80 static inline int ext2_feature_set_ok(struct super_block *sb);
  81 static inline int ext3_feature_set_ok(struct super_block *sb);
  82 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  83 static void ext4_destroy_lazyinit_thread(void);
  84 static void ext4_unregister_li_request(struct super_block *sb);
  85 static void ext4_clear_request_list(void);
  86 static struct inode *ext4_get_journal_inode(struct super_block *sb,
  87                                             unsigned int journal_inum);
  88
  89 /*
  90  * Lock ordering
  91  *
  92  * Note the difference between i_mmap_sem (EXT4_I(inode)->i_mmap_sem) and
  93  * i_mmap_rwsem (inode->i_mmap_rwsem)!
  94  *
  95  * page fault path:
  96  * mmap_lock -> 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_lock
 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_lock
 111  * sb_start_write -> i_mutex -> transaction start -> i_data_sem (rw)
 112  *
 113  * writepages:
 114  * transaction start -> page lock(s) -> i_data_sem (rw)
 115  */
 116
 117 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
 118 static struct file_system_type ext2_fs_type = {
 119         .owner          = THIS_MODULE,
 120         .name           = "ext2",
 121         .mount          = ext4_mount,
 122         .kill_sb        = kill_block_super,
 123         .fs_flags       = FS_REQUIRES_DEV,
 124 };
 125 MODULE_ALIAS_FS("ext2");
 126 MODULE_ALIAS("ext2");
 127 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
 128 #else
 129 #define IS_EXT2_SB(sb) (0)
 130 #endif
 131
 132
 133 static struct file_system_type ext3_fs_type = {
 134         .owner          = THIS_MODULE,
 135         .name           = "ext3",
 136         .mount          = ext4_mount,
 137         .kill_sb        = kill_block_super,
 138         .fs_flags       = FS_REQUIRES_DEV,
 139 };
 140 MODULE_ALIAS_FS("ext3");
 141 MODULE_ALIAS("ext3");
 142 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 143
 144 /*
 145  * This works like sb_bread() except it uses ERR_PTR for error
 146  * returns.  Currently with sb_bread it's impossible to distinguish
 147  * between ENOMEM and EIO situations (since both result in a NULL
 148  * return.
 149  */
 150 struct buffer_head *
 151 ext4_sb_bread(struct super_block *sb, sector_t block, int op_flags)
 152 {
 153         struct buffer_head *bh = sb_getblk(sb, block);
 154
 155         if (bh == NULL)
 156                 return ERR_PTR(-ENOMEM);
 157         if (ext4_buffer_uptodate(bh))
 158                 return bh;
 159         ll_rw_block(REQ_OP_READ, REQ_META | op_flags, 1, &bh);
 160         wait_on_buffer(bh);
 161         if (buffer_uptodate(bh))
 162                 return bh;
 163         put_bh(bh);
 164         return ERR_PTR(-EIO);
 165 }
 166
 167 static int ext4_verify_csum_type(struct super_block *sb,
 168                                  struct ext4_super_block *es)
 169 {
 170         if (!ext4_has_feature_metadata_csum(sb))
 171                 return 1;
 172
 173         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 174 }
 175
 176 static __le32 ext4_superblock_csum(struct super_block *sb,
 177                                    struct ext4_super_block *es)
 178 {
 179         struct ext4_sb_info *sbi = EXT4_SB(sb);
 180         int offset = offsetof(struct ext4_super_block, s_checksum);
 181         __u32 csum;
 182
 183         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 184
 185         return cpu_to_le32(csum);
 186 }
 187
 188 static int ext4_superblock_csum_verify(struct super_block *sb,
 189                                        struct ext4_super_block *es)
 190 {
 191         if (!ext4_has_metadata_csum(sb))
 192                 return 1;
 193
 194         return es->s_checksum == ext4_superblock_csum(sb, es);
 195 }
 196
 197 void ext4_superblock_csum_set(struct super_block *sb)
 198 {
 199         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 200
 201         if (!ext4_has_metadata_csum(sb))
 202                 return;
 203
 204         es->s_checksum = ext4_superblock_csum(sb, es);
 205 }
 206
 207 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 208                                struct ext4_group_desc *bg)
 209 {
 210         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 211                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 212                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 213 }
 214
 215 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 216                                struct ext4_group_desc *bg)
 217 {
 218         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 219                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 220                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 221 }
 222
 223 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 224                               struct ext4_group_desc *bg)
 225 {
 226         return le32_to_cpu(bg->bg_inode_table_lo) |
 227                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 228                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 229 }
 230
 231 __u32 ext4_free_group_clusters(struct super_block *sb,
 232                                struct ext4_group_desc *bg)
 233 {
 234         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 235                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 236                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 237 }
 238
 239 __u32 ext4_free_inodes_count(struct super_block *sb,
 240                               struct ext4_group_desc *bg)
 241 {
 242         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 243                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 244                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 245 }
 246
 247 __u32 ext4_used_dirs_count(struct super_block *sb,
 248                               struct ext4_group_desc *bg)
 249 {
 250         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 251                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 252                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 253 }
 254
 255 __u32 ext4_itable_unused_count(struct super_block *sb,
 256                               struct ext4_group_desc *bg)
 257 {
 258         return le16_to_cpu(bg->bg_itable_unused_lo) |
 259                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 260                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 261 }
 262
 263 void ext4_block_bitmap_set(struct super_block *sb,
 264                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 265 {
 266         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 267         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 268                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 269 }
 270
 271 void ext4_inode_bitmap_set(struct super_block *sb,
 272                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 273 {
 274         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 275         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 276                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 277 }
 278
 279 void ext4_inode_table_set(struct super_block *sb,
 280                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 281 {
 282         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 283         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 284                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 285 }
 286
 287 void ext4_free_group_clusters_set(struct super_block *sb,
 288                                   struct ext4_group_desc *bg, __u32 count)
 289 {
 290         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 291         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 292                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 293 }
 294
 295 void ext4_free_inodes_set(struct super_block *sb,
 296                           struct ext4_group_desc *bg, __u32 count)
 297 {
 298         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 299         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 300                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 301 }
 302
 303 void ext4_used_dirs_set(struct super_block *sb,
 304                           struct ext4_group_desc *bg, __u32 count)
 305 {
 306         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 307         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 308                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 309 }
 310
 311 void ext4_itable_unused_set(struct super_block *sb,
 312                           struct ext4_group_desc *bg, __u32 count)
 313 {
 314         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 315         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 316                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 317 }
 318
 319 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi)
 320 {
 321         time64_t now = ktime_get_real_seconds();
 322
 323         now = clamp_val(now, 0, (1ull << 40) - 1);
 324
 325         *lo = cpu_to_le32(lower_32_bits(now));
 326         *hi = upper_32_bits(now);
 327 }
 328
 329 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi)
 330 {
 331         return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo);
 332 }
 333 #define ext4_update_tstamp(es, tstamp) \
 334         __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
 335 #define ext4_get_tstamp(es, tstamp) \
 336         __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
 337
 338 static void __save_error_info(struct super_block *sb, int error,
 339                               __u32 ino, __u64 block,
 340                               const char *func, unsigned int line)
 341 {
 342         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 343         int err;
 344
 345         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 346         if (bdev_read_only(sb->s_bdev))
 347                 return;
 348         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 349         ext4_update_tstamp(es, s_last_error_time);
 350         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 351         es->s_last_error_line = cpu_to_le32(line);
 352         es->s_last_error_ino = cpu_to_le32(ino);
 353         es->s_last_error_block = cpu_to_le64(block);
 354         switch (error) {
 355         case EIO:
 356                 err = EXT4_ERR_EIO;
 357                 break;
 358         case ENOMEM:
 359                 err = EXT4_ERR_ENOMEM;
 360                 break;
 361         case EFSBADCRC:
 362                 err = EXT4_ERR_EFSBADCRC;
 363                 break;
 364         case 0:
 365         case EFSCORRUPTED:
 366                 err = EXT4_ERR_EFSCORRUPTED;
 367                 break;
 368         case ENOSPC:
 369                 err = EXT4_ERR_ENOSPC;
 370                 break;
 371         case ENOKEY:
 372                 err = EXT4_ERR_ENOKEY;
 373                 break;
 374         case EROFS:
 375                 err = EXT4_ERR_EROFS;
 376                 break;
 377         case EFBIG:
 378                 err = EXT4_ERR_EFBIG;
 379                 break;
 380         case EEXIST:
 381                 err = EXT4_ERR_EEXIST;
 382                 break;
 383         case ERANGE:
 384                 err = EXT4_ERR_ERANGE;
 385                 break;
 386         case EOVERFLOW:
 387                 err = EXT4_ERR_EOVERFLOW;
 388                 break;
 389         case EBUSY:
 390                 err = EXT4_ERR_EBUSY;
 391                 break;
 392         case ENOTDIR:
 393                 err = EXT4_ERR_ENOTDIR;
 394                 break;
 395         case ENOTEMPTY:
 396                 err = EXT4_ERR_ENOTEMPTY;
 397                 break;
 398         case ESHUTDOWN:
 399                 err = EXT4_ERR_ESHUTDOWN;
 400                 break;
 401         case EFAULT:
 402                 err = EXT4_ERR_EFAULT;
 403                 break;
 404         default:
 405                 err = EXT4_ERR_UNKNOWN;
 406         }
 407         es->s_last_error_errcode = err;
 408         if (!es->s_first_error_time) {
 409                 es->s_first_error_time = es->s_last_error_time;
 410                 es->s_first_error_time_hi = es->s_last_error_time_hi;
 411                 strncpy(es->s_first_error_func, func,
 412                         sizeof(es->s_first_error_func));
 413                 es->s_first_error_line = cpu_to_le32(line);
 414                 es->s_first_error_ino = es->s_last_error_ino;
 415                 es->s_first_error_block = es->s_last_error_block;
 416                 es->s_first_error_errcode = es->s_last_error_errcode;
 417         }
 418         /*
 419          * Start the daily error reporting function if it hasn't been
 420          * started already
 421          */
 422         if (!es->s_error_count)
 423                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 424         le32_add_cpu(&es->s_error_count, 1);
 425 }
 426
 427 static void save_error_info(struct super_block *sb, int error,
 428                             __u32 ino, __u64 block,
 429                             const char *func, unsigned int line)
 430 {
 431         __save_error_info(sb, error, ino, block, func, line);
 432         if (!bdev_read_only(sb->s_bdev))
 433                 ext4_commit_super(sb, 1);
 434 }
 435
 436 /*
 437  * The del_gendisk() function uninitializes the disk-specific data
 438  * structures, including the bdi structure, without telling anyone
 439  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 440  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 441  * This is a kludge to prevent these oops until we can put in a proper
 442  * hook in del_gendisk() to inform the VFS and file system layers.
 443  */
 444 static int block_device_ejected(struct super_block *sb)
 445 {
 446         struct inode *bd_inode = sb->s_bdev->bd_inode;
 447         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 448
 449         return bdi->dev == NULL;
 450 }
 451
 452 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 453 {
 454         struct super_block              *sb = journal->j_private;
 455         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 456         int                             error = is_journal_aborted(journal);
 457         struct ext4_journal_cb_entry    *jce;
 458
 459         BUG_ON(txn->t_state == T_FINISHED);
 460
 461         ext4_process_freed_data(sb, txn->t_tid);
 462
 463         spin_lock(&sbi->s_md_lock);
 464         while (!list_empty(&txn->t_private_list)) {
 465                 jce = list_entry(txn->t_private_list.next,
 466                                  struct ext4_journal_cb_entry, jce_list);
 467                 list_del_init(&jce->jce_list);
 468                 spin_unlock(&sbi->s_md_lock);
 469                 jce->jce_func(sb, jce, error);
 470                 spin_lock(&sbi->s_md_lock);
 471         }
 472         spin_unlock(&sbi->s_md_lock);
 473 }
 474
 475 static bool system_going_down(void)
 476 {
 477         return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF
 478                 || system_state == SYSTEM_RESTART;
 479 }
 480
 481 /* Deal with the reporting of failure conditions on a filesystem such as
 482  * inconsistencies detected or read IO failures.
 483  *
 484  * On ext2, we can store the error state of the filesystem in the
 485  * superblock.  That is not possible on ext4, because we may have other
 486  * write ordering constraints on the superblock which prevent us from
 487  * writing it out straight away; and given that the journal is about to
 488  * be aborted, we can't rely on the current, or future, transactions to
 489  * write out the superblock safely.
 490  *
 491  * We'll just use the jbd2_journal_abort() error code to record an error in
 492  * the journal instead.  On recovery, the journal will complain about
 493  * that error until we've noted it down and cleared it.
 494  */
 495
 496 static void ext4_handle_error(struct super_block *sb)
 497 {
 498         if (test_opt(sb, WARN_ON_ERROR))
 499                 WARN_ON_ONCE(1);
 500
 501         if (sb_rdonly(sb))
 502                 return;
 503
 504         if (!test_opt(sb, ERRORS_CONT)) {
 505                 journal_t *journal = EXT4_SB(sb)->s_journal;
 506
 507                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 508                 if (journal)
 509                         jbd2_journal_abort(journal, -EIO);
 510         }
 511         /*
 512          * We force ERRORS_RO behavior when system is rebooting. Otherwise we
 513          * could panic during 'reboot -f' as the underlying device got already
 514          * disabled.
 515          */
 516         if (test_opt(sb, ERRORS_RO) || system_going_down()) {
 517                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 518                 /*
 519                  * Make sure updated value of ->s_mount_flags will be visible
 520                  * before ->s_flags update
 521                  */
 522                 smp_wmb();
 523                 sb->s_flags |= SB_RDONLY;
 524         } else if (test_opt(sb, ERRORS_PANIC)) {
 525                 panic("EXT4-fs (device %s): panic forced after error\n",
 526                         sb->s_id);
 527         }
 528 }
 529
 530 #define ext4_error_ratelimit(sb)                                        \
 531                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 532                              "EXT4-fs error")
 533
 534 void __ext4_error(struct super_block *sb, const char *function,
 535                   unsigned int line, int error, __u64 block,
 536                   const char *fmt, ...)
 537 {
 538         struct va_format vaf;
 539         va_list args;
 540
 541         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 542                 return;
 543
 544         trace_ext4_error(sb, function, line);
 545         if (ext4_error_ratelimit(sb)) {
 546                 va_start(args, fmt);
 547                 vaf.fmt = fmt;
 548                 vaf.va = &args;
 549                 printk(KERN_CRIT
 550                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 551                        sb->s_id, function, line, current->comm, &vaf);
 552                 va_end(args);
 553         }
 554         save_error_info(sb, error, 0, block, function, line);
 555         ext4_handle_error(sb);
 556 }
 557
 558 void __ext4_error_inode(struct inode *inode, const char *function,
 559                         unsigned int line, ext4_fsblk_t block, int error,
 560                         const char *fmt, ...)
 561 {
 562         va_list args;
 563         struct va_format vaf;
 564
 565         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 566                 return;
 567
 568         trace_ext4_error(inode->i_sb, function, line);
 569         if (ext4_error_ratelimit(inode->i_sb)) {
 570                 va_start(args, fmt);
 571                 vaf.fmt = fmt;
 572                 vaf.va = &args;
 573                 if (block)
 574                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 575                                "inode #%lu: block %llu: comm %s: %pV\n",
 576                                inode->i_sb->s_id, function, line, inode->i_ino,
 577                                block, current->comm, &vaf);
 578                 else
 579                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 580                                "inode #%lu: comm %s: %pV\n",
 581                                inode->i_sb->s_id, function, line, inode->i_ino,
 582                                current->comm, &vaf);
 583                 va_end(args);
 584         }
 585         save_error_info(inode->i_sb, error, inode->i_ino, block,
 586                         function, line);
 587         ext4_handle_error(inode->i_sb);
 588 }
 589
 590 void __ext4_error_file(struct file *file, const char *function,
 591                        unsigned int line, ext4_fsblk_t block,
 592                        const char *fmt, ...)
 593 {
 594         va_list args;
 595         struct va_format vaf;
 596         struct inode *inode = file_inode(file);
 597         char pathname[80], *path;
 598
 599         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 600                 return;
 601
 602         trace_ext4_error(inode->i_sb, function, line);
 603         if (ext4_error_ratelimit(inode->i_sb)) {
 604                 path = file_path(file, pathname, sizeof(pathname));
 605                 if (IS_ERR(path))
 606                         path = "(unknown)";
 607                 va_start(args, fmt);
 608                 vaf.fmt = fmt;
 609                 vaf.va = &args;
 610                 if (block)
 611                         printk(KERN_CRIT
 612                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 613                                "block %llu: comm %s: path %s: %pV\n",
 614                                inode->i_sb->s_id, function, line, inode->i_ino,
 615                                block, current->comm, path, &vaf);
 616                 else
 617                         printk(KERN_CRIT
 618                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 619                                "comm %s: path %s: %pV\n",
 620                                inode->i_sb->s_id, function, line, inode->i_ino,
 621                                current->comm, path, &vaf);
 622                 va_end(args);
 623         }
 624         save_error_info(inode->i_sb, EFSCORRUPTED, inode->i_ino, block,
 625                         function, line);
 626         ext4_handle_error(inode->i_sb);
 627 }
 628
 629 const char *ext4_decode_error(struct super_block *sb, int errno,
 630                               char nbuf[16])
 631 {
 632         char *errstr = NULL;
 633
 634         switch (errno) {
 635         case -EFSCORRUPTED:
 636                 errstr = "Corrupt filesystem";
 637                 break;
 638         case -EFSBADCRC:
 639                 errstr = "Filesystem failed CRC";
 640                 break;
 641         case -EIO:
 642                 errstr = "IO failure";
 643                 break;
 644         case -ENOMEM:
 645                 errstr = "Out of memory";
 646                 break;
 647         case -EROFS:
 648                 if (!sb || (EXT4_SB(sb)->s_journal &&
 649                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 650                         errstr = "Journal has aborted";
 651                 else
 652                         errstr = "Readonly filesystem";
 653                 break;
 654         default:
 655                 /* If the caller passed in an extra buffer for unknown
 656                  * errors, textualise them now.  Else we just return
 657                  * NULL. */
 658                 if (nbuf) {
 659                         /* Check for truncated error codes... */
 660                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 661                                 errstr = nbuf;
 662                 }
 663                 break;
 664         }
 665
 666         return errstr;
 667 }
 668
 669 /* __ext4_std_error decodes expected errors from journaling functions
 670  * automatically and invokes the appropriate error response.  */
 671
 672 void __ext4_std_error(struct super_block *sb, const char *function,
 673                       unsigned int line, int errno)
 674 {
 675         char nbuf[16];
 676         const char *errstr;
 677
 678         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 679                 return;
 680
 681         /* Special case: if the error is EROFS, and we're not already
 682          * inside a transaction, then there's really no point in logging
 683          * an error. */
 684         if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb))
 685                 return;
 686
 687         if (ext4_error_ratelimit(sb)) {
 688                 errstr = ext4_decode_error(sb, errno, nbuf);
 689                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 690                        sb->s_id, function, line, errstr);
 691         }
 692
 693         save_error_info(sb, -errno, 0, 0, function, line);
 694         ext4_handle_error(sb);
 695 }
 696
 697 /*
 698  * ext4_abort is a much stronger failure handler than ext4_error.  The
 699  * abort function may be used to deal with unrecoverable failures such
 700  * as journal IO errors or ENOMEM at a critical moment in log management.
 701  *
 702  * We unconditionally force the filesystem into an ABORT|READONLY state,
 703  * unless the error response on the fs has been set to panic in which
 704  * case we take the easy way out and panic immediately.
 705  */
 706
 707 void __ext4_abort(struct super_block *sb, const char *function,
 708                   unsigned int line, int error, const char *fmt, ...)
 709 {
 710         struct va_format vaf;
 711         va_list args;
 712
 713         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 714                 return;
 715
 716         save_error_info(sb, error, 0, 0, function, line);
 717         va_start(args, fmt);
 718         vaf.fmt = fmt;
 719         vaf.va = &args;
 720         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: %pV\n",
 721                sb->s_id, function, line, &vaf);
 722         va_end(args);
 723
 724         if (sb_rdonly(sb) == 0) {
 725                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 726                 if (EXT4_SB(sb)->s_journal)
 727                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 728
 729                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 730                 /*
 731                  * Make sure updated value of ->s_mount_flags will be visible
 732                  * before ->s_flags update
 733                  */
 734                 smp_wmb();
 735                 sb->s_flags |= SB_RDONLY;
 736         }
 737         if (test_opt(sb, ERRORS_PANIC) && !system_going_down())
 738                 panic("EXT4-fs panic from previous error\n");
 739 }
 740
 741 void __ext4_msg(struct super_block *sb,
 742                 const char *prefix, const char *fmt, ...)
 743 {
 744         struct va_format vaf;
 745         va_list args;
 746
 747         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 748                 return;
 749
 750         va_start(args, fmt);
 751         vaf.fmt = fmt;
 752         vaf.va = &args;
 753         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 754         va_end(args);
 755 }
 756
 757 #define ext4_warning_ratelimit(sb)                                      \
 758                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
 759                              "EXT4-fs warning")
 760
 761 void __ext4_warning(struct super_block *sb, const char *function,
 762                     unsigned int line, const char *fmt, ...)
 763 {
 764         struct va_format vaf;
 765         va_list args;
 766
 767         if (!ext4_warning_ratelimit(sb))
 768                 return;
 769
 770         va_start(args, fmt);
 771         vaf.fmt = fmt;
 772         vaf.va = &args;
 773         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 774                sb->s_id, function, line, &vaf);
 775         va_end(args);
 776 }
 777
 778 void __ext4_warning_inode(const struct inode *inode, const char *function,
 779                           unsigned int line, const char *fmt, ...)
 780 {
 781         struct va_format vaf;
 782         va_list args;
 783
 784         if (!ext4_warning_ratelimit(inode->i_sb))
 785                 return;
 786
 787         va_start(args, fmt);
 788         vaf.fmt = fmt;
 789         vaf.va = &args;
 790         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 791                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 792                function, line, inode->i_ino, current->comm, &vaf);
 793         va_end(args);
 794 }
 795
 796 void __ext4_grp_locked_error(const char *function, unsigned int line,
 797                              struct super_block *sb, ext4_group_t grp,
 798                              unsigned long ino, ext4_fsblk_t block,
 799                              const char *fmt, ...)
 800 __releases(bitlock)
 801 __acquires(bitlock)
 802 {
 803         struct va_format vaf;
 804         va_list args;
 805
 806         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 807                 return;
 808
 809         trace_ext4_error(sb, function, line);
 810         __save_error_info(sb, EFSCORRUPTED, ino, block, function, line);
 811
 812         if (ext4_error_ratelimit(sb)) {
 813                 va_start(args, fmt);
 814                 vaf.fmt = fmt;
 815                 vaf.va = &args;
 816                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 817                        sb->s_id, function, line, grp);
 818                 if (ino)
 819                         printk(KERN_CONT "inode %lu: ", ino);
 820                 if (block)
 821                         printk(KERN_CONT "block %llu:",
 822                                (unsigned long long) block);
 823                 printk(KERN_CONT "%pV\n", &vaf);
 824                 va_end(args);
 825         }
 826
 827         if (test_opt(sb, WARN_ON_ERROR))
 828                 WARN_ON_ONCE(1);
 829
 830         if (test_opt(sb, ERRORS_CONT)) {
 831                 ext4_commit_super(sb, 0);
 832                 return;
 833         }
 834
 835         ext4_unlock_group(sb, grp);
 836         ext4_commit_super(sb, 1);
 837         ext4_handle_error(sb);
 838         /*
 839          * We only get here in the ERRORS_RO case; relocking the group
 840          * may be dangerous, but nothing bad will happen since the
 841          * filesystem will have already been marked read/only and the
 842          * journal has been aborted.  We return 1 as a hint to callers
 843          * who might what to use the return value from
 844          * ext4_grp_locked_error() to distinguish between the
 845          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 846          * aggressively from the ext4 function in question, with a
 847          * more appropriate error code.
 848          */
 849         ext4_lock_group(sb, grp);
 850         return;
 851 }
 852
 853 void ext4_mark_group_bitmap_corrupted(struct super_block *sb,
 854                                      ext4_group_t group,
 855                                      unsigned int flags)
 856 {
 857         struct ext4_sb_info *sbi = EXT4_SB(sb);
 858         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
 859         struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
 860         int ret;
 861
 862         if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) {
 863                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT,
 864                                             &grp->bb_state);
 865                 if (!ret)
 866                         percpu_counter_sub(&sbi->s_freeclusters_counter,
 867                                            grp->bb_free);
 868         }
 869
 870         if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) {
 871                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT,
 872                                             &grp->bb_state);
 873                 if (!ret && gdp) {
 874                         int count;
 875
 876                         count = ext4_free_inodes_count(sb, gdp);
 877                         percpu_counter_sub(&sbi->s_freeinodes_counter,
 878                                            count);
 879                 }
 880         }
 881 }
 882
 883 void ext4_update_dynamic_rev(struct super_block *sb)
 884 {
 885         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 886
 887         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 888                 return;
 889
 890         ext4_warning(sb,
 891                      "updating to rev %d because of new feature flag, "
 892                      "running e2fsck is recommended",
 893                      EXT4_DYNAMIC_REV);
 894
 895         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 896         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 897         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 898         /* leave es->s_feature_*compat flags alone */
 899         /* es->s_uuid will be set by e2fsck if empty */
 900
 901         /*
 902          * The rest of the superblock fields should be zero, and if not it
 903          * means they are likely already in use, so leave them alone.  We
 904          * can leave it up to e2fsck to clean up any inconsistencies there.
 905          */
 906 }
 907
 908 /*
 909  * Open the external journal device
 910  */
 911 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 912 {
 913         struct block_device *bdev;
 914
 915         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 916         if (IS_ERR(bdev))
 917                 goto fail;
 918         return bdev;
 919
 920 fail:
 921         ext4_msg(sb, KERN_ERR,
 922                  "failed to open journal device unknown-block(%u,%u) %ld",
 923                  MAJOR(dev), MINOR(dev), PTR_ERR(bdev));
 924         return NULL;
 925 }
 926
 927 /*
 928  * Release the journal device
 929  */
 930 static void ext4_blkdev_put(struct block_device *bdev)
 931 {
 932         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 933 }
 934
 935 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 936 {
 937         struct block_device *bdev;
 938         bdev = sbi->journal_bdev;
 939         if (bdev) {
 940                 ext4_blkdev_put(bdev);
 941                 sbi->journal_bdev = NULL;
 942         }
 943 }
 944
 945 static inline struct inode *orphan_list_entry(struct list_head *l)
 946 {
 947         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 948 }
 949
 950 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 951 {
 952         struct list_head *l;
 953
 954         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 955                  le32_to_cpu(sbi->s_es->s_last_orphan));
 956
 957         printk(KERN_ERR "sb_info orphan list:\n");
 958         list_for_each(l, &sbi->s_orphan) {
 959                 struct inode *inode = orphan_list_entry(l);
 960                 printk(KERN_ERR "  "
 961                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 962                        inode->i_sb->s_id, inode->i_ino, inode,
 963                        inode->i_mode, inode->i_nlink,
 964                        NEXT_ORPHAN(inode));
 965         }
 966 }
 967
 968 #ifdef CONFIG_QUOTA
 969 static int ext4_quota_off(struct super_block *sb, int type);
 970
 971 static inline void ext4_quota_off_umount(struct super_block *sb)
 972 {
 973         int type;
 974
 975         /* Use our quota_off function to clear inode flags etc. */
 976         for (type = 0; type < EXT4_MAXQUOTAS; type++)
 977                 ext4_quota_off(sb, type);
 978 }
 979
 980 /*
 981  * This is a helper function which is used in the mount/remount
 982  * codepaths (which holds s_umount) to fetch the quota file name.
 983  */
 984 static inline char *get_qf_name(struct super_block *sb,
 985                                 struct ext4_sb_info *sbi,
 986                                 int type)
 987 {
 988         return rcu_dereference_protected(sbi->s_qf_names[type],
 989                                          lockdep_is_held(&sb->s_umount));
 990 }
 991 #else
 992 static inline void ext4_quota_off_umount(struct super_block *sb)
 993 {
 994 }
 995 #endif
 996
 997 static void ext4_put_super(struct super_block *sb)
 998 {
 999         struct ext4_sb_info *sbi = EXT4_SB(sb);
1000         struct ext4_super_block *es = sbi->s_es;
1001         struct buffer_head **group_desc;
1002         struct flex_groups **flex_groups;
1003         int aborted = 0;
1004         int i, err;
1005
1006         ext4_unregister_li_request(sb);
1007         ext4_quota_off_umount(sb);
1008
1009         destroy_workqueue(sbi->rsv_conversion_wq);
1010
1011         /*
1012          * Unregister sysfs before destroying jbd2 journal.
1013          * Since we could still access attr_journal_task attribute via sysfs
1014          * path which could have sbi->s_journal->j_task as NULL
1015          */
1016         ext4_unregister_sysfs(sb);
1017
1018         if (sbi->s_journal) {
1019                 aborted = is_journal_aborted(sbi->s_journal);
1020                 err = jbd2_journal_destroy(sbi->s_journal);
1021                 sbi->s_journal = NULL;
1022                 if ((err < 0) && !aborted) {
1023                         ext4_abort(sb, -err, "Couldn't clean up the journal");
1024                 }
1025         }
1026
1027         ext4_es_unregister_shrinker(sbi);
1028         del_timer_sync(&sbi->s_err_report);
1029         ext4_release_system_zone(sb);
1030         ext4_mb_release(sb);
1031         ext4_ext_release(sb);
1032
1033         if (!sb_rdonly(sb) && !aborted) {
1034                 ext4_clear_feature_journal_needs_recovery(sb);
1035                 es->s_state = cpu_to_le16(sbi->s_mount_state);
1036         }
1037         if (!sb_rdonly(sb))
1038                 ext4_commit_super(sb, 1);
1039
1040         rcu_read_lock();
1041         group_desc = rcu_dereference(sbi->s_group_desc);
1042         for (i = 0; i < sbi->s_gdb_count; i++)
1043                 brelse(group_desc[i]);
1044         kvfree(group_desc);
1045         flex_groups = rcu_dereference(sbi->s_flex_groups);
1046         if (flex_groups) {
1047                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
1048                         kvfree(flex_groups[i]);
1049                 kvfree(flex_groups);
1050         }
1051         rcu_read_unlock();
1052         percpu_counter_destroy(&sbi->s_freeclusters_counter);
1053         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1054         percpu_counter_destroy(&sbi->s_dirs_counter);
1055         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
1056         percpu_free_rwsem(&sbi->s_writepages_rwsem);
1057 #ifdef CONFIG_QUOTA
1058         for (i = 0; i < EXT4_MAXQUOTAS; i++)
1059                 kfree(get_qf_name(sb, sbi, i));
1060 #endif
1061
1062         /* Debugging code just in case the in-memory inode orphan list
1063          * isn't empty.  The on-disk one can be non-empty if we've
1064          * detected an error and taken the fs readonly, but the
1065          * in-memory list had better be clean by this point. */
1066         if (!list_empty(&sbi->s_orphan))
1067                 dump_orphan_list(sb, sbi);
1068         J_ASSERT(list_empty(&sbi->s_orphan));
1069
1070         sync_blockdev(sb->s_bdev);
1071         invalidate_bdev(sb->s_bdev);
1072         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
1073                 /*
1074                  * Invalidate the journal device's buffers.  We don't want them
1075                  * floating about in memory - the physical journal device may
1076                  * hotswapped, and it breaks the `ro-after' testing code.
1077                  */
1078                 sync_blockdev(sbi->journal_bdev);
1079                 invalidate_bdev(sbi->journal_bdev);
1080                 ext4_blkdev_remove(sbi);
1081         }
1082
1083         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
1084         sbi->s_ea_inode_cache = NULL;
1085
1086         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
1087         sbi->s_ea_block_cache = NULL;
1088
1089         if (sbi->s_mmp_tsk)
1090                 kthread_stop(sbi->s_mmp_tsk);
1091         brelse(sbi->s_sbh);
1092         sb->s_fs_info = NULL;
1093         /*
1094          * Now that we are completely done shutting down the
1095          * superblock, we need to actually destroy the kobject.
1096          */
1097         kobject_put(&sbi->s_kobj);
1098         wait_for_completion(&sbi->s_kobj_unregister);
1099         if (sbi->s_chksum_driver)
1100                 crypto_free_shash(sbi->s_chksum_driver);
1101         kfree(sbi->s_blockgroup_lock);
1102         fs_put_dax(sbi->s_daxdev);
1103         fscrypt_free_dummy_context(&sbi->s_dummy_enc_ctx);
1104 #ifdef CONFIG_UNICODE
1105         utf8_unload(sbi->s_encoding);
1106 #endif
1107         kfree(sbi);
1108 }
1109
1110 static struct kmem_cache *ext4_inode_cachep;
1111
1112 /*
1113  * Called inside transaction, so use GFP_NOFS
1114  */
1115 static struct inode *ext4_alloc_inode(struct super_block *sb)
1116 {
1117         struct ext4_inode_info *ei;
1118
1119         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
1120         if (!ei)
1121                 return NULL;
1122
1123         inode_set_iversion(&ei->vfs_inode, 1);
1124         spin_lock_init(&ei->i_raw_lock);
1125         INIT_LIST_HEAD(&ei->i_prealloc_list);
1126         spin_lock_init(&ei->i_prealloc_lock);
1127         ext4_es_init_tree(&ei->i_es_tree);
1128         rwlock_init(&ei->i_es_lock);
1129         INIT_LIST_HEAD(&ei->i_es_list);
1130         ei->i_es_all_nr = 0;
1131         ei->i_es_shk_nr = 0;
1132         ei->i_es_shrink_lblk = 0;
1133         ei->i_reserved_data_blocks = 0;
1134         spin_lock_init(&(ei->i_block_reservation_lock));
1135         ext4_init_pending_tree(&ei->i_pending_tree);
1136 #ifdef CONFIG_QUOTA
1137         ei->i_reserved_quota = 0;
1138         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
1139 #endif
1140         ei->jinode = NULL;
1141         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
1142         spin_lock_init(&ei->i_completed_io_lock);
1143         ei->i_sync_tid = 0;
1144         ei->i_datasync_tid = 0;
1145         atomic_set(&ei->i_unwritten, 0);
1146         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
1147         return &ei->vfs_inode;
1148 }
1149
1150 static int ext4_drop_inode(struct inode *inode)
1151 {
1152         int drop = generic_drop_inode(inode);
1153
1154         if (!drop)
1155                 drop = fscrypt_drop_inode(inode);
1156
1157         trace_ext4_drop_inode(inode, drop);
1158         return drop;
1159 }
1160
1161 static void ext4_free_in_core_inode(struct inode *inode)
1162 {
1163         fscrypt_free_inode(inode);
1164         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1165 }
1166
1167 static void ext4_destroy_inode(struct inode *inode)
1168 {
1169         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1170                 ext4_msg(inode->i_sb, KERN_ERR,
1171                          "Inode %lu (%p): orphan list check failed!",
1172                          inode->i_ino, EXT4_I(inode));
1173                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1174                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1175                                 true);
1176                 dump_stack();
1177         }
1178 }
1179
1180 static void init_once(void *foo)
1181 {
1182         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1183
1184         INIT_LIST_HEAD(&ei->i_orphan);
1185         init_rwsem(&ei->xattr_sem);
1186         init_rwsem(&ei->i_data_sem);
1187         init_rwsem(&ei->i_mmap_sem);
1188         inode_init_once(&ei->vfs_inode);
1189 }
1190
1191 static int __init init_inodecache(void)
1192 {
1193         ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache",
1194                                 sizeof(struct ext4_inode_info), 0,
1195                                 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
1196                                         SLAB_ACCOUNT),
1197                                 offsetof(struct ext4_inode_info, i_data),
1198                                 sizeof_field(struct ext4_inode_info, i_data),
1199                                 init_once);
1200         if (ext4_inode_cachep == NULL)
1201                 return -ENOMEM;
1202         return 0;
1203 }
1204
1205 static void destroy_inodecache(void)
1206 {
1207         /*
1208          * Make sure all delayed rcu free inodes are flushed before we
1209          * destroy cache.
1210          */
1211         rcu_barrier();
1212         kmem_cache_destroy(ext4_inode_cachep);
1213 }
1214
1215 void ext4_clear_inode(struct inode *inode)
1216 {
1217         invalidate_inode_buffers(inode);
1218         clear_inode(inode);
1219         ext4_discard_preallocations(inode);
1220         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1221         dquot_drop(inode);
1222         if (EXT4_I(inode)->jinode) {
1223                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1224                                                EXT4_I(inode)->jinode);
1225                 jbd2_free_inode(EXT4_I(inode)->jinode);
1226                 EXT4_I(inode)->jinode = NULL;
1227         }
1228         fscrypt_put_encryption_info(inode);
1229         fsverity_cleanup_inode(inode);
1230 }
1231
1232 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1233                                         u64 ino, u32 generation)
1234 {
1235         struct inode *inode;
1236
1237         /*
1238          * Currently we don't know the generation for parent directory, so
1239          * a generation of 0 means "accept any"
1240          */
1241         inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE);
1242         if (IS_ERR(inode))
1243                 return ERR_CAST(inode);
1244         if (generation && inode->i_generation != generation) {
1245                 iput(inode);
1246                 return ERR_PTR(-ESTALE);
1247         }
1248
1249         return inode;
1250 }
1251
1252 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1253                                         int fh_len, int fh_type)
1254 {
1255         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1256                                     ext4_nfs_get_inode);
1257 }
1258
1259 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1260                                         int fh_len, int fh_type)
1261 {
1262         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1263                                     ext4_nfs_get_inode);
1264 }
1265
1266 static int ext4_nfs_commit_metadata(struct inode *inode)
1267 {
1268         struct writeback_control wbc = {
1269                 .sync_mode = WB_SYNC_ALL
1270         };
1271
1272         trace_ext4_nfs_commit_metadata(inode);
1273         return ext4_write_inode(inode, &wbc);
1274 }
1275
1276 /*
1277  * Try to release metadata pages (indirect blocks, directories) which are
1278  * mapped via the block device.  Since these pages could have journal heads
1279  * which would prevent try_to_free_buffers() from freeing them, we must use
1280  * jbd2 layer's try_to_free_buffers() function to release them.
1281  */
1282 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1283                                  gfp_t wait)
1284 {
1285         journal_t *journal = EXT4_SB(sb)->s_journal;
1286
1287         WARN_ON(PageChecked(page));
1288         if (!page_has_buffers(page))
1289                 return 0;
1290         if (journal)
1291                 return jbd2_journal_try_to_free_buffers(journal, page,
1292                                                 wait & ~__GFP_DIRECT_RECLAIM);
1293         return try_to_free_buffers(page);
1294 }
1295
1296 #ifdef CONFIG_FS_ENCRYPTION
1297 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1298 {
1299         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1300                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1301 }
1302
1303 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1304                                                         void *fs_data)
1305 {
1306         handle_t *handle = fs_data;
1307         int res, res2, credits, retries = 0;
1308
1309         /*
1310          * Encrypting the root directory is not allowed because e2fsck expects
1311          * lost+found to exist and be unencrypted, and encrypting the root
1312          * directory would imply encrypting the lost+found directory as well as
1313          * the filename "lost+found" itself.
1314          */
1315         if (inode->i_ino == EXT4_ROOT_INO)
1316                 return -EPERM;
1317
1318         if (WARN_ON_ONCE(IS_DAX(inode) && i_size_read(inode)))
1319                 return -EINVAL;
1320
1321         if (ext4_test_inode_flag(inode, EXT4_INODE_DAX))
1322                 return -EOPNOTSUPP;
1323
1324         res = ext4_convert_inline_data(inode);
1325         if (res)
1326                 return res;
1327
1328         /*
1329          * If a journal handle was specified, then the encryption context is
1330          * being set on a new inode via inheritance and is part of a larger
1331          * transaction to create the inode.  Otherwise the encryption context is
1332          * being set on an existing inode in its own transaction.  Only in the
1333          * latter case should the "retry on ENOSPC" logic be used.
1334          */
1335
1336         if (handle) {
1337                 res = ext4_xattr_set_handle(handle, inode,
1338                                             EXT4_XATTR_INDEX_ENCRYPTION,
1339                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1340                                             ctx, len, 0);
1341                 if (!res) {
1342                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1343                         ext4_clear_inode_state(inode,
1344                                         EXT4_STATE_MAY_INLINE_DATA);
1345                         /*
1346                          * Update inode->i_flags - S_ENCRYPTED will be enabled,
1347                          * S_DAX may be disabled
1348                          */
1349                         ext4_set_inode_flags(inode, false);
1350                 }
1351                 return res;
1352         }
1353
1354         res = dquot_initialize(inode);
1355         if (res)
1356                 return res;
1357 retry:
1358         res = ext4_xattr_set_credits(inode, len, false /* is_create */,
1359                                      &credits);
1360         if (res)
1361                 return res;
1362
1363         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
1364         if (IS_ERR(handle))
1365                 return PTR_ERR(handle);
1366
1367         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1368                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1369                                     ctx, len, 0);
1370         if (!res) {
1371                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1372                 /*
1373                  * Update inode->i_flags - S_ENCRYPTED will be enabled,
1374                  * S_DAX may be disabled
1375                  */
1376                 ext4_set_inode_flags(inode, false);
1377                 res = ext4_mark_inode_dirty(handle, inode);
1378                 if (res)
1379                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1380         }
1381         res2 = ext4_journal_stop(handle);
1382
1383         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1384                 goto retry;
1385         if (!res)
1386                 res = res2;
1387         return res;
1388 }
1389
1390 static const union fscrypt_context *
1391 ext4_get_dummy_context(struct super_block *sb)
1392 {
1393         return EXT4_SB(sb)->s_dummy_enc_ctx.ctx;
1394 }
1395
1396 static bool ext4_has_stable_inodes(struct super_block *sb)
1397 {
1398         return ext4_has_feature_stable_inodes(sb);
1399 }
1400
1401 static void ext4_get_ino_and_lblk_bits(struct super_block *sb,
1402                                        int *ino_bits_ret, int *lblk_bits_ret)
1403 {
1404         *ino_bits_ret = 8 * sizeof(EXT4_SB(sb)->s_es->s_inodes_count);
1405         *lblk_bits_ret = 8 * sizeof(ext4_lblk_t);
1406 }
1407
1408 static const struct fscrypt_operations ext4_cryptops = {
1409         .key_prefix             = "ext4:",
1410         .get_context            = ext4_get_context,
1411         .set_context            = ext4_set_context,
1412         .get_dummy_context      = ext4_get_dummy_context,
1413         .empty_dir              = ext4_empty_dir,
1414         .max_namelen            = EXT4_NAME_LEN,
1415         .has_stable_inodes      = ext4_has_stable_inodes,
1416         .get_ino_and_lblk_bits  = ext4_get_ino_and_lblk_bits,
1417 };
1418 #endif
1419
1420 #ifdef CONFIG_QUOTA
1421 static const char * const quotatypes[] = INITQFNAMES;
1422 #define QTYPE2NAME(t) (quotatypes[t])
1423
1424 static int ext4_write_dquot(struct dquot *dquot);
1425 static int ext4_acquire_dquot(struct dquot *dquot);
1426 static int ext4_release_dquot(struct dquot *dquot);
1427 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1428 static int ext4_write_info(struct super_block *sb, int type);
1429 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1430                          const struct path *path);
1431 static int ext4_quota_on_mount(struct super_block *sb, int type);
1432 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1433                                size_t len, loff_t off);
1434 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1435                                 const char *data, size_t len, loff_t off);
1436 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1437                              unsigned int flags);
1438 static int ext4_enable_quotas(struct super_block *sb);
1439
1440 static struct dquot **ext4_get_dquots(struct inode *inode)
1441 {
1442         return EXT4_I(inode)->i_dquot;
1443 }
1444
1445 static const struct dquot_operations ext4_quota_operations = {
1446         .get_reserved_space     = ext4_get_reserved_space,
1447         .write_dquot            = ext4_write_dquot,
1448         .acquire_dquot          = ext4_acquire_dquot,
1449         .release_dquot          = ext4_release_dquot,
1450         .mark_dirty             = ext4_mark_dquot_dirty,
1451         .write_info             = ext4_write_info,
1452         .alloc_dquot            = dquot_alloc,
1453         .destroy_dquot          = dquot_destroy,
1454         .get_projid             = ext4_get_projid,
1455         .get_inode_usage        = ext4_get_inode_usage,
1456         .get_next_id            = dquot_get_next_id,
1457 };
1458
1459 static const struct quotactl_ops ext4_qctl_operations = {
1460         .quota_on       = ext4_quota_on,
1461         .quota_off      = ext4_quota_off,
1462         .quota_sync     = dquot_quota_sync,
1463         .get_state      = dquot_get_state,
1464         .set_info       = dquot_set_dqinfo,
1465         .get_dqblk      = dquot_get_dqblk,
1466         .set_dqblk      = dquot_set_dqblk,
1467         .get_nextdqblk  = dquot_get_next_dqblk,
1468 };
1469 #endif
1470
1471 static const struct super_operations ext4_sops = {
1472         .alloc_inode    = ext4_alloc_inode,
1473         .free_inode     = ext4_free_in_core_inode,
1474         .destroy_inode  = ext4_destroy_inode,
1475         .write_inode    = ext4_write_inode,
1476         .dirty_inode    = ext4_dirty_inode,
1477         .drop_inode     = ext4_drop_inode,
1478         .evict_inode    = ext4_evict_inode,
1479         .put_super      = ext4_put_super,
1480         .sync_fs        = ext4_sync_fs,
1481         .freeze_fs      = ext4_freeze,
1482         .unfreeze_fs    = ext4_unfreeze,
1483         .statfs         = ext4_statfs,
1484         .remount_fs     = ext4_remount,
1485         .show_options   = ext4_show_options,
1486 #ifdef CONFIG_QUOTA
1487         .quota_read     = ext4_quota_read,
1488         .quota_write    = ext4_quota_write,
1489         .get_dquots     = ext4_get_dquots,
1490 #endif
1491         .bdev_try_to_free_page = bdev_try_to_free_page,
1492 };
1493
1494 static const struct export_operations ext4_export_ops = {
1495         .fh_to_dentry = ext4_fh_to_dentry,
1496         .fh_to_parent = ext4_fh_to_parent,
1497         .get_parent = ext4_get_parent,
1498         .commit_metadata = ext4_nfs_commit_metadata,
1499 };
1500
1501 enum {
1502         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1503         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1504         Opt_nouid32, Opt_debug, Opt_removed,
1505         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1506         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1507         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1508         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1509         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1510         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1511         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1512         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1513         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1514         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version,
1515         Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never,
1516         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
1517         Opt_nowarn_on_error, Opt_mblk_io_submit,
1518         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1519         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1520         Opt_inode_readahead_blks, Opt_journal_ioprio,
1521         Opt_dioread_nolock, Opt_dioread_lock,
1522         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1523         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1524 };
1525
1526 static const match_table_t tokens = {
1527         {Opt_bsd_df, "bsddf"},
1528         {Opt_minix_df, "minixdf"},
1529         {Opt_grpid, "grpid"},
1530         {Opt_grpid, "bsdgroups"},
1531         {Opt_nogrpid, "nogrpid"},
1532         {Opt_nogrpid, "sysvgroups"},
1533         {Opt_resgid, "resgid=%u"},
1534         {Opt_resuid, "resuid=%u"},
1535         {Opt_sb, "sb=%u"},
1536         {Opt_err_cont, "errors=continue"},
1537         {Opt_err_panic, "errors=panic"},
1538         {Opt_err_ro, "errors=remount-ro"},
1539         {Opt_nouid32, "nouid32"},
1540         {Opt_debug, "debug"},
1541         {Opt_removed, "oldalloc"},
1542         {Opt_removed, "orlov"},
1543         {Opt_user_xattr, "user_xattr"},
1544         {Opt_nouser_xattr, "nouser_xattr"},
1545         {Opt_acl, "acl"},
1546         {Opt_noacl, "noacl"},
1547         {Opt_noload, "norecovery"},
1548         {Opt_noload, "noload"},
1549         {Opt_removed, "nobh"},
1550         {Opt_removed, "bh"},
1551         {Opt_commit, "commit=%u"},
1552         {Opt_min_batch_time, "min_batch_time=%u"},
1553         {Opt_max_batch_time, "max_batch_time=%u"},
1554         {Opt_journal_dev, "journal_dev=%u"},
1555         {Opt_journal_path, "journal_path=%s"},
1556         {Opt_journal_checksum, "journal_checksum"},
1557         {Opt_nojournal_checksum, "nojournal_checksum"},
1558         {Opt_journal_async_commit, "journal_async_commit"},
1559         {Opt_abort, "abort"},
1560         {Opt_data_journal, "data=journal"},
1561         {Opt_data_ordered, "data=ordered"},
1562         {Opt_data_writeback, "data=writeback"},
1563         {Opt_data_err_abort, "data_err=abort"},
1564         {Opt_data_err_ignore, "data_err=ignore"},
1565         {Opt_offusrjquota, "usrjquota="},
1566         {Opt_usrjquota, "usrjquota=%s"},
1567         {Opt_offgrpjquota, "grpjquota="},
1568         {Opt_grpjquota, "grpjquota=%s"},
1569         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1570         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1571         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1572         {Opt_grpquota, "grpquota"},
1573         {Opt_noquota, "noquota"},
1574         {Opt_quota, "quota"},
1575         {Opt_usrquota, "usrquota"},
1576         {Opt_prjquota, "prjquota"},
1577         {Opt_barrier, "barrier=%u"},
1578         {Opt_barrier, "barrier"},
1579         {Opt_nobarrier, "nobarrier"},
1580         {Opt_i_version, "i_version"},
1581         {Opt_dax, "dax"},
1582         {Opt_dax_always, "dax=always"},
1583         {Opt_dax_inode, "dax=inode"},
1584         {Opt_dax_never, "dax=never"},
1585         {Opt_stripe, "stripe=%u"},
1586         {Opt_delalloc, "delalloc"},
1587         {Opt_warn_on_error, "warn_on_error"},
1588         {Opt_nowarn_on_error, "nowarn_on_error"},
1589         {Opt_lazytime, "lazytime"},
1590         {Opt_nolazytime, "nolazytime"},
1591         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1592         {Opt_nodelalloc, "nodelalloc"},
1593         {Opt_removed, "mblk_io_submit"},
1594         {Opt_removed, "nomblk_io_submit"},
1595         {Opt_block_validity, "block_validity"},
1596         {Opt_noblock_validity, "noblock_validity"},
1597         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1598         {Opt_journal_ioprio, "journal_ioprio=%u"},
1599         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1600         {Opt_auto_da_alloc, "auto_da_alloc"},
1601         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1602         {Opt_dioread_nolock, "dioread_nolock"},
1603         {Opt_dioread_lock, "nodioread_nolock"},
1604         {Opt_dioread_lock, "dioread_lock"},
1605         {Opt_discard, "discard"},
1606         {Opt_nodiscard, "nodiscard"},
1607         {Opt_init_itable, "init_itable=%u"},
1608         {Opt_init_itable, "init_itable"},
1609         {Opt_noinit_itable, "noinit_itable"},
1610         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1611         {Opt_test_dummy_encryption, "test_dummy_encryption=%s"},
1612         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1613         {Opt_nombcache, "nombcache"},
1614         {Opt_nombcache, "no_mbcache"},  /* for backward compatibility */
1615         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1616         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1617         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1618         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1619         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1620         {Opt_err, NULL},
1621 };
1622
1623 static ext4_fsblk_t get_sb_block(void **data)
1624 {
1625         ext4_fsblk_t    sb_block;
1626         char            *options = (char *) *data;
1627
1628         if (!options || strncmp(options, "sb=", 3) != 0)
1629                 return 1;       /* Default location */
1630
1631         options += 3;
1632         /* TODO: use simple_strtoll with >32bit ext4 */
1633         sb_block = simple_strtoul(options, &options, 0);
1634         if (*options && *options != ',') {
1635                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1636                        (char *) *data);
1637                 return 1;
1638         }
1639         if (*options == ',')
1640                 options++;
1641         *data = (void *) options;
1642
1643         return sb_block;
1644 }
1645
1646 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1647 static const char deprecated_msg[] =
1648         "Mount option \"%s\" will be removed by %s\n"
1649         "Contact [email protected] if you think we should keep it.\n";
1650
1651 #ifdef CONFIG_QUOTA
1652 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1653 {
1654         struct ext4_sb_info *sbi = EXT4_SB(sb);
1655         char *qname, *old_qname = get_qf_name(sb, sbi, qtype);
1656         int ret = -1;
1657
1658         if (sb_any_quota_loaded(sb) && !old_qname) {
1659                 ext4_msg(sb, KERN_ERR,
1660                         "Cannot change journaled "
1661                         "quota options when quota turned on");
1662                 return -1;
1663         }
1664         if (ext4_has_feature_quota(sb)) {
1665                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1666                          "ignored when QUOTA feature is enabled");
1667                 return 1;
1668         }
1669         qname = match_strdup(args);
1670         if (!qname) {
1671                 ext4_msg(sb, KERN_ERR,
1672                         "Not enough memory for storing quotafile name");
1673                 return -1;
1674         }
1675         if (old_qname) {
1676                 if (strcmp(old_qname, qname) == 0)
1677                         ret = 1;
1678                 else
1679                         ext4_msg(sb, KERN_ERR,
1680                                  "%s quota file already specified",
1681                                  QTYPE2NAME(qtype));
1682                 goto errout;
1683         }
1684         if (strchr(qname, '/')) {
1685                 ext4_msg(sb, KERN_ERR,
1686                         "quotafile must be on filesystem root");
1687                 goto errout;
1688         }
1689         rcu_assign_pointer(sbi->s_qf_names[qtype], qname);
1690         set_opt(sb, QUOTA);
1691         return 1;
1692 errout:
1693         kfree(qname);
1694         return ret;
1695 }
1696
1697 static int clear_qf_name(struct super_block *sb, int qtype)
1698 {
1699
1700         struct ext4_sb_info *sbi = EXT4_SB(sb);
1701         char *old_qname = get_qf_name(sb, sbi, qtype);
1702
1703         if (sb_any_quota_loaded(sb) && old_qname) {
1704                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1705                         " when quota turned on");
1706                 return -1;
1707         }
1708         rcu_assign_pointer(sbi->s_qf_names[qtype], NULL);
1709         synchronize_rcu();
1710         kfree(old_qname);
1711         return 1;
1712 }
1713 #endif
1714
1715 #define MOPT_SET        0x0001
1716 #define MOPT_CLEAR      0x0002
1717 #define MOPT_NOSUPPORT  0x0004
1718 #define MOPT_EXPLICIT   0x0008
1719 #define MOPT_CLEAR_ERR  0x0010
1720 #define MOPT_GTE0       0x0020
1721 #ifdef CONFIG_QUOTA
1722 #define MOPT_Q          0
1723 #define MOPT_QFMT       0x0040
1724 #else
1725 #define MOPT_Q          MOPT_NOSUPPORT
1726 #define MOPT_QFMT       MOPT_NOSUPPORT
1727 #endif
1728 #define MOPT_DATAJ      0x0080
1729 #define MOPT_NO_EXT2    0x0100
1730 #define MOPT_NO_EXT3    0x0200
1731 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1732 #define MOPT_STRING     0x0400
1733 #define MOPT_SKIP       0x0800
1734
1735 static const struct mount_opts {
1736         int     token;
1737         int     mount_opt;
1738         int     flags;
1739 } ext4_mount_opts[] = {
1740         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1741         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1742         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1743         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1744         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1745         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1746         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1747          MOPT_EXT4_ONLY | MOPT_SET},
1748         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1749          MOPT_EXT4_ONLY | MOPT_CLEAR},
1750         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1751         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1752         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1753          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1754         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1755          MOPT_EXT4_ONLY | MOPT_CLEAR},
1756         {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET},
1757         {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR},
1758         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1759          MOPT_EXT4_ONLY | MOPT_CLEAR},
1760         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1761          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1762         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1763                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1764          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1765         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1766         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1767         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1768         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1769         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1770          MOPT_NO_EXT2},
1771         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1772          MOPT_NO_EXT2},
1773         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1774         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1775         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1776         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1777         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1778         {Opt_commit, 0, MOPT_GTE0},
1779         {Opt_max_batch_time, 0, MOPT_GTE0},
1780         {Opt_min_batch_time, 0, MOPT_GTE0},
1781         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1782         {Opt_init_itable, 0, MOPT_GTE0},
1783         {Opt_dax, EXT4_MOUNT_DAX_ALWAYS, MOPT_SET | MOPT_SKIP},
1784         {Opt_dax_always, EXT4_MOUNT_DAX_ALWAYS,
1785                 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP},
1786         {Opt_dax_inode, EXT4_MOUNT2_DAX_INODE,
1787                 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP},
1788         {Opt_dax_never, EXT4_MOUNT2_DAX_NEVER,
1789                 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP},
1790         {Opt_stripe, 0, MOPT_GTE0},
1791         {Opt_resuid, 0, MOPT_GTE0},
1792         {Opt_resgid, 0, MOPT_GTE0},
1793         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1794         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1795         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1796         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1797         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1798         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1799          MOPT_NO_EXT2 | MOPT_DATAJ},
1800         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1801         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1802 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1803         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1804         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1805 #else
1806         {Opt_acl, 0, MOPT_NOSUPPORT},
1807         {Opt_noacl, 0, MOPT_NOSUPPORT},
1808 #endif
1809         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1810         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1811         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1812         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1813         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1814                                                         MOPT_SET | MOPT_Q},
1815         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1816                                                         MOPT_SET | MOPT_Q},
1817         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1818                                                         MOPT_SET | MOPT_Q},
1819         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1820                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1821                                                         MOPT_CLEAR | MOPT_Q},
1822         {Opt_usrjquota, 0, MOPT_Q},
1823         {Opt_grpjquota, 0, MOPT_Q},
1824         {Opt_offusrjquota, 0, MOPT_Q},
1825         {Opt_offgrpjquota, 0, MOPT_Q},
1826         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1827         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1828         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1829         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1830         {Opt_test_dummy_encryption, 0, MOPT_STRING},
1831         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1832         {Opt_err, 0, 0}
1833 };
1834
1835 #ifdef CONFIG_UNICODE
1836 static const struct ext4_sb_encodings {
1837         __u16 magic;
1838         char *name;
1839         char *version;
1840 } ext4_sb_encoding_map[] = {
1841         {EXT4_ENC_UTF8_12_1, "utf8", "12.1.0"},
1842 };
1843
1844 static int ext4_sb_read_encoding(const struct ext4_super_block *es,
1845                                  const struct ext4_sb_encodings **encoding,
1846                                  __u16 *flags)
1847 {
1848         __u16 magic = le16_to_cpu(es->s_encoding);
1849         int i;
1850
1851         for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++)
1852                 if (magic == ext4_sb_encoding_map[i].magic)
1853                         break;
1854
1855         if (i >= ARRAY_SIZE(ext4_sb_encoding_map))
1856                 return -EINVAL;
1857
1858         *encoding = &ext4_sb_encoding_map[i];
1859         *flags = le16_to_cpu(es->s_encoding_flags);
1860
1861         return 0;
1862 }
1863 #endif
1864
1865 static int ext4_set_test_dummy_encryption(struct super_block *sb,
1866                                           const char *opt,
1867                                           const substring_t *arg,
1868                                           bool is_remount)
1869 {
1870 #ifdef CONFIG_FS_ENCRYPTION
1871         struct ext4_sb_info *sbi = EXT4_SB(sb);
1872         int err;
1873
1874         /*
1875          * This mount option is just for testing, and it's not worthwhile to
1876          * implement the extra complexity (e.g. RCU protection) that would be
1877          * needed to allow it to be set or changed during remount.  We do allow
1878          * it to be specified during remount, but only if there is no change.
1879          */
1880         if (is_remount && !sbi->s_dummy_enc_ctx.ctx) {
1881                 ext4_msg(sb, KERN_WARNING,
1882                          "Can't set test_dummy_encryption on remount");
1883                 return -1;
1884         }
1885         err = fscrypt_set_test_dummy_encryption(sb, arg, &sbi->s_dummy_enc_ctx);
1886         if (err) {
1887                 if (err == -EEXIST)
1888                         ext4_msg(sb, KERN_WARNING,
1889                                  "Can't change test_dummy_encryption on remount");
1890                 else if (err == -EINVAL)
1891                         ext4_msg(sb, KERN_WARNING,
1892                                  "Value of option \"%s\" is unrecognized", opt);
1893                 else
1894                         ext4_msg(sb, KERN_WARNING,
1895                                  "Error processing option \"%s\" [%d]",
1896                                  opt, err);
1897                 return -1;
1898         }
1899         ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled");
1900 #else
1901         ext4_msg(sb, KERN_WARNING,
1902                  "Test dummy encryption mount option ignored");
1903 #endif
1904         return 1;
1905 }
1906
1907 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1908                             substring_t *args, unsigned long *journal_devnum,
1909                             unsigned int *journal_ioprio, int is_remount)
1910 {
1911         struct ext4_sb_info *sbi = EXT4_SB(sb);
1912         const struct mount_opts *m;
1913         kuid_t uid;
1914         kgid_t gid;
1915         int arg = 0;
1916
1917 #ifdef CONFIG_QUOTA
1918         if (token == Opt_usrjquota)
1919                 return set_qf_name(sb, USRQUOTA, &args[0]);
1920         else if (token == Opt_grpjquota)
1921                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1922         else if (token == Opt_offusrjquota)
1923                 return clear_qf_name(sb, USRQUOTA);
1924         else if (token == Opt_offgrpjquota)
1925                 return clear_qf_name(sb, GRPQUOTA);
1926 #endif
1927         switch (token) {
1928         case Opt_noacl:
1929         case Opt_nouser_xattr:
1930                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1931                 break;
1932         case Opt_sb:
1933                 return 1;       /* handled by get_sb_block() */
1934         case Opt_removed:
1935                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1936                 return 1;
1937         case Opt_abort:
1938                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1939                 return 1;
1940         case Opt_i_version:
1941                 sb->s_flags |= SB_I_VERSION;
1942                 return 1;
1943         case Opt_lazytime:
1944                 sb->s_flags |= SB_LAZYTIME;
1945                 return 1;
1946         case Opt_nolazytime:
1947                 sb->s_flags &= ~SB_LAZYTIME;
1948                 return 1;
1949         }
1950
1951         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1952                 if (token == m->token)
1953                         break;
1954
1955         if (m->token == Opt_err) {
1956                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1957                          "or missing value", opt);
1958                 return -1;
1959         }
1960
1961         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1962                 ext4_msg(sb, KERN_ERR,
1963                          "Mount option \"%s\" incompatible with ext2", opt);
1964                 return -1;
1965         }
1966         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1967                 ext4_msg(sb, KERN_ERR,
1968                          "Mount option \"%s\" incompatible with ext3", opt);
1969                 return -1;
1970         }
1971
1972         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1973                 return -1;
1974         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1975                 return -1;
1976         if (m->flags & MOPT_EXPLICIT) {
1977                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1978                         set_opt2(sb, EXPLICIT_DELALLOC);
1979                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1980                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1981                 } else
1982                         return -1;
1983         }
1984         if (m->flags & MOPT_CLEAR_ERR)
1985                 clear_opt(sb, ERRORS_MASK);
1986         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1987                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1988                          "options when quota turned on");
1989                 return -1;
1990         }
1991
1992         if (m->flags & MOPT_NOSUPPORT) {
1993                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1994         } else if (token == Opt_commit) {
1995                 if (arg == 0)
1996                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1997                 else if (arg > INT_MAX / HZ) {
1998                         ext4_msg(sb, KERN_ERR,
1999                                  "Invalid commit interval %d, "
2000                                  "must be smaller than %d",
2001                                  arg, INT_MAX / HZ);
2002                         return -1;
2003                 }
2004                 sbi->s_commit_interval = HZ * arg;
2005         } else if (token == Opt_debug_want_extra_isize) {
2006                 if ((arg & 1) ||
2007                     (arg < 4) ||
2008                     (arg > (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE))) {
2009                         ext4_msg(sb, KERN_ERR,
2010                                  "Invalid want_extra_isize %d", arg);
2011                         return -1;
2012                 }
2013                 sbi->s_want_extra_isize = arg;
2014         } else if (token == Opt_max_batch_time) {
2015                 sbi->s_max_batch_time = arg;
2016         } else if (token == Opt_min_batch_time) {
2017                 sbi->s_min_batch_time = arg;
2018         } else if (token == Opt_inode_readahead_blks) {
2019                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
2020                         ext4_msg(sb, KERN_ERR,
2021                                  "EXT4-fs: inode_readahead_blks must be "
2022                                  "0 or a power of 2 smaller than 2^31");
2023                         return -1;
2024                 }
2025                 sbi->s_inode_readahead_blks = arg;
2026         } else if (token == Opt_init_itable) {
2027                 set_opt(sb, INIT_INODE_TABLE);
2028                 if (!args->from)
2029                         arg = EXT4_DEF_LI_WAIT_MULT;
2030                 sbi->s_li_wait_mult = arg;
2031         } else if (token == Opt_max_dir_size_kb) {
2032                 sbi->s_max_dir_size_kb = arg;
2033         } else if (token == Opt_stripe) {
2034                 sbi->s_stripe = arg;
2035         } else if (token == Opt_resuid) {
2036                 uid = make_kuid(current_user_ns(), arg);
2037                 if (!uid_valid(uid)) {
2038                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
2039                         return -1;
2040                 }
2041                 sbi->s_resuid = uid;
2042         } else if (token == Opt_resgid) {
2043                 gid = make_kgid(current_user_ns(), arg);
2044                 if (!gid_valid(gid)) {
2045                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
2046                         return -1;
2047                 }
2048                 sbi->s_resgid = gid;
2049         } else if (token == Opt_journal_dev) {
2050                 if (is_remount) {
2051                         ext4_msg(sb, KERN_ERR,
2052                                  "Cannot specify journal on remount");
2053                         return -1;
2054                 }
2055                 *journal_devnum = arg;
2056         } else if (token == Opt_journal_path) {
2057                 char *journal_path;
2058                 struct inode *journal_inode;
2059                 struct path path;
2060                 int error;
2061
2062                 if (is_remount) {
2063                         ext4_msg(sb, KERN_ERR,
2064                                  "Cannot specify journal on remount");
2065                         return -1;
2066                 }
2067                 journal_path = match_strdup(&args[0]);
2068                 if (!journal_path) {
2069                         ext4_msg(sb, KERN_ERR, "error: could not dup "
2070                                 "journal device string");
2071                         return -1;
2072                 }
2073
2074                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
2075                 if (error) {
2076                         ext4_msg(sb, KERN_ERR, "error: could not find "
2077                                 "journal device path: error %d", error);
2078                         kfree(journal_path);
2079                         return -1;
2080                 }
2081
2082                 journal_inode = d_inode(path.dentry);
2083                 if (!S_ISBLK(journal_inode->i_mode)) {
2084                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
2085                                 "is not a block device", journal_path);
2086                         path_put(&path);
2087                         kfree(journal_path);
2088                         return -1;
2089                 }
2090
2091                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
2092                 path_put(&path);
2093                 kfree(journal_path);
2094         } else if (token == Opt_journal_ioprio) {
2095                 if (arg > 7) {
2096                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
2097                                  " (must be 0-7)");
2098                         return -1;
2099                 }
2100                 *journal_ioprio =
2101                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
2102         } else if (token == Opt_test_dummy_encryption) {
2103                 return ext4_set_test_dummy_encryption(sb, opt, &args[0],
2104                                                       is_remount);
2105         } else if (m->flags & MOPT_DATAJ) {
2106                 if (is_remount) {
2107                         if (!sbi->s_journal)
2108                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
2109                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
2110                                 ext4_msg(sb, KERN_ERR,
2111                                          "Cannot change data mode on remount");
2112                                 return -1;
2113                         }
2114                 } else {
2115                         clear_opt(sb, DATA_FLAGS);
2116                         sbi->s_mount_opt |= m->mount_opt;
2117                 }
2118 #ifdef CONFIG_QUOTA
2119         } else if (m->flags & MOPT_QFMT) {
2120                 if (sb_any_quota_loaded(sb) &&
2121                     sbi->s_jquota_fmt != m->mount_opt) {
2122                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
2123                                  "quota options when quota turned on");
2124                         return -1;
2125                 }
2126                 if (ext4_has_feature_quota(sb)) {
2127                         ext4_msg(sb, KERN_INFO,
2128                                  "Quota format mount options ignored "
2129                                  "when QUOTA feature is enabled");
2130                         return 1;
2131                 }
2132                 sbi->s_jquota_fmt = m->mount_opt;
2133 #endif
2134         } else if (token == Opt_dax || token == Opt_dax_always ||
2135                    token == Opt_dax_inode || token == Opt_dax_never) {
2136 #ifdef CONFIG_FS_DAX
2137                 switch (token) {
2138                 case Opt_dax:
2139                 case Opt_dax_always:
2140                         if (is_remount &&
2141                             (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2142                              (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) {
2143                         fail_dax_change_remount:
2144                                 ext4_msg(sb, KERN_ERR, "can't change "
2145                                          "dax mount option while remounting");
2146                                 return -1;
2147                         }
2148                         if (is_remount &&
2149                             (test_opt(sb, DATA_FLAGS) ==
2150                              EXT4_MOUNT_JOURNAL_DATA)) {
2151                                     ext4_msg(sb, KERN_ERR, "can't mount with "
2152                                              "both data=journal and dax");
2153                                     return -1;
2154                         }
2155                         ext4_msg(sb, KERN_WARNING,
2156                                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
2157                         sbi->s_mount_opt |= EXT4_MOUNT_DAX_ALWAYS;
2158                         sbi->s_mount_opt2 &= ~EXT4_MOUNT2_DAX_NEVER;
2159                         break;
2160                 case Opt_dax_never:
2161                         if (is_remount &&
2162                             (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2163                              (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS)))
2164                                 goto fail_dax_change_remount;
2165                         sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_NEVER;
2166                         sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS;
2167                         break;
2168                 case Opt_dax_inode:
2169                         if (is_remount &&
2170                             ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2171                              (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2172                              !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE)))
2173                                 goto fail_dax_change_remount;
2174                         sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS;
2175                         sbi->s_mount_opt2 &= ~EXT4_MOUNT2_DAX_NEVER;
2176                         /* Strictly for printing options */
2177                         sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_INODE;
2178                         break;
2179                 }
2180 #else
2181                 ext4_msg(sb, KERN_INFO, "dax option not supported");
2182                 sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_NEVER;
2183                 sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS;
2184                 return -1;
2185 #endif
2186         } else if (token == Opt_data_err_abort) {
2187                 sbi->s_mount_opt |= m->mount_opt;
2188         } else if (token == Opt_data_err_ignore) {
2189                 sbi->s_mount_opt &= ~m->mount_opt;
2190         } else {
2191                 if (!args->from)
2192                         arg = 1;
2193                 if (m->flags & MOPT_CLEAR)
2194                         arg = !arg;
2195                 else if (unlikely(!(m->flags & MOPT_SET))) {
2196                         ext4_msg(sb, KERN_WARNING,
2197                                  "buggy handling of option %s", opt);
2198                         WARN_ON(1);
2199                         return -1;
2200                 }
2201                 if (arg != 0)
2202                         sbi->s_mount_opt |= m->mount_opt;
2203                 else
2204                         sbi->s_mount_opt &= ~m->mount_opt;
2205         }
2206         return 1;
2207 }
2208
2209 static int parse_options(char *options, struct super_block *sb,
2210                          unsigned long *journal_devnum,
2211                          unsigned int *journal_ioprio,
2212                          int is_remount)
2213 {
2214         struct ext4_sb_info __maybe_unused *sbi = EXT4_SB(sb);
2215         char *p, __maybe_unused *usr_qf_name, __maybe_unused *grp_qf_name;
2216         substring_t args[MAX_OPT_ARGS];
2217         int token;
2218
2219         if (!options)
2220                 return 1;
2221
2222         while ((p = strsep(&options, ",")) != NULL) {
2223                 if (!*p)
2224                         continue;
2225                 /*
2226                  * Initialize args struct so we know whether arg was
2227                  * found; some options take optional arguments.
2228                  */
2229                 args[0].to = args[0].from = NULL;
2230                 token = match_token(p, tokens, args);
2231                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
2232                                      journal_ioprio, is_remount) < 0)
2233                         return 0;
2234         }
2235 #ifdef CONFIG_QUOTA
2236         /*
2237          * We do the test below only for project quotas. 'usrquota' and
2238          * 'grpquota' mount options are allowed even without quota feature
2239          * to support legacy quotas in quota files.
2240          */
2241         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
2242                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
2243                          "Cannot enable project quota enforcement.");
2244                 return 0;
2245         }
2246         usr_qf_name = get_qf_name(sb, sbi, USRQUOTA);
2247         grp_qf_name = get_qf_name(sb, sbi, GRPQUOTA);
2248         if (usr_qf_name || grp_qf_name) {
2249                 if (test_opt(sb, USRQUOTA) && usr_qf_name)
2250                         clear_opt(sb, USRQUOTA);
2251
2252                 if (test_opt(sb, GRPQUOTA) && grp_qf_name)
2253                         clear_opt(sb, GRPQUOTA);
2254
2255                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
2256                         ext4_msg(sb, KERN_ERR, "old and new quota "
2257                                         "format mixing");
2258                         return 0;
2259                 }
2260
2261                 if (!sbi->s_jquota_fmt) {
2262                         ext4_msg(sb, KERN_ERR, "journaled quota format "
2263                                         "not specified");
2264                         return 0;
2265                 }
2266         }
2267 #endif
2268         if (test_opt(sb, DIOREAD_NOLOCK)) {
2269                 int blocksize =
2270                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
2271                 if (blocksize < PAGE_SIZE)
2272                         ext4_msg(sb, KERN_WARNING, "Warning: mounting with an "
2273                                  "experimental mount option 'dioread_nolock' "
2274                                  "for blocksize < PAGE_SIZE");
2275         }
2276         return 1;
2277 }
2278
2279 static inline void ext4_show_quota_options(struct seq_file *seq,
2280                                            struct super_block *sb)
2281 {
2282 #if defined(CONFIG_QUOTA)
2283         struct ext4_sb_info *sbi = EXT4_SB(sb);
2284         char *usr_qf_name, *grp_qf_name;
2285
2286         if (sbi->s_jquota_fmt) {
2287                 char *fmtname = "";
2288
2289                 switch (sbi->s_jquota_fmt) {
2290                 case QFMT_VFS_OLD:
2291                         fmtname = "vfsold";
2292                         break;
2293                 case QFMT_VFS_V0:
2294                         fmtname = "vfsv0";
2295                         break;
2296                 case QFMT_VFS_V1:
2297                         fmtname = "vfsv1";
2298                         break;
2299                 }
2300                 seq_printf(seq, ",jqfmt=%s", fmtname);
2301         }
2302
2303         rcu_read_lock();
2304         usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]);
2305         grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]);
2306         if (usr_qf_name)
2307                 seq_show_option(seq, "usrjquota", usr_qf_name);
2308         if (grp_qf_name)
2309                 seq_show_option(seq, "grpjquota", grp_qf_name);
2310         rcu_read_unlock();
2311 #endif
2312 }
2313
2314 static const char *token2str(int token)
2315 {
2316         const struct match_token *t;
2317
2318         for (t = tokens; t->token != Opt_err; t++)
2319                 if (t->token == token && !strchr(t->pattern, '='))
2320                         break;
2321         return t->pattern;
2322 }
2323
2324 /*
2325  * Show an option if
2326  *  - it's set to a non-default value OR
2327  *  - if the per-sb default is different from the global default
2328  */
2329 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2330                               int nodefs)
2331 {
2332         struct ext4_sb_info *sbi = EXT4_SB(sb);
2333         struct ext4_super_block *es = sbi->s_es;
2334         int def_errors, def_mount_opt = sbi->s_def_mount_opt;
2335         const struct mount_opts *m;
2336         char sep = nodefs ? '\n' : ',';
2337
2338 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2339 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2340
2341         if (sbi->s_sb_block != 1)
2342                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2343
2344         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2345                 int want_set = m->flags & MOPT_SET;
2346                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2347                     (m->flags & MOPT_CLEAR_ERR) || m->flags & MOPT_SKIP)
2348                         continue;
2349                 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2350                         continue; /* skip if same as the default */
2351                 if ((want_set &&
2352                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2353                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2354                         continue; /* select Opt_noFoo vs Opt_Foo */
2355                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2356         }
2357
2358         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2359             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2360                 SEQ_OPTS_PRINT("resuid=%u",
2361                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2362         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2363             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2364                 SEQ_OPTS_PRINT("resgid=%u",
2365                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2366         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2367         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2368                 SEQ_OPTS_PUTS("errors=remount-ro");
2369         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2370                 SEQ_OPTS_PUTS("errors=continue");
2371         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2372                 SEQ_OPTS_PUTS("errors=panic");
2373         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2374                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2375         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2376                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2377         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2378                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2379         if (sb->s_flags & SB_I_VERSION)
2380                 SEQ_OPTS_PUTS("i_version");
2381         if (nodefs || sbi->s_stripe)
2382                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2383         if (nodefs || EXT4_MOUNT_DATA_FLAGS &
2384                         (sbi->s_mount_opt ^ def_mount_opt)) {
2385                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2386                         SEQ_OPTS_PUTS("data=journal");
2387                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2388                         SEQ_OPTS_PUTS("data=ordered");
2389                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2390                         SEQ_OPTS_PUTS("data=writeback");
2391         }
2392         if (nodefs ||
2393             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2394                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2395                                sbi->s_inode_readahead_blks);
2396
2397         if (test_opt(sb, INIT_INODE_TABLE) && (nodefs ||
2398                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2399                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2400         if (nodefs || sbi->s_max_dir_size_kb)
2401                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2402         if (test_opt(sb, DATA_ERR_ABORT))
2403                 SEQ_OPTS_PUTS("data_err=abort");
2404
2405         fscrypt_show_test_dummy_encryption(seq, sep, sb);
2406
2407         if (test_opt(sb, DAX_ALWAYS)) {
2408                 if (IS_EXT2_SB(sb))
2409                         SEQ_OPTS_PUTS("dax");
2410                 else
2411                         SEQ_OPTS_PUTS("dax=always");
2412         } else if (test_opt2(sb, DAX_NEVER)) {
2413                 SEQ_OPTS_PUTS("dax=never");
2414         } else if (test_opt2(sb, DAX_INODE)) {
2415                 SEQ_OPTS_PUTS("dax=inode");
2416         }
2417
2418         ext4_show_quota_options(seq, sb);
2419         return 0;
2420 }
2421
2422 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2423 {
2424         return _ext4_show_options(seq, root->d_sb, 0);
2425 }
2426
2427 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2428 {
2429         struct super_block *sb = seq->private;
2430         int rc;
2431
2432         seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
2433         rc = _ext4_show_options(seq, sb, 1);
2434         seq_puts(seq, "\n");
2435         return rc;
2436 }
2437
2438 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2439                             int read_only)
2440 {
2441         struct ext4_sb_info *sbi = EXT4_SB(sb);
2442         int err = 0;
2443
2444         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2445                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2446                          "forcing read-only mode");
2447                 err = -EROFS;
2448                 goto done;
2449         }
2450         if (read_only)
2451                 goto done;
2452         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2453                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2454                          "running e2fsck is recommended");
2455         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2456                 ext4_msg(sb, KERN_WARNING,
2457                          "warning: mounting fs with errors, "
2458                          "running e2fsck is recommended");
2459         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2460                  le16_to_cpu(es->s_mnt_count) >=
2461                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2462                 ext4_msg(sb, KERN_WARNING,
2463                          "warning: maximal mount count reached, "
2464                          "running e2fsck is recommended");
2465         else if (le32_to_cpu(es->s_checkinterval) &&
2466                  (ext4_get_tstamp(es, s_lastcheck) +
2467                   le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds()))
2468                 ext4_msg(sb, KERN_WARNING,
2469                          "warning: checktime reached, "
2470                          "running e2fsck is recommended");
2471         if (!sbi->s_journal)
2472                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2473         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2474                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2475         le16_add_cpu(&es->s_mnt_count, 1);
2476         ext4_update_tstamp(es, s_mtime);
2477         if (sbi->s_journal)
2478                 ext4_set_feature_journal_needs_recovery(sb);
2479
2480         err = ext4_commit_super(sb, 1);
2481 done:
2482         if (test_opt(sb, DEBUG))
2483                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2484                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2485                         sb->s_blocksize,
2486                         sbi->s_groups_count,
2487                         EXT4_BLOCKS_PER_GROUP(sb),
2488                         EXT4_INODES_PER_GROUP(sb),
2489                         sbi->s_mount_opt, sbi->s_mount_opt2);
2490
2491         cleancache_init_fs(sb);
2492         return err;
2493 }
2494
2495 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2496 {
2497         struct ext4_sb_info *sbi = EXT4_SB(sb);
2498         struct flex_groups **old_groups, **new_groups;
2499         int size, i, j;
2500
2501         if (!sbi->s_log_groups_per_flex)
2502                 return 0;
2503
2504         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2505         if (size <= sbi->s_flex_groups_allocated)
2506                 return 0;
2507
2508         new_groups = kvzalloc(roundup_pow_of_two(size *
2509                               sizeof(*sbi->s_flex_groups)), GFP_KERNEL);
2510         if (!new_groups) {
2511                 ext4_msg(sb, KERN_ERR,
2512                          "not enough memory for %d flex group pointers", size);
2513                 return -ENOMEM;
2514         }
2515         for (i = sbi->s_flex_groups_allocated; i < size; i++) {
2516                 new_groups[i] = kvzalloc(roundup_pow_of_two(
2517                                          sizeof(struct flex_groups)),
2518                                          GFP_KERNEL);
2519                 if (!new_groups[i]) {
2520                         for (j = sbi->s_flex_groups_allocated; j < i; j++)
2521                                 kvfree(new_groups[j]);
2522                         kvfree(new_groups);
2523                         ext4_msg(sb, KERN_ERR,
2524                                  "not enough memory for %d flex groups", size);
2525                         return -ENOMEM;
2526                 }
2527         }
2528         rcu_read_lock();
2529         old_groups = rcu_dereference(sbi->s_flex_groups);
2530         if (old_groups)
2531                 memcpy(new_groups, old_groups,
2532                        (sbi->s_flex_groups_allocated *
2533                         sizeof(struct flex_groups *)));
2534         rcu_read_unlock();
2535         rcu_assign_pointer(sbi->s_flex_groups, new_groups);
2536         sbi->s_flex_groups_allocated = size;
2537         if (old_groups)
2538                 ext4_kvfree_array_rcu(old_groups);
2539         return 0;
2540 }
2541
2542 static int ext4_fill_flex_info(struct super_block *sb)
2543 {
2544         struct ext4_sb_info *sbi = EXT4_SB(sb);
2545         struct ext4_group_desc *gdp = NULL;
2546         struct flex_groups *fg;
2547         ext4_group_t flex_group;
2548         int i, err;
2549
2550         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2551         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2552                 sbi->s_log_groups_per_flex = 0;
2553                 return 1;
2554         }
2555
2556         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2557         if (err)
2558                 goto failed;
2559
2560         for (i = 0; i < sbi->s_groups_count; i++) {
2561                 gdp = ext4_get_group_desc(sb, i, NULL);
2562
2563                 flex_group = ext4_flex_group(sbi, i);
2564                 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
2565                 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes);
2566                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2567                              &fg->free_clusters);
2568                 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs);
2569         }
2570
2571         return 1;
2572 failed:
2573         return 0;
2574 }
2575
2576 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2577                                    struct ext4_group_desc *gdp)
2578 {
2579         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2580         __u16 crc = 0;
2581         __le32 le_group = cpu_to_le32(block_group);
2582         struct ext4_sb_info *sbi = EXT4_SB(sb);
2583
2584         if (ext4_has_metadata_csum(sbi->s_sb)) {
2585                 /* Use new metadata_csum algorithm */
2586                 __u32 csum32;
2587                 __u16 dummy_csum = 0;
2588
2589                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2590                                      sizeof(le_group));
2591                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2592                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2593                                      sizeof(dummy_csum));
2594                 offset += sizeof(dummy_csum);
2595                 if (offset < sbi->s_desc_size)
2596                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2597                                              sbi->s_desc_size - offset);
2598
2599                 crc = csum32 & 0xFFFF;
2600                 goto out;
2601         }
2602
2603         /* old crc16 code */
2604         if (!ext4_has_feature_gdt_csum(sb))
2605                 return 0;
2606
2607         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2608         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2609         crc = crc16(crc, (__u8 *)gdp, offset);
2610         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2611         /* for checksum of struct ext4_group_desc do the rest...*/
2612         if (ext4_has_feature_64bit(sb) &&
2613             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2614                 crc = crc16(crc, (__u8 *)gdp + offset,
2615                             le16_to_cpu(sbi->s_es->s_desc_size) -
2616                                 offset);
2617
2618 out:
2619         return cpu_to_le16(crc);
2620 }
2621
2622 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2623                                 struct ext4_group_desc *gdp)
2624 {
2625         if (ext4_has_group_desc_csum(sb) &&
2626             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2627                 return 0;
2628
2629         return 1;
2630 }
2631
2632 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2633                               struct ext4_group_desc *gdp)
2634 {
2635         if (!ext4_has_group_desc_csum(sb))
2636                 return;
2637         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2638 }
2639
2640 /* Called at mount-time, super-block is locked */
2641 static int ext4_check_descriptors(struct super_block *sb,
2642                                   ext4_fsblk_t sb_block,
2643                                   ext4_group_t *first_not_zeroed)
2644 {
2645         struct ext4_sb_info *sbi = EXT4_SB(sb);
2646         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2647         ext4_fsblk_t last_block;
2648         ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0);
2649         ext4_fsblk_t block_bitmap;
2650         ext4_fsblk_t inode_bitmap;
2651         ext4_fsblk_t inode_table;
2652         int flexbg_flag = 0;
2653         ext4_group_t i, grp = sbi->s_groups_count;
2654
2655         if (ext4_has_feature_flex_bg(sb))
2656                 flexbg_flag = 1;
2657
2658         ext4_debug("Checking group descriptors");
2659
2660         for (i = 0; i < sbi->s_groups_count; i++) {
2661                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2662
2663                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2664                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2665                 else
2666                         last_block = first_block +
2667                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2668
2669                 if ((grp == sbi->s_groups_count) &&
2670                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2671                         grp = i;
2672
2673                 block_bitmap = ext4_block_bitmap(sb, gdp);
2674                 if (block_bitmap == sb_block) {
2675                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2676                                  "Block bitmap for group %u overlaps "
2677                                  "superblock", i);
2678                         if (!sb_rdonly(sb))
2679                                 return 0;
2680                 }
2681                 if (block_bitmap >= sb_block + 1 &&
2682                     block_bitmap <= last_bg_block) {
2683                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2684                                  "Block bitmap for group %u overlaps "
2685                                  "block group descriptors", i);
2686                         if (!sb_rdonly(sb))
2687                                 return 0;
2688                 }
2689                 if (block_bitmap < first_block || block_bitmap > last_block) {
2690                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2691                                "Block bitmap for group %u not in group "
2692                                "(block %llu)!", i, block_bitmap);
2693                         return 0;
2694                 }
2695                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2696                 if (inode_bitmap == sb_block) {
2697                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2698                                  "Inode bitmap for group %u overlaps "
2699                                  "superblock", i);
2700                         if (!sb_rdonly(sb))
2701                                 return 0;
2702                 }
2703                 if (inode_bitmap >= sb_block + 1 &&
2704                     inode_bitmap <= last_bg_block) {
2705                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2706                                  "Inode bitmap for group %u overlaps "
2707                                  "block group descriptors", i);
2708                         if (!sb_rdonly(sb))
2709                                 return 0;
2710                 }
2711                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2712                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2713                                "Inode bitmap for group %u not in group "
2714                                "(block %llu)!", i, inode_bitmap);
2715                         return 0;
2716                 }
2717                 inode_table = ext4_inode_table(sb, gdp);
2718                 if (inode_table == sb_block) {
2719                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2720                                  "Inode table for group %u overlaps "
2721                                  "superblock", i);
2722                         if (!sb_rdonly(sb))
2723                                 return 0;
2724                 }
2725                 if (inode_table >= sb_block + 1 &&
2726                     inode_table <= last_bg_block) {
2727                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2728                                  "Inode table for group %u overlaps "
2729                                  "block group descriptors", i);
2730                         if (!sb_rdonly(sb))
2731                                 return 0;
2732                 }
2733                 if (inode_table < first_block ||
2734                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2735                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2736                                "Inode table for group %u not in group "
2737                                "(block %llu)!", i, inode_table);
2738                         return 0;
2739                 }
2740                 ext4_lock_group(sb, i);
2741                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2742                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2743                                  "Checksum for group %u failed (%u!=%u)",
2744                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2745                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2746                         if (!sb_rdonly(sb)) {
2747                                 ext4_unlock_group(sb, i);
2748                                 return 0;
2749                         }
2750                 }
2751                 ext4_unlock_group(sb, i);
2752                 if (!flexbg_flag)
2753                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2754         }
2755         if (NULL != first_not_zeroed)
2756                 *first_not_zeroed = grp;
2757         return 1;
2758 }
2759
2760 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2761  * the superblock) which were deleted from all directories, but held open by
2762  * a process at the time of a crash.  We walk the list and try to delete these
2763  * inodes at recovery time (only with a read-write filesystem).
2764  *
2765  * In order to keep the orphan inode chain consistent during traversal (in
2766  * case of crash during recovery), we link each inode into the superblock
2767  * orphan list_head and handle it the same way as an inode deletion during
2768  * normal operation (which journals the operations for us).
2769  *
2770  * We only do an iget() and an iput() on each inode, which is very safe if we
2771  * accidentally point at an in-use or already deleted inode.  The worst that
2772  * can happen in this case is that we get a "bit already cleared" message from
2773  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2774  * e2fsck was run on this filesystem, and it must have already done the orphan
2775  * inode cleanup for us, so we can safely abort without any further action.
2776  */
2777 static void ext4_orphan_cleanup(struct super_block *sb,
2778                                 struct ext4_super_block *es)
2779 {
2780         unsigned int s_flags = sb->s_flags;
2781         int ret, nr_orphans = 0, nr_truncates = 0;
2782 #ifdef CONFIG_QUOTA
2783         int quota_update = 0;
2784         int i;
2785 #endif
2786         if (!es->s_last_orphan) {
2787                 jbd_debug(4, "no orphan inodes to clean up\n");
2788                 return;
2789         }
2790
2791         if (bdev_read_only(sb->s_bdev)) {
2792                 ext4_msg(sb, KERN_ERR, "write access "
2793                         "unavailable, skipping orphan cleanup");
2794                 return;
2795         }
2796
2797         /* Check if feature set would not allow a r/w mount */
2798         if (!ext4_feature_set_ok(sb, 0)) {
2799                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2800                          "unknown ROCOMPAT features");
2801                 return;
2802         }
2803
2804         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2805                 /* don't clear list on RO mount w/ errors */
2806                 if (es->s_last_orphan && !(s_flags & SB_RDONLY)) {
2807                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2808                                   "clearing orphan list.\n");
2809                         es->s_last_orphan = 0;
2810                 }
2811                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2812                 return;
2813         }
2814
2815         if (s_flags & SB_RDONLY) {
2816                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2817                 sb->s_flags &= ~SB_RDONLY;
2818         }
2819 #ifdef CONFIG_QUOTA
2820         /* Needed for iput() to work correctly and not trash data */
2821         sb->s_flags |= SB_ACTIVE;
2822
2823         /*
2824          * Turn on quotas which were not enabled for read-only mounts if
2825          * filesystem has quota feature, so that they are updated correctly.
2826          */
2827         if (ext4_has_feature_quota(sb) && (s_flags & SB_RDONLY)) {
2828                 int ret = ext4_enable_quotas(sb);
2829
2830                 if (!ret)
2831                         quota_update = 1;
2832                 else
2833                         ext4_msg(sb, KERN_ERR,
2834                                 "Cannot turn on quotas: error %d", ret);
2835         }
2836
2837         /* Turn on journaled quotas used for old sytle */
2838         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2839                 if (EXT4_SB(sb)->s_qf_names[i]) {
2840                         int ret = ext4_quota_on_mount(sb, i);
2841
2842                         if (!ret)
2843                                 quota_update = 1;
2844                         else
2845                                 ext4_msg(sb, KERN_ERR,
2846                                         "Cannot turn on journaled "
2847                                         "quota: type %d: error %d", i, ret);
2848                 }
2849         }
2850 #endif
2851
2852         while (es->s_last_orphan) {
2853                 struct inode *inode;
2854
2855                 /*
2856                  * We may have encountered an error during cleanup; if
2857                  * so, skip the rest.
2858                  */
2859                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2860                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2861                         es->s_last_orphan = 0;
2862                         break;
2863                 }
2864
2865                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2866                 if (IS_ERR(inode)) {
2867                         es->s_last_orphan = 0;
2868                         break;
2869                 }
2870
2871                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2872                 dquot_initialize(inode);
2873                 if (inode->i_nlink) {
2874                         if (test_opt(sb, DEBUG))
2875                                 ext4_msg(sb, KERN_DEBUG,
2876                                         "%s: truncating inode %lu to %lld bytes",
2877                                         __func__, inode->i_ino, inode->i_size);
2878                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2879                                   inode->i_ino, inode->i_size);
2880                         inode_lock(inode);
2881                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2882                         ret = ext4_truncate(inode);
2883                         if (ret)
2884                                 ext4_std_error(inode->i_sb, ret);
2885                         inode_unlock(inode);
2886                         nr_truncates++;
2887                 } else {
2888                         if (test_opt(sb, DEBUG))
2889                                 ext4_msg(sb, KERN_DEBUG,
2890                                         "%s: deleting unreferenced inode %lu",
2891                                         __func__, inode->i_ino);
2892                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2893                                   inode->i_ino);
2894                         nr_orphans++;
2895                 }
2896                 iput(inode);  /* The delete magic happens here! */
2897         }
2898
2899 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2900
2901         if (nr_orphans)
2902                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2903                        PLURAL(nr_orphans));
2904         if (nr_truncates)
2905                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2906                        PLURAL(nr_truncates));
2907 #ifdef CONFIG_QUOTA
2908         /* Turn off quotas if they were enabled for orphan cleanup */
2909         if (quota_update) {
2910                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2911                         if (sb_dqopt(sb)->files[i])
2912                                 dquot_quota_off(sb, i);
2913                 }
2914         }
2915 #endif
2916         sb->s_flags = s_flags; /* Restore SB_RDONLY status */
2917 }
2918
2919 /*
2920  * Maximal extent format file size.
2921  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2922  * extent format containers, within a sector_t, and within i_blocks
2923  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2924  * so that won't be a limiting factor.
2925  *
2926  * However there is other limiting factor. We do store extents in the form
2927  * of starting block and length, hence the resulting length of the extent
2928  * covering maximum file size must fit into on-disk format containers as
2929  * well. Given that length is always by 1 unit bigger than max unit (because
2930  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2931  *
2932  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2933  */
2934 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2935 {
2936         loff_t res;
2937         loff_t upper_limit = MAX_LFS_FILESIZE;
2938
2939         BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64));
2940
2941         if (!has_huge_files) {
2942                 upper_limit = (1LL << 32) - 1;
2943
2944                 /* total blocks in file system block size */
2945                 upper_limit >>= (blkbits - 9);
2946                 upper_limit <<= blkbits;
2947         }
2948
2949         /*
2950          * 32-bit extent-start container, ee_block. We lower the maxbytes
2951          * by one fs block, so ee_len can cover the extent of maximum file
2952          * size
2953          */
2954         res = (1LL << 32) - 1;
2955         res <<= blkbits;
2956
2957         /* Sanity check against vm- & vfs- imposed limits */
2958         if (res > upper_limit)
2959                 res = upper_limit;
2960
2961         return res;
2962 }
2963
2964 /*
2965  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2966  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2967  * We need to be 1 filesystem block less than the 2^48 sector limit.
2968  */
2969 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2970 {
2971         loff_t res = EXT4_NDIR_BLOCKS;
2972         int meta_blocks;
2973         loff_t upper_limit;
2974         /* This is calculated to be the largest file size for a dense, block
2975          * mapped file such that the file's total number of 512-byte sectors,
2976          * including data and all indirect blocks, does not exceed (2^48 - 1).
2977          *
2978          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2979          * number of 512-byte sectors of the file.
2980          */
2981
2982         if (!has_huge_files) {
2983                 /*
2984                  * !has_huge_files or implies that the inode i_block field
2985                  * represents total file blocks in 2^32 512-byte sectors ==
2986                  * size of vfs inode i_blocks * 8
2987                  */
2988                 upper_limit = (1LL << 32) - 1;
2989
2990                 /* total blocks in file system block size */
2991                 upper_limit >>= (bits - 9);
2992
2993         } else {
2994                 /*
2995                  * We use 48 bit ext4_inode i_blocks
2996                  * With EXT4_HUGE_FILE_FL set the i_blocks
2997                  * represent total number of blocks in
2998                  * file system block size
2999                  */
3000                 upper_limit = (1LL << 48) - 1;
3001
3002         }
3003
3004         /* indirect blocks */
3005         meta_blocks = 1;
3006         /* double indirect blocks */
3007         meta_blocks += 1 + (1LL << (bits-2));
3008         /* tripple indirect blocks */
3009         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
3010
3011         upper_limit -= meta_blocks;
3012         upper_limit <<= bits;
3013
3014         res += 1LL << (bits-2);
3015         res += 1LL << (2*(bits-2));
3016         res += 1LL << (3*(bits-2));
3017         res <<= bits;
3018         if (res > upper_limit)
3019                 res = upper_limit;
3020
3021         if (res > MAX_LFS_FILESIZE)
3022                 res = MAX_LFS_FILESIZE;
3023
3024         return res;
3025 }
3026
3027 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
3028                                    ext4_fsblk_t logical_sb_block, int nr)
3029 {
3030         struct ext4_sb_info *sbi = EXT4_SB(sb);
3031         ext4_group_t bg, first_meta_bg;
3032         int has_super = 0;
3033
3034         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
3035
3036         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
3037                 return logical_sb_block + nr + 1;
3038         bg = sbi->s_desc_per_block * nr;
3039         if (ext4_bg_has_super(sb, bg))
3040                 has_super = 1;
3041
3042         /*
3043          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
3044          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
3045          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
3046          * compensate.
3047          */
3048         if (sb->s_blocksize == 1024 && nr == 0 &&
3049             le32_to_cpu(sbi->s_es->s_first_data_block) == 0)
3050                 has_super++;
3051
3052         return (has_super + ext4_group_first_block_no(sb, bg));
3053 }
3054
3055 /**
3056  * ext4_get_stripe_size: Get the stripe size.
3057  * @sbi: In memory super block info
3058  *
3059  * If we have specified it via mount option, then
3060  * use the mount option value. If the value specified at mount time is
3061  * greater than the blocks per group use the super block value.
3062  * If the super block value is greater than blocks per group return 0.
3063  * Allocator needs it be less than blocks per group.
3064  *
3065  */
3066 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
3067 {
3068         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
3069         unsigned long stripe_width =
3070                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
3071         int ret;
3072
3073         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
3074                 ret = sbi->s_stripe;
3075         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
3076                 ret = stripe_width;
3077         else if (stride && stride <= sbi->s_blocks_per_group)
3078                 ret = stride;
3079         else
3080                 ret = 0;
3081
3082         /*
3083          * If the stripe width is 1, this makes no sense and
3084          * we set it to 0 to turn off stripe handling code.
3085          */
3086         if (ret <= 1)
3087                 ret = 0;
3088
3089         return ret;
3090 }
3091
3092 /*
3093  * Check whether this filesystem can be mounted based on
3094  * the features present and the RDONLY/RDWR mount requested.
3095  * Returns 1 if this filesystem can be mounted as requested,
3096  * 0 if it cannot be.
3097  */
3098 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
3099 {
3100         if (ext4_has_unknown_ext4_incompat_features(sb)) {
3101                 ext4_msg(sb, KERN_ERR,
3102                         "Couldn't mount because of "
3103                         "unsupported optional features (%x)",
3104                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
3105                         ~EXT4_FEATURE_INCOMPAT_SUPP));
3106                 return 0;
3107         }
3108
3109 #ifndef CONFIG_UNICODE
3110         if (ext4_has_feature_casefold(sb)) {
3111                 ext4_msg(sb, KERN_ERR,
3112                          "Filesystem with casefold feature cannot be "
3113                          "mounted without CONFIG_UNICODE");
3114                 return 0;
3115         }
3116 #endif
3117
3118         if (readonly)
3119                 return 1;
3120
3121         if (ext4_has_feature_readonly(sb)) {
3122                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
3123                 sb->s_flags |= SB_RDONLY;
3124                 return 1;
3125         }
3126
3127         /* Check that feature set is OK for a read-write mount */
3128         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
3129                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
3130                          "unsupported optional features (%x)",
3131                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
3132                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
3133                 return 0;
3134         }
3135         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
3136                 ext4_msg(sb, KERN_ERR,
3137                          "Can't support bigalloc feature without "
3138                          "extents feature\n");
3139                 return 0;
3140         }
3141
3142 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2)
3143         if (!readonly && (ext4_has_feature_quota(sb) ||
3144                           ext4_has_feature_project(sb))) {
3145                 ext4_msg(sb, KERN_ERR,
3146                          "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2");
3147                 return 0;
3148         }
3149 #endif  /* CONFIG_QUOTA */
3150         return 1;
3151 }
3152
3153 /*
3154  * This function is called once a day if we have errors logged
3155  * on the file system
3156  */
3157 static void print_daily_error_info(struct timer_list *t)
3158 {
3159         struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report);
3160         struct super_block *sb = sbi->s_sb;
3161         struct ext4_super_block *es = sbi->s_es;
3162
3163         if (es->s_error_count)
3164                 /* fsck newer than v1.41.13 is needed to clean this condition. */
3165                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
3166                          le32_to_cpu(es->s_error_count));
3167         if (es->s_first_error_time) {
3168                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d",
3169                        sb->s_id,
3170                        ext4_get_tstamp(es, s_first_error_time),
3171                        (int) sizeof(es->s_first_error_func),
3172                        es->s_first_error_func,
3173                        le32_to_cpu(es->s_first_error_line));
3174                 if (es->s_first_error_ino)
3175                         printk(KERN_CONT ": inode %u",
3176                                le32_to_cpu(es->s_first_error_ino));
3177                 if (es->s_first_error_block)
3178                         printk(KERN_CONT ": block %llu", (unsigned long long)
3179                                le64_to_cpu(es->s_first_error_block));
3180                 printk(KERN_CONT "\n");
3181         }
3182         if (es->s_last_error_time) {
3183                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d",
3184                        sb->s_id,
3185                        ext4_get_tstamp(es, s_last_error_time),
3186                        (int) sizeof(es->s_last_error_func),
3187                        es->s_last_error_func,
3188                        le32_to_cpu(es->s_last_error_line));
3189                 if (es->s_last_error_ino)
3190                         printk(KERN_CONT ": inode %u",
3191                                le32_to_cpu(es->s_last_error_ino));
3192                 if (es->s_last_error_block)
3193                         printk(KERN_CONT ": block %llu", (unsigned long long)
3194                                le64_to_cpu(es->s_last_error_block));
3195                 printk(KERN_CONT "\n");
3196         }
3197         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
3198 }
3199
3200 /* Find next suitable group and run ext4_init_inode_table */
3201 static int ext4_run_li_request(struct ext4_li_request *elr)
3202 {
3203         struct ext4_group_desc *gdp = NULL;
3204         ext4_group_t group, ngroups;
3205         struct super_block *sb;
3206         unsigned long timeout = 0;
3207         int ret = 0;
3208
3209         sb = elr->lr_super;
3210         ngroups = EXT4_SB(sb)->s_groups_count;
3211
3212         for (group = elr->lr_next_group; group < ngroups; group++) {
3213                 gdp = ext4_get_group_desc(sb, group, NULL);
3214                 if (!gdp) {
3215                         ret = 1;
3216                         break;
3217                 }
3218
3219                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3220                         break;
3221         }
3222
3223         if (group >= ngroups)
3224                 ret = 1;
3225
3226         if (!ret) {
3227                 timeout = jiffies;
3228                 ret = ext4_init_inode_table(sb, group,
3229                                             elr->lr_timeout ? 0 : 1);
3230                 if (elr->lr_timeout == 0) {
3231                         timeout = (jiffies - timeout) *
3232                                   elr->lr_sbi->s_li_wait_mult;
3233                         elr->lr_timeout = timeout;
3234                 }
3235                 elr->lr_next_sched = jiffies + elr->lr_timeout;
3236                 elr->lr_next_group = group + 1;
3237         }
3238         return ret;
3239 }
3240
3241 /*
3242  * Remove lr_request from the list_request and free the
3243  * request structure. Should be called with li_list_mtx held
3244  */
3245 static void ext4_remove_li_request(struct ext4_li_request *elr)
3246 {
3247         struct ext4_sb_info *sbi;
3248
3249         if (!elr)
3250                 return;
3251
3252         sbi = elr->lr_sbi;
3253
3254         list_del(&elr->lr_request);
3255         sbi->s_li_request = NULL;
3256         kfree(elr);
3257 }
3258
3259 static void ext4_unregister_li_request(struct super_block *sb)
3260 {
3261         mutex_lock(&ext4_li_mtx);
3262         if (!ext4_li_info) {
3263                 mutex_unlock(&ext4_li_mtx);
3264                 return;
3265         }
3266
3267         mutex_lock(&ext4_li_info->li_list_mtx);
3268         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
3269         mutex_unlock(&ext4_li_info->li_list_mtx);
3270         mutex_unlock(&ext4_li_mtx);
3271 }
3272
3273 static struct task_struct *ext4_lazyinit_task;
3274
3275 /*
3276  * This is the function where ext4lazyinit thread lives. It walks
3277  * through the request list searching for next scheduled filesystem.
3278  * When such a fs is found, run the lazy initialization request
3279  * (ext4_rn_li_request) and keep track of the time spend in this
3280  * function. Based on that time we compute next schedule time of
3281  * the request. When walking through the list is complete, compute
3282  * next waking time and put itself into sleep.
3283  */
3284 static int ext4_lazyinit_thread(void *arg)
3285 {
3286         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
3287         struct list_head *pos, *n;
3288         struct ext4_li_request *elr;
3289         unsigned long next_wakeup, cur;
3290
3291         BUG_ON(NULL == eli);
3292
3293 cont_thread:
3294         while (true) {
3295                 next_wakeup = MAX_JIFFY_OFFSET;
3296
3297                 mutex_lock(&eli->li_list_mtx);
3298                 if (list_empty(&eli->li_request_list)) {
3299                         mutex_unlock(&eli->li_list_mtx);
3300                         goto exit_thread;
3301                 }
3302                 list_for_each_safe(pos, n, &eli->li_request_list) {
3303                         int err = 0;
3304                         int progress = 0;
3305                         elr = list_entry(pos, struct ext4_li_request,
3306                                          lr_request);
3307
3308                         if (time_before(jiffies, elr->lr_next_sched)) {
3309                                 if (time_before(elr->lr_next_sched, next_wakeup))
3310                                         next_wakeup = elr->lr_next_sched;
3311                                 continue;
3312                         }
3313                         if (down_read_trylock(&elr->lr_super->s_umount)) {
3314                                 if (sb_start_write_trylock(elr->lr_super)) {
3315                                         progress = 1;
3316                                         /*
3317                                          * We hold sb->s_umount, sb can not
3318                                          * be removed from the list, it is
3319                                          * now safe to drop li_list_mtx
3320                                          */
3321                                         mutex_unlock(&eli->li_list_mtx);
3322                                         err = ext4_run_li_request(elr);
3323                                         sb_end_write(elr->lr_super);
3324                                         mutex_lock(&eli->li_list_mtx);
3325                                         n = pos->next;
3326                                 }
3327                                 up_read((&elr->lr_super->s_umount));
3328                         }
3329                         /* error, remove the lazy_init job */
3330                         if (err) {
3331                                 ext4_remove_li_request(elr);
3332                                 continue;
3333                         }
3334                         if (!progress) {
3335                                 elr->lr_next_sched = jiffies +
3336                                         (prandom_u32()
3337                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3338                         }
3339                         if (time_before(elr->lr_next_sched, next_wakeup))
3340                                 next_wakeup = elr->lr_next_sched;
3341                 }
3342                 mutex_unlock(&eli->li_list_mtx);
3343
3344                 try_to_freeze();
3345
3346                 cur = jiffies;
3347                 if ((time_after_eq(cur, next_wakeup)) ||
3348                     (MAX_JIFFY_OFFSET == next_wakeup)) {
3349                         cond_resched();
3350                         continue;
3351                 }
3352
3353                 schedule_timeout_interruptible(next_wakeup - cur);
3354
3355                 if (kthread_should_stop()) {
3356                         ext4_clear_request_list();
3357                         goto exit_thread;
3358                 }
3359         }
3360
3361 exit_thread:
3362         /*
3363          * It looks like the request list is empty, but we need
3364          * to check it under the li_list_mtx lock, to prevent any
3365          * additions into it, and of course we should lock ext4_li_mtx
3366          * to atomically free the list and ext4_li_info, because at
3367          * this point another ext4 filesystem could be registering
3368          * new one.
3369          */
3370         mutex_lock(&ext4_li_mtx);
3371         mutex_lock(&eli->li_list_mtx);
3372         if (!list_empty(&eli->li_request_list)) {
3373                 mutex_unlock(&eli->li_list_mtx);
3374                 mutex_unlock(&ext4_li_mtx);
3375                 goto cont_thread;
3376         }
3377         mutex_unlock(&eli->li_list_mtx);
3378         kfree(ext4_li_info);
3379         ext4_li_info = NULL;
3380         mutex_unlock(&ext4_li_mtx);
3381
3382         return 0;
3383 }
3384
3385 static void ext4_clear_request_list(void)
3386 {
3387         struct list_head *pos, *n;
3388         struct ext4_li_request *elr;
3389
3390         mutex_lock(&ext4_li_info->li_list_mtx);
3391         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3392                 elr = list_entry(pos, struct ext4_li_request,
3393                                  lr_request);
3394                 ext4_remove_li_request(elr);
3395         }
3396         mutex_unlock(&ext4_li_info->li_list_mtx);
3397 }
3398
3399 static int ext4_run_lazyinit_thread(void)
3400 {
3401         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3402                                          ext4_li_info, "ext4lazyinit");
3403         if (IS_ERR(ext4_lazyinit_task)) {
3404                 int err = PTR_ERR(ext4_lazyinit_task);
3405                 ext4_clear_request_list();
3406                 kfree(ext4_li_info);
3407                 ext4_li_info = NULL;
3408                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3409                                  "initialization thread\n",
3410                                  err);
3411                 return err;
3412         }
3413         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3414         return 0;
3415 }
3416
3417 /*
3418  * Check whether it make sense to run itable init. thread or not.
3419  * If there is at least one uninitialized inode table, return
3420  * corresponding group number, else the loop goes through all
3421  * groups and return total number of groups.
3422  */
3423 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3424 {
3425         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3426         struct ext4_group_desc *gdp = NULL;
3427
3428         if (!ext4_has_group_desc_csum(sb))
3429                 return ngroups;
3430
3431         for (group = 0; group < ngroups; group++) {
3432                 gdp = ext4_get_group_desc(sb, group, NULL);
3433                 if (!gdp)
3434                         continue;
3435
3436                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3437                         break;
3438         }
3439
3440         return group;
3441 }
3442
3443 static int ext4_li_info_new(void)
3444 {
3445         struct ext4_lazy_init *eli = NULL;
3446
3447         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3448         if (!eli)
3449                 return -ENOMEM;
3450
3451         INIT_LIST_HEAD(&eli->li_request_list);
3452         mutex_init(&eli->li_list_mtx);
3453
3454         eli->li_state |= EXT4_LAZYINIT_QUIT;
3455
3456         ext4_li_info = eli;
3457
3458         return 0;
3459 }
3460
3461 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3462                                             ext4_group_t start)
3463 {
3464         struct ext4_sb_info *sbi = EXT4_SB(sb);
3465         struct ext4_li_request *elr;
3466
3467         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3468         if (!elr)
3469                 return NULL;
3470
3471         elr->lr_super = sb;
3472         elr->lr_sbi = sbi;
3473         elr->lr_next_group = start;
3474
3475         /*
3476          * Randomize first schedule time of the request to
3477          * spread the inode table initialization requests
3478          * better.
3479          */
3480         elr->lr_next_sched = jiffies + (prandom_u32() %
3481                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3482         return elr;
3483 }
3484
3485 int ext4_register_li_request(struct super_block *sb,
3486                              ext4_group_t first_not_zeroed)
3487 {
3488         struct ext4_sb_info *sbi = EXT4_SB(sb);
3489         struct ext4_li_request *elr = NULL;
3490         ext4_group_t ngroups = sbi->s_groups_count;
3491         int ret = 0;
3492
3493         mutex_lock(&ext4_li_mtx);
3494         if (sbi->s_li_request != NULL) {
3495                 /*
3496                  * Reset timeout so it can be computed again, because
3497                  * s_li_wait_mult might have changed.
3498                  */
3499                 sbi->s_li_request->lr_timeout = 0;
3500                 goto out;
3501         }
3502
3503         if (first_not_zeroed == ngroups || sb_rdonly(sb) ||
3504             !test_opt(sb, INIT_INODE_TABLE))
3505                 goto out;
3506
3507         elr = ext4_li_request_new(sb, first_not_zeroed);
3508         if (!elr) {
3509                 ret = -ENOMEM;
3510                 goto out;
3511         }
3512
3513         if (NULL == ext4_li_info) {
3514                 ret = ext4_li_info_new();
3515                 if (ret)
3516                         goto out;
3517         }
3518
3519         mutex_lock(&ext4_li_info->li_list_mtx);
3520         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3521         mutex_unlock(&ext4_li_info->li_list_mtx);
3522
3523         sbi->s_li_request = elr;
3524         /*
3525          * set elr to NULL here since it has been inserted to
3526          * the request_list and the removal and free of it is
3527          * handled by ext4_clear_request_list from now on.
3528          */
3529         elr = NULL;
3530
3531         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3532                 ret = ext4_run_lazyinit_thread();
3533                 if (ret)
3534                         goto out;
3535         }
3536 out:
3537         mutex_unlock(&ext4_li_mtx);
3538         if (ret)
3539                 kfree(elr);
3540         return ret;
3541 }
3542
3543 /*
3544  * We do not need to lock anything since this is called on
3545  * module unload.
3546  */
3547 static void ext4_destroy_lazyinit_thread(void)
3548 {
3549         /*
3550          * If thread exited earlier
3551          * there's nothing to be done.
3552          */
3553         if (!ext4_li_info || !ext4_lazyinit_task)
3554                 return;
3555
3556         kthread_stop(ext4_lazyinit_task);
3557 }
3558
3559 static int set_journal_csum_feature_set(struct super_block *sb)
3560 {
3561         int ret = 1;
3562         int compat, incompat;
3563         struct ext4_sb_info *sbi = EXT4_SB(sb);
3564
3565         if (ext4_has_metadata_csum(sb)) {
3566                 /* journal checksum v3 */
3567                 compat = 0;
3568                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3569         } else {
3570                 /* journal checksum v1 */
3571                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3572                 incompat = 0;
3573         }
3574
3575         jbd2_journal_clear_features(sbi->s_journal,
3576                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3577                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3578                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3579         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3580                 ret = jbd2_journal_set_features(sbi->s_journal,
3581                                 compat, 0,
3582                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3583                                 incompat);
3584         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3585                 ret = jbd2_journal_set_features(sbi->s_journal,
3586                                 compat, 0,
3587                                 incompat);
3588                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3589                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3590         } else {
3591                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3592                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3593         }
3594
3595         return ret;
3596 }
3597
3598 /*
3599  * Note: calculating the overhead so we can be compatible with
3600  * historical BSD practice is quite difficult in the face of
3601  * clusters/bigalloc.  This is because multiple metadata blocks from
3602  * different block group can end up in the same allocation cluster.
3603  * Calculating the exact overhead in the face of clustered allocation
3604  * requires either O(all block bitmaps) in memory or O(number of block
3605  * groups**2) in time.  We will still calculate the superblock for
3606  * older file systems --- and if we come across with a bigalloc file
3607  * system with zero in s_overhead_clusters the estimate will be close to
3608  * correct especially for very large cluster sizes --- but for newer
3609  * file systems, it's better to calculate this figure once at mkfs
3610  * time, and store it in the superblock.  If the superblock value is
3611  * present (even for non-bigalloc file systems), we will use it.
3612  */
3613 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3614                           char *buf)
3615 {
3616         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3617         struct ext4_group_desc  *gdp;
3618         ext4_fsblk_t            first_block, last_block, b;
3619         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3620         int                     s, j, count = 0;
3621
3622         if (!ext4_has_feature_bigalloc(sb))
3623                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3624                         sbi->s_itb_per_group + 2);
3625
3626         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3627                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3628         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3629         for (i = 0; i < ngroups; i++) {
3630                 gdp = ext4_get_group_desc(sb, i, NULL);
3631                 b = ext4_block_bitmap(sb, gdp);
3632                 if (b >= first_block && b <= last_block) {
3633                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3634                         count++;
3635                 }
3636                 b = ext4_inode_bitmap(sb, gdp);
3637                 if (b >= first_block && b <= last_block) {
3638                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3639                         count++;
3640                 }
3641                 b = ext4_inode_table(sb, gdp);
3642                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3643                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3644                                 int c = EXT4_B2C(sbi, b - first_block);
3645                                 ext4_set_bit(c, buf);
3646                                 count++;
3647                         }
3648                 if (i != grp)
3649                         continue;
3650                 s = 0;
3651                 if (ext4_bg_has_super(sb, grp)) {
3652                         ext4_set_bit(s++, buf);
3653                         count++;
3654                 }
3655                 j = ext4_bg_num_gdb(sb, grp);
3656                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3657                         ext4_error(sb, "Invalid number of block group "
3658                                    "descriptor blocks: %d", j);
3659                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3660                 }
3661                 count += j;
3662                 for (; j > 0; j--)
3663                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3664         }
3665         if (!count)
3666                 return 0;
3667         return EXT4_CLUSTERS_PER_GROUP(sb) -
3668                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3669 }
3670
3671 /*
3672  * Compute the overhead and stash it in sbi->s_overhead
3673  */
3674 int ext4_calculate_overhead(struct super_block *sb)
3675 {
3676         struct ext4_sb_info *sbi = EXT4_SB(sb);
3677         struct ext4_super_block *es = sbi->s_es;
3678         struct inode *j_inode;
3679         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3680         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3681         ext4_fsblk_t overhead = 0;
3682         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3683
3684         if (!buf)
3685                 return -ENOMEM;
3686
3687         /*
3688          * Compute the overhead (FS structures).  This is constant
3689          * for a given filesystem unless the number of block groups
3690          * changes so we cache the previous value until it does.
3691          */
3692
3693         /*
3694          * All of the blocks before first_data_block are overhead
3695          */
3696         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3697
3698         /*
3699          * Add the overhead found in each block group
3700          */
3701         for (i = 0; i < ngroups; i++) {
3702                 int blks;
3703
3704                 blks = count_overhead(sb, i, buf);
3705                 overhead += blks;
3706                 if (blks)
3707                         memset(buf, 0, PAGE_SIZE);
3708                 cond_resched();
3709         }
3710
3711         /*
3712          * Add the internal journal blocks whether the journal has been
3713          * loaded or not
3714          */
3715         if (sbi->s_journal && !sbi->journal_bdev)
3716                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3717         else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) {
3718                 /* j_inum for internal journal is non-zero */
3719                 j_inode = ext4_get_journal_inode(sb, j_inum);
3720                 if (j_inode) {
3721                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3722                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
3723                         iput(j_inode);
3724                 } else {
3725                         ext4_msg(sb, KERN_ERR, "can't get journal size");
3726                 }
3727         }
3728         sbi->s_overhead = overhead;
3729         smp_wmb();
3730         free_page((unsigned long) buf);
3731         return 0;
3732 }
3733
3734 static void ext4_set_resv_clusters(struct super_block *sb)
3735 {
3736         ext4_fsblk_t resv_clusters;
3737         struct ext4_sb_info *sbi = EXT4_SB(sb);
3738
3739         /*
3740          * There's no need to reserve anything when we aren't using extents.
3741          * The space estimates are exact, there are no unwritten extents,
3742          * hole punching doesn't need new metadata... This is needed especially
3743          * to keep ext2/3 backward compatibility.
3744          */
3745         if (!ext4_has_feature_extents(sb))
3746                 return;
3747         /*
3748          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3749          * This should cover the situations where we can not afford to run
3750          * out of space like for example punch hole, or converting
3751          * unwritten extents in delalloc path. In most cases such
3752          * allocation would require 1, or 2 blocks, higher numbers are
3753          * very rare.
3754          */
3755         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3756                          sbi->s_cluster_bits);
3757
3758         do_div(resv_clusters, 50);
3759         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3760
3761         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3762 }
3763
3764 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3765 {
3766         struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev);
3767         char *orig_data = kstrdup(data, GFP_KERNEL);
3768         struct buffer_head *bh, **group_desc;
3769         struct ext4_super_block *es = NULL;
3770         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3771         struct flex_groups **flex_groups;
3772         ext4_fsblk_t block;
3773         ext4_fsblk_t sb_block = get_sb_block(&data);
3774         ext4_fsblk_t logical_sb_block;
3775         unsigned long offset = 0;
3776         unsigned long journal_devnum = 0;
3777         unsigned long def_mount_opts;
3778         struct inode *root;
3779         const char *descr;
3780         int ret = -ENOMEM;
3781         int blocksize, clustersize;
3782         unsigned int db_count;
3783         unsigned int i;
3784         int needs_recovery, has_huge_files;
3785         __u64 blocks_count;
3786         int err = 0;
3787         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3788         ext4_group_t first_not_zeroed;
3789
3790         if ((data && !orig_data) || !sbi)
3791                 goto out_free_base;
3792
3793         sbi->s_daxdev = dax_dev;
3794         sbi->s_blockgroup_lock =
3795                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3796         if (!sbi->s_blockgroup_lock)
3797                 goto out_free_base;
3798
3799         sb->s_fs_info = sbi;
3800         sbi->s_sb = sb;
3801         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3802         sbi->s_sb_block = sb_block;
3803         if (sb->s_bdev->bd_part)
3804                 sbi->s_sectors_written_start =
3805                         part_stat_read(sb->s_bdev->bd_part, sectors[STAT_WRITE]);
3806
3807         /* Cleanup superblock name */
3808         strreplace(sb->s_id, '/', '!');
3809
3810         /* -EINVAL is default */
3811         ret = -EINVAL;
3812         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3813         if (!blocksize) {
3814                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3815                 goto out_fail;
3816         }
3817
3818         /*
3819          * The ext4 superblock will not be buffer aligned for other than 1kB
3820          * block sizes.  We need to calculate the offset from buffer start.
3821          */
3822         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3823                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3824                 offset = do_div(logical_sb_block, blocksize);
3825         } else {
3826                 logical_sb_block = sb_block;
3827         }
3828
3829         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3830                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3831                 goto out_fail;
3832         }
3833         /*
3834          * Note: s_es must be initialized as soon as possible because
3835          *       some ext4 macro-instructions depend on its value
3836          */
3837         es = (struct ext4_super_block *) (bh->b_data + offset);
3838         sbi->s_es = es;
3839         sb->s_magic = le16_to_cpu(es->s_magic);
3840         if (sb->s_magic != EXT4_SUPER_MAGIC)
3841                 goto cantfind_ext4;
3842         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3843
3844         /* Warn if metadata_csum and gdt_csum are both set. */
3845         if (ext4_has_feature_metadata_csum(sb) &&
3846             ext4_has_feature_gdt_csum(sb))
3847                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3848                              "redundant flags; please run fsck.");
3849
3850         /* Check for a known checksum algorithm */
3851         if (!ext4_verify_csum_type(sb, es)) {
3852                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3853                          "unknown checksum algorithm.");
3854                 silent = 1;
3855                 goto cantfind_ext4;
3856         }
3857
3858         /* Load the checksum driver */
3859         sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3860         if (IS_ERR(sbi->s_chksum_driver)) {
3861                 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3862                 ret = PTR_ERR(sbi->s_chksum_driver);
3863                 sbi->s_chksum_driver = NULL;
3864                 goto failed_mount;
3865         }
3866
3867         /* Check superblock checksum */
3868         if (!ext4_superblock_csum_verify(sb, es)) {
3869                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3870                          "invalid superblock checksum.  Run e2fsck?");
3871                 silent = 1;
3872                 ret = -EFSBADCRC;
3873                 goto cantfind_ext4;
3874         }
3875
3876         /* Precompute checksum seed for all metadata */
3877         if (ext4_has_feature_csum_seed(sb))
3878                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3879         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
3880                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3881                                                sizeof(es->s_uuid));
3882
3883         /* Set defaults before we parse the mount options */
3884         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3885         set_opt(sb, INIT_INODE_TABLE);
3886         if (def_mount_opts & EXT4_DEFM_DEBUG)
3887                 set_opt(sb, DEBUG);
3888         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3889                 set_opt(sb, GRPID);
3890         if (def_mount_opts & EXT4_DEFM_UID16)
3891                 set_opt(sb, NO_UID32);
3892         /* xattr user namespace & acls are now defaulted on */
3893         set_opt(sb, XATTR_USER);
3894 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3895         set_opt(sb, POSIX_ACL);
3896 #endif
3897         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3898         if (ext4_has_metadata_csum(sb))
3899                 set_opt(sb, JOURNAL_CHECKSUM);
3900
3901         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3902                 set_opt(sb, JOURNAL_DATA);
3903         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3904                 set_opt(sb, ORDERED_DATA);
3905         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3906                 set_opt(sb, WRITEBACK_DATA);
3907
3908         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3909                 set_opt(sb, ERRORS_PANIC);
3910         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3911                 set_opt(sb, ERRORS_CONT);
3912         else
3913                 set_opt(sb, ERRORS_RO);
3914         /* block_validity enabled by default; disable with noblock_validity */
3915         set_opt(sb, BLOCK_VALIDITY);
3916         if (def_mount_opts & EXT4_DEFM_DISCARD)
3917                 set_opt(sb, DISCARD);
3918
3919         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3920         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3921         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3922         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3923         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3924
3925         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3926                 set_opt(sb, BARRIER);
3927
3928         /*
3929          * enable delayed allocation by default
3930          * Use -o nodelalloc to turn it off
3931          */
3932         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3933             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3934                 set_opt(sb, DELALLOC);
3935
3936         /*
3937          * set default s_li_wait_mult for lazyinit, for the case there is
3938          * no mount option specified.
3939          */
3940         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3941
3942         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3943
3944         if (blocksize == PAGE_SIZE)
3945                 set_opt(sb, DIOREAD_NOLOCK);
3946
3947         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3948             blocksize > EXT4_MAX_BLOCK_SIZE) {
3949                 ext4_msg(sb, KERN_ERR,
3950                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3951                          blocksize, le32_to_cpu(es->s_log_block_size));
3952                 goto failed_mount;
3953         }
3954
3955         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3956                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3957                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3958         } else {
3959                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3960                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3961                 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) {
3962                         ext4_msg(sb, KERN_ERR, "invalid first ino: %u",
3963                                  sbi->s_first_ino);
3964                         goto failed_mount;
3965                 }
3966                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3967                     (!is_power_of_2(sbi->s_inode_size)) ||
3968                     (sbi->s_inode_size > blocksize)) {
3969                         ext4_msg(sb, KERN_ERR,
3970                                "unsupported inode size: %d",
3971                                sbi->s_inode_size);
3972                         ext4_msg(sb, KERN_ERR, "blocksize: %d", blocksize);
3973                         goto failed_mount;
3974                 }
3975                 /*
3976                  * i_atime_extra is the last extra field available for
3977                  * [acm]times in struct ext4_inode. Checking for that
3978                  * field should suffice to ensure we have extra space
3979                  * for all three.
3980                  */
3981                 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) +
3982                         sizeof(((struct ext4_inode *)0)->i_atime_extra)) {
3983                         sb->s_time_gran = 1;
3984                         sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX;
3985                 } else {
3986                         sb->s_time_gran = NSEC_PER_SEC;
3987                         sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX;
3988                 }
3989                 sb->s_time_min = EXT4_TIMESTAMP_MIN;
3990         }
3991         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3992                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3993                         EXT4_GOOD_OLD_INODE_SIZE;
3994                 if (ext4_has_feature_extra_isize(sb)) {
3995                         unsigned v, max = (sbi->s_inode_size -
3996                                            EXT4_GOOD_OLD_INODE_SIZE);
3997
3998                         v = le16_to_cpu(es->s_want_extra_isize);
3999                         if (v > max) {
4000                                 ext4_msg(sb, KERN_ERR,
4001                                          "bad s_want_extra_isize: %d", v);
4002                                 goto failed_mount;
4003                         }
4004                         if (sbi->s_want_extra_isize < v)
4005                                 sbi->s_want_extra_isize = v;
4006
4007                         v = le16_to_cpu(es->s_min_extra_isize);
4008                         if (v > max) {
4009                                 ext4_msg(sb, KERN_ERR,
4010                                          "bad s_min_extra_isize: %d", v);
4011                                 goto failed_mount;
4012                         }
4013                         if (sbi->s_want_extra_isize < v)
4014                                 sbi->s_want_extra_isize = v;
4015                 }
4016         }
4017
4018         if (sbi->s_es->s_mount_opts[0]) {
4019                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
4020                                               sizeof(sbi->s_es->s_mount_opts),
4021                                               GFP_KERNEL);
4022                 if (!s_mount_opts)
4023                         goto failed_mount;
4024                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
4025                                    &journal_ioprio, 0)) {
4026                         ext4_msg(sb, KERN_WARNING,
4027                                  "failed to parse options in superblock: %s",
4028                                  s_mount_opts);
4029                 }
4030                 kfree(s_mount_opts);
4031         }
4032         sbi->s_def_mount_opt = sbi->s_mount_opt;
4033         if (!parse_options((char *) data, sb, &journal_devnum,
4034                            &journal_ioprio, 0))
4035                 goto failed_mount;
4036
4037 #ifdef CONFIG_UNICODE
4038         if (ext4_has_feature_casefold(sb) && !sbi->s_encoding) {
4039                 const struct ext4_sb_encodings *encoding_info;
4040                 struct unicode_map *encoding;
4041                 __u16 encoding_flags;
4042
4043                 if (ext4_has_feature_encrypt(sb)) {
4044                         ext4_msg(sb, KERN_ERR,
4045                                  "Can't mount with encoding and encryption");
4046                         goto failed_mount;
4047                 }
4048
4049                 if (ext4_sb_read_encoding(es, &encoding_info,
4050                                           &encoding_flags)) {
4051                         ext4_msg(sb, KERN_ERR,
4052                                  "Encoding requested by superblock is unknown");
4053                         goto failed_mount;
4054                 }
4055
4056                 encoding = utf8_load(encoding_info->version);
4057                 if (IS_ERR(encoding)) {
4058                         ext4_msg(sb, KERN_ERR,
4059                                  "can't mount with superblock charset: %s-%s "
4060                                  "not supported by the kernel. flags: 0x%x.",
4061                                  encoding_info->name, encoding_info->version,
4062                                  encoding_flags);
4063                         goto failed_mount;
4064                 }
4065                 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: "
4066                          "%s-%s with flags 0x%hx", encoding_info->name,
4067                          encoding_info->version?:"\b", encoding_flags);
4068
4069                 sbi->s_encoding = encoding;
4070                 sbi->s_encoding_flags = encoding_flags;
4071         }
4072 #endif
4073
4074         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4075                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with data=journal disables delayed allocation, dioread_nolock, and O_DIRECT support!\n");
4076                 /* can't mount with both data=journal and dioread_nolock. */
4077                 clear_opt(sb, DIOREAD_NOLOCK);
4078                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4079                         ext4_msg(sb, KERN_ERR, "can't mount with "
4080                                  "both data=journal and delalloc");
4081                         goto failed_mount;
4082                 }
4083                 if (test_opt(sb, DAX_ALWAYS)) {
4084                         ext4_msg(sb, KERN_ERR, "can't mount with "
4085                                  "both data=journal and dax");
4086                         goto failed_mount;
4087                 }
4088                 if (ext4_has_feature_encrypt(sb)) {
4089                         ext4_msg(sb, KERN_WARNING,
4090                                  "encrypted files will use data=ordered "
4091                                  "instead of data journaling mode");
4092                 }
4093                 if (test_opt(sb, DELALLOC))
4094                         clear_opt(sb, DELALLOC);
4095         } else {
4096                 sb->s_iflags |= SB_I_CGROUPWB;
4097         }
4098
4099         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
4100                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
4101
4102         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
4103             (ext4_has_compat_features(sb) ||
4104              ext4_has_ro_compat_features(sb) ||
4105              ext4_has_incompat_features(sb)))
4106                 ext4_msg(sb, KERN_WARNING,
4107                        "feature flags set on rev 0 fs, "
4108                        "running e2fsck is recommended");
4109
4110         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
4111                 set_opt2(sb, HURD_COMPAT);
4112                 if (ext4_has_feature_64bit(sb)) {
4113                         ext4_msg(sb, KERN_ERR,
4114                                  "The Hurd can't support 64-bit file systems");
4115                         goto failed_mount;
4116                 }
4117
4118                 /*
4119                  * ea_inode feature uses l_i_version field which is not
4120                  * available in HURD_COMPAT mode.
4121                  */
4122                 if (ext4_has_feature_ea_inode(sb)) {
4123                         ext4_msg(sb, KERN_ERR,
4124                                  "ea_inode feature is not supported for Hurd");
4125                         goto failed_mount;
4126                 }
4127         }
4128
4129         if (IS_EXT2_SB(sb)) {
4130                 if (ext2_feature_set_ok(sb))
4131                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
4132                                  "using the ext4 subsystem");
4133                 else {
4134                         /*
4135                          * If we're probing be silent, if this looks like
4136                          * it's actually an ext[34] filesystem.
4137                          */
4138                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4139                                 goto failed_mount;
4140                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
4141                                  "to feature incompatibilities");
4142                         goto failed_mount;
4143                 }
4144         }
4145
4146         if (IS_EXT3_SB(sb)) {
4147                 if (ext3_feature_set_ok(sb))
4148                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
4149                                  "using the ext4 subsystem");
4150                 else {
4151                         /*
4152                          * If we're probing be silent, if this looks like
4153                          * it's actually an ext4 filesystem.
4154                          */
4155                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4156                                 goto failed_mount;
4157                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
4158                                  "to feature incompatibilities");
4159                         goto failed_mount;
4160                 }
4161         }
4162
4163         /*
4164          * Check feature flags regardless of the revision level, since we
4165          * previously didn't change the revision level when setting the flags,
4166          * so there is a chance incompat flags are set on a rev 0 filesystem.
4167          */
4168         if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
4169                 goto failed_mount;
4170
4171         if (le32_to_cpu(es->s_log_block_size) >
4172             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4173                 ext4_msg(sb, KERN_ERR,
4174                          "Invalid log block size: %u",
4175                          le32_to_cpu(es->s_log_block_size));
4176                 goto failed_mount;
4177         }
4178         if (le32_to_cpu(es->s_log_cluster_size) >
4179             (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4180                 ext4_msg(sb, KERN_ERR,
4181                          "Invalid log cluster size: %u",
4182                          le32_to_cpu(es->s_log_cluster_size));
4183                 goto failed_mount;
4184         }
4185
4186         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
4187                 ext4_msg(sb, KERN_ERR,
4188                          "Number of reserved GDT blocks insanely large: %d",
4189                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
4190                 goto failed_mount;
4191         }
4192
4193         if (bdev_dax_supported(sb->s_bdev, blocksize))
4194                 set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags);
4195
4196         if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) {
4197                 if (ext4_has_feature_inline_data(sb)) {
4198                         ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
4199                                         " that may contain inline data");
4200                         goto failed_mount;
4201                 }
4202                 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) {
4203                         ext4_msg(sb, KERN_ERR,
4204                                 "DAX unsupported by block device.");
4205                         goto failed_mount;
4206                 }
4207         }
4208
4209         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
4210                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
4211                          es->s_encryption_level);
4212                 goto failed_mount;
4213         }
4214
4215         if (sb->s_blocksize != blocksize) {
4216                 /* Validate the filesystem blocksize */
4217                 if (!sb_set_blocksize(sb, blocksize)) {
4218                         ext4_msg(sb, KERN_ERR, "bad block size %d",
4219                                         blocksize);
4220                         goto failed_mount;
4221                 }
4222
4223                 brelse(bh);
4224                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
4225                 offset = do_div(logical_sb_block, blocksize);
4226                 bh = sb_bread_unmovable(sb, logical_sb_block);
4227                 if (!bh) {
4228                         ext4_msg(sb, KERN_ERR,
4229                                "Can't read superblock on 2nd try");
4230                         goto failed_mount;
4231                 }
4232                 es = (struct ext4_super_block *)(bh->b_data + offset);
4233                 sbi->s_es = es;
4234                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
4235                         ext4_msg(sb, KERN_ERR,
4236                                "Magic mismatch, very weird!");
4237                         goto failed_mount;
4238                 }
4239         }
4240
4241         has_huge_files = ext4_has_feature_huge_file(sb);
4242         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
4243                                                       has_huge_files);
4244         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
4245
4246         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
4247         if (ext4_has_feature_64bit(sb)) {
4248                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
4249                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
4250                     !is_power_of_2(sbi->s_desc_size)) {
4251                         ext4_msg(sb, KERN_ERR,
4252                                "unsupported descriptor size %lu",
4253                                sbi->s_desc_size);
4254                         goto failed_mount;
4255                 }
4256         } else
4257                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
4258
4259         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
4260         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
4261
4262         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
4263         if (sbi->s_inodes_per_block == 0)
4264                 goto cantfind_ext4;
4265         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
4266             sbi->s_inodes_per_group > blocksize * 8) {
4267                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
4268                          sbi->s_inodes_per_group);
4269                 goto failed_mount;
4270         }
4271         sbi->s_itb_per_group = sbi->s_inodes_per_group /
4272                                         sbi->s_inodes_per_block;
4273         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
4274         sbi->s_sbh = bh;
4275         sbi->s_mount_state = le16_to_cpu(es->s_state);
4276         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
4277         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
4278
4279         for (i = 0; i < 4; i++)
4280                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
4281         sbi->s_def_hash_version = es->s_def_hash_version;
4282         if (ext4_has_feature_dir_index(sb)) {
4283                 i = le32_to_cpu(es->s_flags);
4284                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
4285                         sbi->s_hash_unsigned = 3;
4286                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
4287 #ifdef __CHAR_UNSIGNED__
4288                         if (!sb_rdonly(sb))
4289                                 es->s_flags |=
4290                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
4291                         sbi->s_hash_unsigned = 3;
4292 #else
4293                         if (!sb_rdonly(sb))
4294                                 es->s_flags |=
4295                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
4296 #endif
4297                 }
4298         }
4299
4300         /* Handle clustersize */
4301         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
4302         if (ext4_has_feature_bigalloc(sb)) {
4303                 if (clustersize < blocksize) {
4304                         ext4_msg(sb, KERN_ERR,
4305                                  "cluster size (%d) smaller than "
4306                                  "block size (%d)", clustersize, blocksize);
4307                         goto failed_mount;
4308                 }
4309                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
4310                         le32_to_cpu(es->s_log_block_size);
4311                 sbi->s_clusters_per_group =
4312                         le32_to_cpu(es->s_clusters_per_group);
4313                 if (sbi->s_clusters_per_group > blocksize * 8) {
4314                         ext4_msg(sb, KERN_ERR,
4315                                  "#clusters per group too big: %lu",
4316                                  sbi->s_clusters_per_group);
4317                         goto failed_mount;
4318                 }
4319                 if (sbi->s_blocks_per_group !=
4320                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
4321                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
4322                                  "clusters per group (%lu) inconsistent",
4323                                  sbi->s_blocks_per_group,
4324                                  sbi->s_clusters_per_group);
4325                         goto failed_mount;
4326                 }
4327         } else {
4328                 if (clustersize != blocksize) {
4329                         ext4_msg(sb, KERN_ERR,
4330                                  "fragment/cluster size (%d) != "
4331                                  "block size (%d)", clustersize, blocksize);
4332                         goto failed_mount;
4333                 }
4334                 if (sbi->s_blocks_per_group > blocksize * 8) {
4335                         ext4_msg(sb, KERN_ERR,
4336                                  "#blocks per group too big: %lu",
4337                                  sbi->s_blocks_per_group);
4338                         goto failed_mount;
4339                 }
4340                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
4341                 sbi->s_cluster_bits = 0;
4342         }
4343         sbi->s_cluster_ratio = clustersize / blocksize;
4344
4345         /* Do we have standard group size of clustersize * 8 blocks ? */
4346         if (sbi->s_blocks_per_group == clustersize << 3)
4347                 set_opt2(sb, STD_GROUP_SIZE);
4348
4349         /*
4350          * Test whether we have more sectors than will fit in sector_t,
4351          * and whether the max offset is addressable by the page cache.
4352          */
4353         err = generic_check_addressable(sb->s_blocksize_bits,
4354                                         ext4_blocks_count(es));
4355         if (err) {
4356                 ext4_msg(sb, KERN_ERR, "filesystem"
4357                          " too large to mount safely on this system");
4358                 goto failed_mount;
4359         }
4360
4361         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
4362                 goto cantfind_ext4;
4363
4364         /* check blocks count against device size */
4365         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
4366         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
4367                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
4368                        "exceeds size of device (%llu blocks)",
4369                        ext4_blocks_count(es), blocks_count);
4370                 goto failed_mount;
4371         }
4372
4373         /*
4374          * It makes no sense for the first data block to be beyond the end
4375          * of the filesystem.
4376          */
4377         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
4378                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4379                          "block %u is beyond end of filesystem (%llu)",
4380                          le32_to_cpu(es->s_first_data_block),
4381                          ext4_blocks_count(es));
4382                 goto failed_mount;
4383         }
4384         if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) &&
4385             (sbi->s_cluster_ratio == 1)) {
4386                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4387                          "block is 0 with a 1k block and cluster size");
4388                 goto failed_mount;
4389         }
4390
4391         blocks_count = (ext4_blocks_count(es) -
4392                         le32_to_cpu(es->s_first_data_block) +
4393                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
4394         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
4395         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
4396                 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu "
4397                        "(block count %llu, first data block %u, "
4398                        "blocks per group %lu)", blocks_count,
4399                        ext4_blocks_count(es),
4400                        le32_to_cpu(es->s_first_data_block),
4401                        EXT4_BLOCKS_PER_GROUP(sb));
4402                 goto failed_mount;
4403         }
4404         sbi->s_groups_count = blocks_count;
4405         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
4406                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
4407         if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) !=
4408             le32_to_cpu(es->s_inodes_count)) {
4409                 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu",
4410                          le32_to_cpu(es->s_inodes_count),
4411                          ((u64)sbi->s_groups_count * sbi->s_inodes_per_group));
4412                 ret = -EINVAL;
4413                 goto failed_mount;
4414         }
4415         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
4416                    EXT4_DESC_PER_BLOCK(sb);
4417         if (ext4_has_feature_meta_bg(sb)) {
4418                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
4419                         ext4_msg(sb, KERN_WARNING,
4420                                  "first meta block group too large: %u "
4421                                  "(group descriptor block count %u)",
4422                                  le32_to_cpu(es->s_first_meta_bg), db_count);
4423                         goto failed_mount;
4424                 }
4425         }
4426         rcu_assign_pointer(sbi->s_group_desc,
4427                            kvmalloc_array(db_count,
4428                                           sizeof(struct buffer_head *),
4429                                           GFP_KERNEL));
4430         if (sbi->s_group_desc == NULL) {
4431                 ext4_msg(sb, KERN_ERR, "not enough memory");
4432                 ret = -ENOMEM;
4433                 goto failed_mount;
4434         }
4435
4436         bgl_lock_init(sbi->s_blockgroup_lock);
4437
4438         /* Pre-read the descriptors into the buffer cache */
4439         for (i = 0; i < db_count; i++) {
4440                 block = descriptor_loc(sb, logical_sb_block, i);
4441                 sb_breadahead_unmovable(sb, block);
4442         }
4443
4444         for (i = 0; i < db_count; i++) {
4445                 struct buffer_head *bh;
4446
4447                 block = descriptor_loc(sb, logical_sb_block, i);
4448                 bh = sb_bread_unmovable(sb, block);
4449                 if (!bh) {
4450                         ext4_msg(sb, KERN_ERR,
4451                                "can't read group descriptor %d", i);
4452                         db_count = i;
4453                         goto failed_mount2;
4454                 }
4455                 rcu_read_lock();
4456                 rcu_dereference(sbi->s_group_desc)[i] = bh;
4457                 rcu_read_unlock();
4458         }
4459         sbi->s_gdb_count = db_count;
4460         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
4461                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
4462                 ret = -EFSCORRUPTED;
4463                 goto failed_mount2;
4464         }
4465
4466         timer_setup(&sbi->s_err_report, print_daily_error_info, 0);
4467
4468         /* Register extent status tree shrinker */
4469         if (ext4_es_register_shrinker(sbi))
4470                 goto failed_mount3;
4471
4472         sbi->s_stripe = ext4_get_stripe_size(sbi);
4473         sbi->s_extent_max_zeroout_kb = 32;
4474
4475         /*
4476          * set up enough so that it can read an inode
4477          */
4478         sb->s_op = &ext4_sops;
4479         sb->s_export_op = &ext4_export_ops;
4480         sb->s_xattr = ext4_xattr_handlers;
4481 #ifdef CONFIG_FS_ENCRYPTION
4482         sb->s_cop = &ext4_cryptops;
4483 #endif
4484 #ifdef CONFIG_FS_VERITY
4485         sb->s_vop = &ext4_verityops;
4486 #endif
4487 #ifdef CONFIG_QUOTA
4488         sb->dq_op = &ext4_quota_operations;
4489         if (ext4_has_feature_quota(sb))
4490                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4491         else
4492                 sb->s_qcop = &ext4_qctl_operations;
4493         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
4494 #endif
4495         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
4496
4497         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
4498         mutex_init(&sbi->s_orphan_lock);
4499
4500         sb->s_root = NULL;
4501
4502         needs_recovery = (es->s_last_orphan != 0 ||
4503                           ext4_has_feature_journal_needs_recovery(sb));
4504
4505         if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb))
4506                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
4507                         goto failed_mount3a;
4508
4509         /*
4510          * The first inode we look at is the journal inode.  Don't try
4511          * root first: it may be modified in the journal!
4512          */
4513         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
4514                 err = ext4_load_journal(sb, es, journal_devnum);
4515                 if (err)
4516                         goto failed_mount3a;
4517         } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
4518                    ext4_has_feature_journal_needs_recovery(sb)) {
4519                 ext4_msg(sb, KERN_ERR, "required journal recovery "
4520                        "suppressed and not mounted read-only");
4521                 goto failed_mount_wq;
4522         } else {
4523                 /* Nojournal mode, all journal mount options are illegal */
4524                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
4525                         ext4_msg(sb, KERN_ERR, "can't mount with "
4526                                  "journal_checksum, fs mounted w/o journal");
4527                         goto failed_mount_wq;
4528                 }
4529                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4530                         ext4_msg(sb, KERN_ERR, "can't mount with "
4531                                  "journal_async_commit, fs mounted w/o journal");
4532                         goto failed_mount_wq;
4533                 }
4534                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
4535                         ext4_msg(sb, KERN_ERR, "can't mount with "
4536                                  "commit=%lu, fs mounted w/o journal",
4537                                  sbi->s_commit_interval / HZ);
4538                         goto failed_mount_wq;
4539                 }
4540                 if (EXT4_MOUNT_DATA_FLAGS &
4541                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
4542                         ext4_msg(sb, KERN_ERR, "can't mount with "
4543                                  "data=, fs mounted w/o journal");
4544                         goto failed_mount_wq;
4545                 }
4546                 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM;
4547                 clear_opt(sb, JOURNAL_CHECKSUM);
4548                 clear_opt(sb, DATA_FLAGS);
4549                 sbi->s_journal = NULL;
4550                 needs_recovery = 0;
4551                 goto no_journal;
4552         }
4553
4554         if (ext4_has_feature_64bit(sb) &&
4555             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4556                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4557                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4558                 goto failed_mount_wq;
4559         }
4560
4561         if (!set_journal_csum_feature_set(sb)) {
4562                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4563                          "feature set");
4564                 goto failed_mount_wq;
4565         }
4566
4567         /* We have now updated the journal if required, so we can
4568          * validate the data journaling mode. */
4569         switch (test_opt(sb, DATA_FLAGS)) {
4570         case 0:
4571                 /* No mode set, assume a default based on the journal
4572                  * capabilities: ORDERED_DATA if the journal can
4573                  * cope, else JOURNAL_DATA
4574                  */
4575                 if (jbd2_journal_check_available_features
4576                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4577                         set_opt(sb, ORDERED_DATA);
4578                         sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
4579                 } else {
4580                         set_opt(sb, JOURNAL_DATA);
4581                         sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
4582                 }
4583                 break;
4584
4585         case EXT4_MOUNT_ORDERED_DATA:
4586         case EXT4_MOUNT_WRITEBACK_DATA:
4587                 if (!jbd2_journal_check_available_features
4588                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4589                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4590                                "requested data journaling mode");
4591                         goto failed_mount_wq;
4592                 }
4593         default:
4594                 break;
4595         }
4596
4597         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4598             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4599                 ext4_msg(sb, KERN_ERR, "can't mount with "
4600                         "journal_async_commit in data=ordered mode");
4601                 goto failed_mount_wq;
4602         }
4603
4604         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4605
4606         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4607
4608 no_journal:
4609         if (!test_opt(sb, NO_MBCACHE)) {
4610                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
4611                 if (!sbi->s_ea_block_cache) {
4612                         ext4_msg(sb, KERN_ERR,
4613                                  "Failed to create ea_block_cache");
4614                         goto failed_mount_wq;
4615                 }
4616
4617                 if (ext4_has_feature_ea_inode(sb)) {
4618                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
4619                         if (!sbi->s_ea_inode_cache) {
4620                                 ext4_msg(sb, KERN_ERR,
4621                                          "Failed to create ea_inode_cache");
4622                                 goto failed_mount_wq;
4623                         }
4624                 }
4625         }
4626
4627         if (ext4_has_feature_verity(sb) && blocksize != PAGE_SIZE) {
4628                 ext4_msg(sb, KERN_ERR, "Unsupported blocksize for fs-verity");
4629                 goto failed_mount_wq;
4630         }
4631
4632         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) &&
4633             !ext4_has_feature_encrypt(sb)) {
4634                 ext4_set_feature_encrypt(sb);
4635                 ext4_commit_super(sb, 1);
4636         }
4637
4638         /*
4639          * Get the # of file system overhead blocks from the
4640          * superblock if present.
4641          */
4642         if (es->s_overhead_clusters)
4643                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4644         else {
4645                 err = ext4_calculate_overhead(sb);
4646                 if (err)
4647                         goto failed_mount_wq;
4648         }
4649
4650         /*
4651          * The maximum number of concurrent works can be high and
4652          * concurrency isn't really necessary.  Limit it to 1.
4653          */
4654         EXT4_SB(sb)->rsv_conversion_wq =
4655                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4656         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4657                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4658                 ret = -ENOMEM;
4659                 goto failed_mount4;
4660         }
4661
4662         /*
4663          * The jbd2_journal_load will have done any necessary log recovery,
4664          * so we can safely mount the rest of the filesystem now.
4665          */
4666
4667         root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL);
4668         if (IS_ERR(root)) {
4669                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4670                 ret = PTR_ERR(root);
4671                 root = NULL;
4672                 goto failed_mount4;
4673         }
4674         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4675                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4676                 iput(root);
4677                 goto failed_mount4;
4678         }
4679
4680 #ifdef CONFIG_UNICODE
4681         if (sbi->s_encoding)
4682                 sb->s_d_op = &ext4_dentry_ops;
4683 #endif
4684
4685         sb->s_root = d_make_root(root);
4686         if (!sb->s_root) {
4687                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4688                 ret = -ENOMEM;
4689                 goto failed_mount4;
4690         }
4691
4692         ret = ext4_setup_super(sb, es, sb_rdonly(sb));
4693         if (ret == -EROFS) {
4694                 sb->s_flags |= SB_RDONLY;
4695                 ret = 0;
4696         } else if (ret)
4697                 goto failed_mount4a;
4698
4699         ext4_set_resv_clusters(sb);
4700
4701         err = ext4_setup_system_zone(sb);
4702         if (err) {
4703                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4704                          "zone (%d)", err);
4705                 goto failed_mount4a;
4706         }
4707
4708         ext4_ext_init(sb);
4709         err = ext4_mb_init(sb);
4710         if (err) {
4711                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4712                          err);
4713                 goto failed_mount5;
4714         }
4715
4716         block = ext4_count_free_clusters(sb);
4717         ext4_free_blocks_count_set(sbi->s_es,
4718                                    EXT4_C2B(sbi, block));
4719         ext4_superblock_csum_set(sb);
4720         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4721                                   GFP_KERNEL);
4722         if (!err) {
4723                 unsigned long freei = ext4_count_free_inodes(sb);
4724                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4725                 ext4_superblock_csum_set(sb);
4726                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4727                                           GFP_KERNEL);
4728         }
4729         if (!err)
4730                 err = percpu_counter_init(&sbi->s_dirs_counter,
4731                                           ext4_count_dirs(sb), GFP_KERNEL);
4732         if (!err)
4733                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4734                                           GFP_KERNEL);
4735         if (!err)
4736                 err = percpu_init_rwsem(&sbi->s_writepages_rwsem);
4737
4738         if (err) {
4739                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4740                 goto failed_mount6;
4741         }
4742
4743         if (ext4_has_feature_flex_bg(sb))
4744                 if (!ext4_fill_flex_info(sb)) {
4745                         ext4_msg(sb, KERN_ERR,
4746                                "unable to initialize "
4747                                "flex_bg meta info!");
4748                         goto failed_mount6;
4749                 }
4750
4751         err = ext4_register_li_request(sb, first_not_zeroed);
4752         if (err)
4753                 goto failed_mount6;
4754
4755         err = ext4_register_sysfs(sb);
4756         if (err)
4757                 goto failed_mount7;
4758
4759 #ifdef CONFIG_QUOTA
4760         /* Enable quota usage during mount. */
4761         if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
4762                 err = ext4_enable_quotas(sb);
4763                 if (err)
4764                         goto failed_mount8;
4765         }
4766 #endif  /* CONFIG_QUOTA */
4767
4768         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4769         ext4_orphan_cleanup(sb, es);
4770         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4771         if (needs_recovery) {
4772                 ext4_msg(sb, KERN_INFO, "recovery complete");
4773                 ext4_mark_recovery_complete(sb, es);
4774         }
4775         if (EXT4_SB(sb)->s_journal) {
4776                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4777                         descr = " journalled data mode";
4778                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4779                         descr = " ordered data mode";
4780                 else
4781                         descr = " writeback data mode";
4782         } else
4783                 descr = "out journal";
4784
4785         if (test_opt(sb, DISCARD)) {
4786                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4787                 if (!blk_queue_discard(q))
4788                         ext4_msg(sb, KERN_WARNING,
4789                                  "mounting with \"discard\" option, but "
4790                                  "the device does not support discard");
4791         }
4792
4793         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4794                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4795                          "Opts: %.*s%s%s", descr,
4796                          (int) sizeof(sbi->s_es->s_mount_opts),
4797                          sbi->s_es->s_mount_opts,
4798                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4799
4800         if (es->s_error_count)
4801                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4802
4803         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4804         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4805         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4806         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4807
4808         kfree(orig_data);
4809         return 0;
4810
4811 cantfind_ext4:
4812         if (!silent)
4813                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4814         goto failed_mount;
4815
4816 #ifdef CONFIG_QUOTA
4817 failed_mount8:
4818         ext4_unregister_sysfs(sb);
4819 #endif
4820 failed_mount7:
4821         ext4_unregister_li_request(sb);
4822 failed_mount6:
4823         ext4_mb_release(sb);
4824         rcu_read_lock();
4825         flex_groups = rcu_dereference(sbi->s_flex_groups);
4826         if (flex_groups) {
4827                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
4828                         kvfree(flex_groups[i]);
4829                 kvfree(flex_groups);
4830         }
4831         rcu_read_unlock();
4832         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4833         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4834         percpu_counter_destroy(&sbi->s_dirs_counter);
4835         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4836         percpu_free_rwsem(&sbi->s_writepages_rwsem);
4837 failed_mount5:
4838         ext4_ext_release(sb);
4839         ext4_release_system_zone(sb);
4840 failed_mount4a:
4841         dput(sb->s_root);
4842         sb->s_root = NULL;
4843 failed_mount4:
4844         ext4_msg(sb, KERN_ERR, "mount failed");
4845         if (EXT4_SB(sb)->rsv_conversion_wq)
4846                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4847 failed_mount_wq:
4848         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
4849         sbi->s_ea_inode_cache = NULL;
4850
4851         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
4852         sbi->s_ea_block_cache = NULL;
4853
4854         if (sbi->s_journal) {
4855                 jbd2_journal_destroy(sbi->s_journal);
4856                 sbi->s_journal = NULL;
4857         }
4858 failed_mount3a:
4859         ext4_es_unregister_shrinker(sbi);
4860 failed_mount3:
4861         del_timer_sync(&sbi->s_err_report);
4862         if (sbi->s_mmp_tsk)
4863                 kthread_stop(sbi->s_mmp_tsk);
4864 failed_mount2:
4865         rcu_read_lock();
4866         group_desc = rcu_dereference(sbi->s_group_desc);
4867         for (i = 0; i < db_count; i++)
4868                 brelse(group_desc[i]);
4869         kvfree(group_desc);
4870         rcu_read_unlock();
4871 failed_mount:
4872         if (sbi->s_chksum_driver)
4873                 crypto_free_shash(sbi->s_chksum_driver);
4874
4875 #ifdef CONFIG_UNICODE
4876         utf8_unload(sbi->s_encoding);
4877 #endif
4878
4879 #ifdef CONFIG_QUOTA
4880         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4881                 kfree(get_qf_name(sb, sbi, i));
4882 #endif
4883         fscrypt_free_dummy_context(&sbi->s_dummy_enc_ctx);
4884         ext4_blkdev_remove(sbi);
4885         brelse(bh);
4886 out_fail:
4887         sb->s_fs_info = NULL;
4888         kfree(sbi->s_blockgroup_lock);
4889 out_free_base:
4890         kfree(sbi);
4891         kfree(orig_data);
4892         fs_put_dax(dax_dev);
4893         return err ? err : ret;
4894 }
4895
4896 /*
4897  * Setup any per-fs journal parameters now.  We'll do this both on
4898  * initial mount, once the journal has been initialised but before we've
4899  * done any recovery; and again on any subsequent remount.
4900  */
4901 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4902 {
4903         struct ext4_sb_info *sbi = EXT4_SB(sb);
4904
4905         journal->j_commit_interval = sbi->s_commit_interval;
4906         journal->j_min_batch_time = sbi->s_min_batch_time;
4907         journal->j_max_batch_time = sbi->s_max_batch_time;
4908
4909         write_lock(&journal->j_state_lock);
4910         if (test_opt(sb, BARRIER))
4911                 journal->j_flags |= JBD2_BARRIER;
4912         else
4913                 journal->j_flags &= ~JBD2_BARRIER;
4914         if (test_opt(sb, DATA_ERR_ABORT))
4915                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4916         else
4917                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4918         write_unlock(&journal->j_state_lock);
4919 }
4920
4921 static struct inode *ext4_get_journal_inode(struct super_block *sb,
4922                                              unsigned int journal_inum)
4923 {
4924         struct inode *journal_inode;
4925
4926         /*
4927          * Test for the existence of a valid inode on disk.  Bad things
4928          * happen if we iget() an unused inode, as the subsequent iput()
4929          * will try to delete it.
4930          */
4931         journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL);
4932         if (IS_ERR(journal_inode)) {
4933                 ext4_msg(sb, KERN_ERR, "no journal found");
4934                 return NULL;
4935         }
4936         if (!journal_inode->i_nlink) {
4937                 make_bad_inode(journal_inode);
4938                 iput(journal_inode);
4939                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4940                 return NULL;
4941         }
4942
4943         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4944                   journal_inode, journal_inode->i_size);
4945         if (!S_ISREG(journal_inode->i_mode)) {
4946                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4947                 iput(journal_inode);
4948                 return NULL;
4949         }
4950         return journal_inode;
4951 }
4952
4953 static journal_t *ext4_get_journal(struct super_block *sb,
4954                                    unsigned int journal_inum)
4955 {
4956         struct inode *journal_inode;
4957         journal_t *journal;
4958
4959         BUG_ON(!ext4_has_feature_journal(sb));
4960
4961         journal_inode = ext4_get_journal_inode(sb, journal_inum);
4962         if (!journal_inode)
4963                 return NULL;
4964
4965         journal = jbd2_journal_init_inode(journal_inode);
4966         if (!journal) {
4967                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4968                 iput(journal_inode);
4969                 return NULL;
4970         }
4971         journal->j_private = sb;
4972         ext4_init_journal_params(sb, journal);
4973         return journal;
4974 }
4975
4976 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4977                                        dev_t j_dev)
4978 {
4979         struct buffer_head *bh;
4980         journal_t *journal;
4981         ext4_fsblk_t start;
4982         ext4_fsblk_t len;
4983         int hblock, blocksize;
4984         ext4_fsblk_t sb_block;
4985         unsigned long offset;
4986         struct ext4_super_block *es;
4987         struct block_device *bdev;
4988
4989         BUG_ON(!ext4_has_feature_journal(sb));
4990
4991         bdev = ext4_blkdev_get(j_dev, sb);
4992         if (bdev == NULL)
4993                 return NULL;
4994
4995         blocksize = sb->s_blocksize;
4996         hblock = bdev_logical_block_size(bdev);
4997         if (blocksize < hblock) {
4998                 ext4_msg(sb, KERN_ERR,
4999                         "blocksize too small for journal device");
5000                 goto out_bdev;
5001         }
5002
5003         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
5004         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
5005         set_blocksize(bdev, blocksize);
5006         if (!(bh = __bread(bdev, sb_block, blocksize))) {
5007                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
5008                        "external journal");
5009                 goto out_bdev;
5010         }
5011
5012         es = (struct ext4_super_block *) (bh->b_data + offset);
5013         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
5014             !(le32_to_cpu(es->s_feature_incompat) &
5015               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
5016                 ext4_msg(sb, KERN_ERR, "external journal has "
5017                                         "bad superblock");
5018                 brelse(bh);
5019                 goto out_bdev;
5020         }
5021
5022         if ((le32_to_cpu(es->s_feature_ro_compat) &
5023              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
5024             es->s_checksum != ext4_superblock_csum(sb, es)) {
5025                 ext4_msg(sb, KERN_ERR, "external journal has "
5026                                        "corrupt superblock");
5027                 brelse(bh);
5028                 goto out_bdev;
5029         }
5030
5031         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
5032                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
5033                 brelse(bh);
5034                 goto out_bdev;
5035         }
5036
5037         len = ext4_blocks_count(es);
5038         start = sb_block + 1;
5039         brelse(bh);     /* we're done with the superblock */
5040
5041         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
5042                                         start, len, blocksize);
5043         if (!journal) {
5044                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
5045                 goto out_bdev;
5046         }
5047         journal->j_private = sb;
5048         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
5049         wait_on_buffer(journal->j_sb_buffer);
5050         if (!buffer_uptodate(journal->j_sb_buffer)) {
5051                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
5052                 goto out_journal;
5053         }
5054         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
5055                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
5056                                         "user (unsupported) - %d",
5057                         be32_to_cpu(journal->j_superblock->s_nr_users));
5058                 goto out_journal;
5059         }
5060         EXT4_SB(sb)->journal_bdev = bdev;
5061         ext4_init_journal_params(sb, journal);
5062         return journal;
5063
5064 out_journal:
5065         jbd2_journal_destroy(journal);
5066 out_bdev:
5067         ext4_blkdev_put(bdev);
5068         return NULL;
5069 }
5070
5071 static int ext4_load_journal(struct super_block *sb,
5072                              struct ext4_super_block *es,
5073                              unsigned long journal_devnum)
5074 {
5075         journal_t *journal;
5076         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
5077         dev_t journal_dev;
5078         int err = 0;
5079         int really_read_only;
5080
5081         BUG_ON(!ext4_has_feature_journal(sb));
5082
5083         if (journal_devnum &&
5084             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5085                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
5086                         "numbers have changed");
5087                 journal_dev = new_decode_dev(journal_devnum);
5088         } else
5089                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
5090
5091         really_read_only = bdev_read_only(sb->s_bdev);
5092
5093         /*
5094          * Are we loading a blank journal or performing recovery after a
5095          * crash?  For recovery, we need to check in advance whether we
5096          * can get read-write access to the device.
5097          */
5098         if (ext4_has_feature_journal_needs_recovery(sb)) {
5099                 if (sb_rdonly(sb)) {
5100                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
5101                                         "required on readonly filesystem");
5102                         if (really_read_only) {
5103                                 ext4_msg(sb, KERN_ERR, "write access "
5104                                         "unavailable, cannot proceed "
5105                                         "(try mounting with noload)");
5106                                 return -EROFS;
5107                         }
5108                         ext4_msg(sb, KERN_INFO, "write access will "
5109                                "be enabled during recovery");
5110                 }
5111         }
5112
5113         if (journal_inum && journal_dev) {
5114                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
5115                        "and inode journals!");
5116                 return -EINVAL;
5117         }
5118
5119         if (journal_inum) {
5120                 if (!(journal = ext4_get_journal(sb, journal_inum)))
5121                         return -EINVAL;
5122         } else {
5123                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
5124                         return -EINVAL;
5125         }
5126
5127         if (!(journal->j_flags & JBD2_BARRIER))
5128                 ext4_msg(sb, KERN_INFO, "barriers disabled");
5129
5130         if (!ext4_has_feature_journal_needs_recovery(sb))
5131                 err = jbd2_journal_wipe(journal, !really_read_only);
5132         if (!err) {
5133                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
5134                 if (save)
5135                         memcpy(save, ((char *) es) +
5136                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
5137                 err = jbd2_journal_load(journal);
5138                 if (save)
5139                         memcpy(((char *) es) + EXT4_S_ERR_START,
5140                                save, EXT4_S_ERR_LEN);
5141                 kfree(save);
5142         }
5143
5144         if (err) {
5145                 ext4_msg(sb, KERN_ERR, "error loading journal");
5146                 jbd2_journal_destroy(journal);
5147                 return err;
5148         }
5149
5150         EXT4_SB(sb)->s_journal = journal;
5151         ext4_clear_journal_err(sb, es);
5152
5153         if (!really_read_only && journal_devnum &&
5154             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5155                 es->s_journal_dev = cpu_to_le32(journal_devnum);
5156
5157                 /* Make sure we flush the recovery flag to disk. */
5158                 ext4_commit_super(sb, 1);
5159         }
5160
5161         return 0;
5162 }
5163
5164 static int ext4_commit_super(struct super_block *sb, int sync)
5165 {
5166         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
5167         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
5168         int error = 0;
5169
5170         if (!sbh || block_device_ejected(sb))
5171                 return error;
5172
5173         /*
5174          * The superblock bh should be mapped, but it might not be if the
5175          * device was hot-removed. Not much we can do but fail the I/O.
5176          */
5177         if (!buffer_mapped(sbh))
5178                 return error;
5179
5180         /*
5181          * If the file system is mounted read-only, don't update the
5182          * superblock write time.  This avoids updating the superblock
5183          * write time when we are mounting the root file system
5184          * read/only but we need to replay the journal; at that point,
5185          * for people who are east of GMT and who make their clock
5186          * tick in localtime for Windows bug-for-bug compatibility,
5187          * the clock is set in the future, and this will cause e2fsck
5188          * to complain and force a full file system check.
5189          */
5190         if (!(sb->s_flags & SB_RDONLY))
5191                 ext4_update_tstamp(es, s_wtime);
5192         if (sb->s_bdev->bd_part)
5193                 es->s_kbytes_written =
5194                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
5195                             ((part_stat_read(sb->s_bdev->bd_part,
5196                                              sectors[STAT_WRITE]) -
5197                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
5198         else
5199                 es->s_kbytes_written =
5200                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
5201         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
5202                 ext4_free_blocks_count_set(es,
5203                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
5204                                 &EXT4_SB(sb)->s_freeclusters_counter)));
5205         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
5206                 es->s_free_inodes_count =
5207                         cpu_to_le32(percpu_counter_sum_positive(
5208                                 &EXT4_SB(sb)->s_freeinodes_counter));
5209         BUFFER_TRACE(sbh, "marking dirty");
5210         ext4_superblock_csum_set(sb);
5211         if (sync)
5212                 lock_buffer(sbh);
5213         if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
5214                 /*
5215                  * Oh, dear.  A previous attempt to write the
5216                  * superblock failed.  This could happen because the
5217                  * USB device was yanked out.  Or it could happen to
5218                  * be a transient write error and maybe the block will
5219                  * be remapped.  Nothing we can do but to retry the
5220                  * write and hope for the best.
5221                  */
5222                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
5223                        "superblock detected");
5224                 clear_buffer_write_io_error(sbh);
5225                 set_buffer_uptodate(sbh);
5226         }
5227         mark_buffer_dirty(sbh);
5228         if (sync) {
5229                 unlock_buffer(sbh);
5230                 error = __sync_dirty_buffer(sbh,
5231                         REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
5232                 if (buffer_write_io_error(sbh)) {
5233                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
5234                                "superblock");
5235                         clear_buffer_write_io_error(sbh);
5236                         set_buffer_uptodate(sbh);
5237                 }
5238         }
5239         return error;
5240 }
5241
5242 /*
5243  * Have we just finished recovery?  If so, and if we are mounting (or
5244  * remounting) the filesystem readonly, then we will end up with a
5245  * consistent fs on disk.  Record that fact.
5246  */
5247 static void ext4_mark_recovery_complete(struct super_block *sb,
5248                                         struct ext4_super_block *es)
5249 {
5250         journal_t *journal = EXT4_SB(sb)->s_journal;
5251
5252         if (!ext4_has_feature_journal(sb)) {
5253                 BUG_ON(journal != NULL);
5254                 return;
5255         }
5256         jbd2_journal_lock_updates(journal);
5257         if (jbd2_journal_flush(journal) < 0)
5258                 goto out;
5259
5260         if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) {
5261                 ext4_clear_feature_journal_needs_recovery(sb);
5262                 ext4_commit_super(sb, 1);
5263         }
5264
5265 out:
5266         jbd2_journal_unlock_updates(journal);
5267 }
5268
5269 /*
5270  * If we are mounting (or read-write remounting) a filesystem whose journal
5271  * has recorded an error from a previous lifetime, move that error to the
5272  * main filesystem now.
5273  */
5274 static void ext4_clear_journal_err(struct super_block *sb,
5275                                    struct ext4_super_block *es)
5276 {
5277         journal_t *journal;
5278         int j_errno;
5279         const char *errstr;
5280
5281         BUG_ON(!ext4_has_feature_journal(sb));
5282
5283         journal = EXT4_SB(sb)->s_journal;
5284
5285         /*
5286          * Now check for any error status which may have been recorded in the
5287          * journal by a prior ext4_error() or ext4_abort()
5288          */
5289
5290         j_errno = jbd2_journal_errno(journal);
5291         if (j_errno) {
5292                 char nbuf[16];
5293
5294                 errstr = ext4_decode_error(sb, j_errno, nbuf);
5295                 ext4_warning(sb, "Filesystem error recorded "
5296                              "from previous mount: %s", errstr);
5297                 ext4_warning(sb, "Marking fs in need of filesystem check.");
5298
5299                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
5300                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
5301                 ext4_commit_super(sb, 1);
5302
5303                 jbd2_journal_clear_err(journal);
5304                 jbd2_journal_update_sb_errno(journal);
5305         }
5306 }
5307
5308 /*
5309  * Force the running and committing transactions to commit,
5310  * and wait on the commit.
5311  */
5312 int ext4_force_commit(struct super_block *sb)
5313 {
5314         journal_t *journal;
5315
5316         if (sb_rdonly(sb))
5317                 return 0;
5318
5319         journal = EXT4_SB(sb)->s_journal;
5320         return ext4_journal_force_commit(journal);
5321 }
5322
5323 static int ext4_sync_fs(struct super_block *sb, int wait)
5324 {
5325         int ret = 0;
5326         tid_t target;
5327         bool needs_barrier = false;
5328         struct ext4_sb_info *sbi = EXT4_SB(sb);
5329
5330         if (unlikely(ext4_forced_shutdown(sbi)))
5331                 return 0;
5332
5333         trace_ext4_sync_fs(sb, wait);
5334         flush_workqueue(sbi->rsv_conversion_wq);
5335         /*
5336          * Writeback quota in non-journalled quota case - journalled quota has
5337          * no dirty dquots
5338          */
5339         dquot_writeback_dquots(sb, -1);
5340         /*
5341          * Data writeback is possible w/o journal transaction, so barrier must
5342          * being sent at the end of the function. But we can skip it if
5343          * transaction_commit will do it for us.
5344          */
5345         if (sbi->s_journal) {
5346                 target = jbd2_get_latest_transaction(sbi->s_journal);
5347                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
5348                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
5349                         needs_barrier = true;
5350
5351                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
5352                         if (wait)
5353                                 ret = jbd2_log_wait_commit(sbi->s_journal,
5354                                                            target);
5355                 }
5356         } else if (wait && test_opt(sb, BARRIER))
5357                 needs_barrier = true;
5358         if (needs_barrier) {
5359                 int err;
5360                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL);
5361                 if (!ret)
5362                         ret = err;
5363         }
5364
5365         return ret;
5366 }
5367
5368 /*
5369  * LVM calls this function before a (read-only) snapshot is created.  This
5370  * gives us a chance to flush the journal completely and mark the fs clean.
5371  *
5372  * Note that only this function cannot bring a filesystem to be in a clean
5373  * state independently. It relies on upper layer to stop all data & metadata
5374  * modifications.
5375  */
5376 static int ext4_freeze(struct super_block *sb)
5377 {
5378         int error = 0;
5379         journal_t *journal;
5380
5381         if (sb_rdonly(sb))
5382                 return 0;
5383
5384         journal = EXT4_SB(sb)->s_journal;
5385
5386         if (journal) {
5387                 /* Now we set up the journal barrier. */
5388                 jbd2_journal_lock_updates(journal);
5389
5390                 /*
5391                  * Don't clear the needs_recovery flag if we failed to
5392                  * flush the journal.
5393                  */
5394                 error = jbd2_journal_flush(journal);
5395                 if (error < 0)
5396                         goto out;
5397
5398                 /* Journal blocked and flushed, clear needs_recovery flag. */
5399                 ext4_clear_feature_journal_needs_recovery(sb);
5400         }
5401
5402         error = ext4_commit_super(sb, 1);
5403 out:
5404         if (journal)
5405                 /* we rely on upper layer to stop further updates */
5406                 jbd2_journal_unlock_updates(journal);
5407         return error;
5408 }
5409
5410 /*
5411  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
5412  * flag here, even though the filesystem is not technically dirty yet.
5413  */
5414 static int ext4_unfreeze(struct super_block *sb)
5415 {
5416         if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
5417                 return 0;
5418
5419         if (EXT4_SB(sb)->s_journal) {
5420                 /* Reset the needs_recovery flag before the fs is unlocked. */
5421                 ext4_set_feature_journal_needs_recovery(sb);
5422         }
5423
5424         ext4_commit_super(sb, 1);
5425         return 0;
5426 }
5427
5428 /*
5429  * Structure to save mount options for ext4_remount's benefit
5430  */
5431 struct ext4_mount_options {
5432         unsigned long s_mount_opt;
5433         unsigned long s_mount_opt2;
5434         kuid_t s_resuid;
5435         kgid_t s_resgid;
5436         unsigned long s_commit_interval;
5437         u32 s_min_batch_time, s_max_batch_time;
5438 #ifdef CONFIG_QUOTA
5439         int s_jquota_fmt;
5440         char *s_qf_names[EXT4_MAXQUOTAS];
5441 #endif
5442 };
5443
5444 static int ext4_remount(struct super_block *sb, int *flags, char *data)
5445 {
5446         struct ext4_super_block *es;
5447         struct ext4_sb_info *sbi = EXT4_SB(sb);
5448         unsigned long old_sb_flags;
5449         struct ext4_mount_options old_opts;
5450         int enable_quota = 0;
5451         ext4_group_t g;
5452         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
5453         int err = 0;
5454 #ifdef CONFIG_QUOTA
5455         int i, j;
5456         char *to_free[EXT4_MAXQUOTAS];
5457 #endif
5458         char *orig_data = kstrdup(data, GFP_KERNEL);
5459
5460         if (data && !orig_data)
5461                 return -ENOMEM;
5462
5463         /* Store the original options */
5464         old_sb_flags = sb->s_flags;
5465         old_opts.s_mount_opt = sbi->s_mount_opt;
5466         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
5467         old_opts.s_resuid = sbi->s_resuid;
5468         old_opts.s_resgid = sbi->s_resgid;
5469         old_opts.s_commit_interval = sbi->s_commit_interval;
5470         old_opts.s_min_batch_time = sbi->s_min_batch_time;
5471         old_opts.s_max_batch_time = sbi->s_max_batch_time;
5472 #ifdef CONFIG_QUOTA
5473         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
5474         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5475                 if (sbi->s_qf_names[i]) {
5476                         char *qf_name = get_qf_name(sb, sbi, i);
5477
5478                         old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL);
5479                         if (!old_opts.s_qf_names[i]) {
5480                                 for (j = 0; j < i; j++)
5481                                         kfree(old_opts.s_qf_names[j]);
5482                                 kfree(orig_data);
5483                                 return -ENOMEM;
5484                         }
5485                 } else
5486                         old_opts.s_qf_names[i] = NULL;
5487 #endif
5488         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
5489                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
5490
5491         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
5492                 err = -EINVAL;
5493                 goto restore_opts;
5494         }
5495
5496         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
5497             test_opt(sb, JOURNAL_CHECKSUM)) {
5498                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
5499                          "during remount not supported; ignoring");
5500                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
5501         }
5502
5503         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
5504                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
5505                         ext4_msg(sb, KERN_ERR, "can't mount with "
5506                                  "both data=journal and delalloc");
5507                         err = -EINVAL;
5508                         goto restore_opts;
5509                 }
5510                 if (test_opt(sb, DIOREAD_NOLOCK)) {
5511                         ext4_msg(sb, KERN_ERR, "can't mount with "
5512                                  "both data=journal and dioread_nolock");
5513                         err = -EINVAL;
5514                         goto restore_opts;
5515                 }
5516         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
5517                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5518                         ext4_msg(sb, KERN_ERR, "can't mount with "
5519                                 "journal_async_commit in data=ordered mode");
5520                         err = -EINVAL;
5521                         goto restore_opts;
5522                 }
5523         }
5524
5525         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
5526                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
5527                 err = -EINVAL;
5528                 goto restore_opts;
5529         }
5530
5531         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
5532                 ext4_abort(sb, EXT4_ERR_ESHUTDOWN, "Abort forced by user");
5533
5534         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
5535                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
5536
5537         es = sbi->s_es;
5538
5539         if (sbi->s_journal) {
5540                 ext4_init_journal_params(sb, sbi->s_journal);
5541                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
5542         }
5543
5544         if (*flags & SB_LAZYTIME)
5545                 sb->s_flags |= SB_LAZYTIME;
5546
5547         if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) {
5548                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
5549                         err = -EROFS;
5550                         goto restore_opts;
5551                 }
5552
5553                 if (*flags & SB_RDONLY) {
5554                         err = sync_filesystem(sb);
5555                         if (err < 0)
5556                                 goto restore_opts;
5557                         err = dquot_suspend(sb, -1);
5558                         if (err < 0)
5559                                 goto restore_opts;
5560
5561                         /*
5562                          * First of all, the unconditional stuff we have to do
5563                          * to disable replay of the journal when we next remount
5564                          */
5565                         sb->s_flags |= SB_RDONLY;
5566
5567                         /*
5568                          * OK, test if we are remounting a valid rw partition
5569                          * readonly, and if so set the rdonly flag and then
5570                          * mark the partition as valid again.
5571                          */
5572                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5573                             (sbi->s_mount_state & EXT4_VALID_FS))
5574                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5575
5576                         if (sbi->s_journal)
5577                                 ext4_mark_recovery_complete(sb, es);
5578                         if (sbi->s_mmp_tsk)
5579                                 kthread_stop(sbi->s_mmp_tsk);
5580                 } else {
5581                         /* Make sure we can mount this feature set readwrite */
5582                         if (ext4_has_feature_readonly(sb) ||
5583                             !ext4_feature_set_ok(sb, 0)) {
5584                                 err = -EROFS;
5585                                 goto restore_opts;
5586                         }
5587                         /*
5588                          * Make sure the group descriptor checksums
5589                          * are sane.  If they aren't, refuse to remount r/w.
5590                          */
5591                         for (g = 0; g < sbi->s_groups_count; g++) {
5592                                 struct ext4_group_desc *gdp =
5593                                         ext4_get_group_desc(sb, g, NULL);
5594
5595                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5596                                         ext4_msg(sb, KERN_ERR,
5597                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5598                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
5599                                                le16_to_cpu(gdp->bg_checksum));
5600                                         err = -EFSBADCRC;
5601                                         goto restore_opts;
5602                                 }
5603                         }
5604
5605                         /*
5606                          * If we have an unprocessed orphan list hanging
5607                          * around from a previously readonly bdev mount,
5608                          * require a full umount/remount for now.
5609                          */
5610                         if (es->s_last_orphan) {
5611                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5612                                        "remount RDWR because of unprocessed "
5613                                        "orphan inode list.  Please "
5614                                        "umount/remount instead");
5615                                 err = -EINVAL;
5616                                 goto restore_opts;
5617                         }
5618
5619                         /*
5620                          * Mounting a RDONLY partition read-write, so reread
5621                          * and store the current valid flag.  (It may have
5622                          * been changed by e2fsck since we originally mounted
5623                          * the partition.)
5624                          */
5625                         if (sbi->s_journal)
5626                                 ext4_clear_journal_err(sb, es);
5627                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5628
5629                         err = ext4_setup_super(sb, es, 0);
5630                         if (err)
5631                                 goto restore_opts;
5632
5633                         sb->s_flags &= ~SB_RDONLY;
5634                         if (ext4_has_feature_mmp(sb))
5635                                 if (ext4_multi_mount_protect(sb,
5636                                                 le64_to_cpu(es->s_mmp_block))) {
5637                                         err = -EROFS;
5638                                         goto restore_opts;
5639                                 }
5640                         enable_quota = 1;
5641                 }
5642         }
5643
5644         /*
5645          * Reinitialize lazy itable initialization thread based on
5646          * current settings
5647          */
5648         if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
5649                 ext4_unregister_li_request(sb);
5650         else {
5651                 ext4_group_t first_not_zeroed;
5652                 first_not_zeroed = ext4_has_uninit_itable(sb);
5653                 ext4_register_li_request(sb, first_not_zeroed);
5654         }
5655
5656         ext4_setup_system_zone(sb);
5657         if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) {
5658                 err = ext4_commit_super(sb, 1);
5659                 if (err)
5660                         goto restore_opts;
5661         }
5662
5663 #ifdef CONFIG_QUOTA
5664         /* Release old quota file names */
5665         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5666                 kfree(old_opts.s_qf_names[i]);
5667         if (enable_quota) {
5668                 if (sb_any_quota_suspended(sb))
5669                         dquot_resume(sb, -1);
5670                 else if (ext4_has_feature_quota(sb)) {
5671                         err = ext4_enable_quotas(sb);
5672                         if (err)
5673                                 goto restore_opts;
5674                 }
5675         }
5676 #endif
5677
5678         *flags = (*flags & ~SB_LAZYTIME) | (sb->s_flags & SB_LAZYTIME);
5679         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5680         kfree(orig_data);
5681         return 0;
5682
5683 restore_opts:
5684         sb->s_flags = old_sb_flags;
5685         sbi->s_mount_opt = old_opts.s_mount_opt;
5686         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5687         sbi->s_resuid = old_opts.s_resuid;
5688         sbi->s_resgid = old_opts.s_resgid;
5689         sbi->s_commit_interval = old_opts.s_commit_interval;
5690         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5691         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5692 #ifdef CONFIG_QUOTA
5693         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5694         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5695                 to_free[i] = get_qf_name(sb, sbi, i);
5696                 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]);
5697         }
5698         synchronize_rcu();
5699         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5700                 kfree(to_free[i]);
5701 #endif
5702         kfree(orig_data);
5703         return err;
5704 }
5705
5706 #ifdef CONFIG_QUOTA
5707 static int ext4_statfs_project(struct super_block *sb,
5708                                kprojid_t projid, struct kstatfs *buf)
5709 {
5710         struct kqid qid;
5711         struct dquot *dquot;
5712         u64 limit;
5713         u64 curblock;
5714
5715         qid = make_kqid_projid(projid);
5716         dquot = dqget(sb, qid);
5717         if (IS_ERR(dquot))
5718                 return PTR_ERR(dquot);
5719         spin_lock(&dquot->dq_dqb_lock);
5720
5721         limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit,
5722                              dquot->dq_dqb.dqb_bhardlimit);
5723         limit >>= sb->s_blocksize_bits;
5724
5725         if (limit && buf->f_blocks > limit) {
5726                 curblock = (dquot->dq_dqb.dqb_curspace +
5727                             dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits;
5728                 buf->f_blocks = limit;
5729                 buf->f_bfree = buf->f_bavail =
5730                         (buf->f_blocks > curblock) ?
5731                          (buf->f_blocks - curblock) : 0;
5732         }
5733
5734         limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit,
5735                              dquot->dq_dqb.dqb_ihardlimit);
5736         if (limit && buf->f_files > limit) {
5737                 buf->f_files = limit;
5738                 buf->f_ffree =
5739                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
5740                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
5741         }
5742
5743         spin_unlock(&dquot->dq_dqb_lock);
5744         dqput(dquot);
5745         return 0;
5746 }
5747 #endif
5748
5749 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5750 {
5751         struct super_block *sb = dentry->d_sb;
5752         struct ext4_sb_info *sbi = EXT4_SB(sb);
5753         struct ext4_super_block *es = sbi->s_es;
5754         ext4_fsblk_t overhead = 0, resv_blocks;
5755         u64 fsid;
5756         s64 bfree;
5757         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5758
5759         if (!test_opt(sb, MINIX_DF))
5760                 overhead = sbi->s_overhead;
5761
5762         buf->f_type = EXT4_SUPER_MAGIC;
5763         buf->f_bsize = sb->s_blocksize;
5764         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5765         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5766                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5767         /* prevent underflow in case that few free space is available */
5768         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5769         buf->f_bavail = buf->f_bfree -
5770                         (ext4_r_blocks_count(es) + resv_blocks);
5771         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5772                 buf->f_bavail = 0;
5773         buf->f_files = le32_to_cpu(es->s_inodes_count);
5774         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5775         buf->f_namelen = EXT4_NAME_LEN;
5776         fsid = le64_to_cpup((void *)es->s_uuid) ^
5777                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5778         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5779         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5780
5781 #ifdef CONFIG_QUOTA
5782         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
5783             sb_has_quota_limits_enabled(sb, PRJQUOTA))
5784                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
5785 #endif
5786         return 0;
5787 }
5788
5789
5790 #ifdef CONFIG_QUOTA
5791
5792 /*
5793  * Helper functions so that transaction is started before we acquire dqio_sem
5794  * to keep correct lock ordering of transaction > dqio_sem
5795  */
5796 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5797 {
5798         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5799 }
5800
5801 static int ext4_write_dquot(struct dquot *dquot)
5802 {
5803         int ret, err;
5804         handle_t *handle;
5805         struct inode *inode;
5806
5807         inode = dquot_to_inode(dquot);
5808         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5809                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5810         if (IS_ERR(handle))
5811                 return PTR_ERR(handle);
5812         ret = dquot_commit(dquot);
5813         err = ext4_journal_stop(handle);
5814         if (!ret)
5815                 ret = err;
5816         return ret;
5817 }
5818
5819 static int ext4_acquire_dquot(struct dquot *dquot)
5820 {
5821         int ret, err;
5822         handle_t *handle;
5823
5824         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5825                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5826         if (IS_ERR(handle))
5827                 return PTR_ERR(handle);
5828         ret = dquot_acquire(dquot);
5829         err = ext4_journal_stop(handle);
5830         if (!ret)
5831                 ret = err;
5832         return ret;
5833 }
5834
5835 static int ext4_release_dquot(struct dquot *dquot)
5836 {
5837         int ret, err;
5838         handle_t *handle;
5839
5840         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5841                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5842         if (IS_ERR(handle)) {
5843                 /* Release dquot anyway to avoid endless cycle in dqput() */
5844                 dquot_release(dquot);
5845                 return PTR_ERR(handle);
5846         }
5847         ret = dquot_release(dquot);
5848         err = ext4_journal_stop(handle);
5849         if (!ret)
5850                 ret = err;
5851         return ret;
5852 }
5853
5854 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5855 {
5856         struct super_block *sb = dquot->dq_sb;
5857         struct ext4_sb_info *sbi = EXT4_SB(sb);
5858
5859         /* Are we journaling quotas? */
5860         if (ext4_has_feature_quota(sb) ||
5861             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5862                 dquot_mark_dquot_dirty(dquot);
5863                 return ext4_write_dquot(dquot);
5864         } else {
5865                 return dquot_mark_dquot_dirty(dquot);
5866         }
5867 }
5868
5869 static int ext4_write_info(struct super_block *sb, int type)
5870 {
5871         int ret, err;
5872         handle_t *handle;
5873
5874         /* Data block + inode block */
5875         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5876         if (IS_ERR(handle))
5877                 return PTR_ERR(handle);
5878         ret = dquot_commit_info(sb, type);
5879         err = ext4_journal_stop(handle);
5880         if (!ret)
5881                 ret = err;
5882         return ret;
5883 }
5884
5885 /*
5886  * Turn on quotas during mount time - we need to find
5887  * the quota file and such...
5888  */
5889 static int ext4_quota_on_mount(struct super_block *sb, int type)
5890 {
5891         return dquot_quota_on_mount(sb, get_qf_name(sb, EXT4_SB(sb), type),
5892                                         EXT4_SB(sb)->s_jquota_fmt, type);
5893 }
5894
5895 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5896 {
5897         struct ext4_inode_info *ei = EXT4_I(inode);
5898
5899         /* The first argument of lockdep_set_subclass has to be
5900          * *exactly* the same as the argument to init_rwsem() --- in
5901          * this case, in init_once() --- or lockdep gets unhappy
5902          * because the name of the lock is set using the
5903          * stringification of the argument to init_rwsem().
5904          */
5905         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5906         lockdep_set_subclass(&ei->i_data_sem, subclass);
5907 }
5908
5909 /*
5910  * Standard function to be called on quota_on
5911  */
5912 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5913                          const struct path *path)
5914 {
5915         int err;
5916
5917         if (!test_opt(sb, QUOTA))
5918                 return -EINVAL;
5919
5920         /* Quotafile not on the same filesystem? */
5921         if (path->dentry->d_sb != sb)
5922                 return -EXDEV;
5923         /* Journaling quota? */
5924         if (EXT4_SB(sb)->s_qf_names[type]) {
5925                 /* Quotafile not in fs root? */
5926                 if (path->dentry->d_parent != sb->s_root)
5927                         ext4_msg(sb, KERN_WARNING,
5928                                 "Quota file not on filesystem root. "
5929                                 "Journaled quota will not work");
5930                 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
5931         } else {
5932                 /*
5933                  * Clear the flag just in case mount options changed since
5934                  * last time.
5935                  */
5936                 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
5937         }
5938
5939         /*
5940          * When we journal data on quota file, we have to flush journal to see
5941          * all updates to the file when we bypass pagecache...
5942          */
5943         if (EXT4_SB(sb)->s_journal &&
5944             ext4_should_journal_data(d_inode(path->dentry))) {
5945                 /*
5946                  * We don't need to lock updates but journal_flush() could
5947                  * otherwise be livelocked...
5948                  */
5949                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5950                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5951                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5952                 if (err)
5953                         return err;
5954         }
5955
5956         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5957         err = dquot_quota_on(sb, type, format_id, path);
5958         if (err) {
5959                 lockdep_set_quota_inode(path->dentry->d_inode,
5960                                              I_DATA_SEM_NORMAL);
5961         } else {
5962                 struct inode *inode = d_inode(path->dentry);
5963                 handle_t *handle;
5964
5965                 /*
5966                  * Set inode flags to prevent userspace from messing with quota
5967                  * files. If this fails, we return success anyway since quotas
5968                  * are already enabled and this is not a hard failure.
5969                  */
5970                 inode_lock(inode);
5971                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5972                 if (IS_ERR(handle))
5973                         goto unlock_inode;
5974                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
5975                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
5976                                 S_NOATIME | S_IMMUTABLE);
5977                 err = ext4_mark_inode_dirty(handle, inode);
5978                 ext4_journal_stop(handle);
5979         unlock_inode:
5980                 inode_unlock(inode);
5981         }
5982         return err;
5983 }
5984
5985 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5986                              unsigned int flags)
5987 {
5988         int err;
5989         struct inode *qf_inode;
5990         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5991                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5992                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5993                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5994         };
5995
5996         BUG_ON(!ext4_has_feature_quota(sb));
5997
5998         if (!qf_inums[type])
5999                 return -EPERM;
6000
6001         qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL);
6002         if (IS_ERR(qf_inode)) {
6003                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
6004                 return PTR_ERR(qf_inode);
6005         }
6006
6007         /* Don't account quota for quota files to avoid recursion */
6008         qf_inode->i_flags |= S_NOQUOTA;
6009         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
6010         err = dquot_load_quota_inode(qf_inode, type, format_id, flags);
6011         if (err)
6012                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
6013         iput(qf_inode);
6014
6015         return err;
6016 }
6017
6018 /* Enable usage tracking for all quota types. */
6019 static int ext4_enable_quotas(struct super_block *sb)
6020 {
6021         int type, err = 0;
6022         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
6023                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
6024                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
6025                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
6026         };
6027         bool quota_mopt[EXT4_MAXQUOTAS] = {
6028                 test_opt(sb, USRQUOTA),
6029                 test_opt(sb, GRPQUOTA),
6030                 test_opt(sb, PRJQUOTA),
6031         };
6032
6033         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
6034         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
6035                 if (qf_inums[type]) {
6036                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
6037                                 DQUOT_USAGE_ENABLED |
6038                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
6039                         if (err) {
6040                                 ext4_warning(sb,
6041                                         "Failed to enable quota tracking "
6042                                         "(type=%d, err=%d). Please run "
6043                                         "e2fsck to fix.", type, err);
6044                                 for (type--; type >= 0; type--)
6045                                         dquot_quota_off(sb, type);
6046
6047                                 return err;
6048                         }
6049                 }
6050         }
6051         return 0;
6052 }
6053
6054 static int ext4_quota_off(struct super_block *sb, int type)
6055 {
6056         struct inode *inode = sb_dqopt(sb)->files[type];
6057         handle_t *handle;
6058         int err;
6059
6060         /* Force all delayed allocation blocks to be allocated.
6061          * Caller already holds s_umount sem */
6062         if (test_opt(sb, DELALLOC))
6063                 sync_filesystem(sb);
6064
6065         if (!inode || !igrab(inode))
6066                 goto out;
6067
6068         err = dquot_quota_off(sb, type);
6069         if (err || ext4_has_feature_quota(sb))
6070                 goto out_put;
6071
6072         inode_lock(inode);
6073         /*
6074          * Update modification times of quota files when userspace can
6075          * start looking at them. If we fail, we return success anyway since
6076          * this is not a hard failure and quotas are already disabled.
6077          */
6078         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
6079         if (IS_ERR(handle)) {
6080                 err = PTR_ERR(handle);
6081                 goto out_unlock;
6082         }
6083         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
6084         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
6085         inode->i_mtime = inode->i_ctime = current_time(inode);
6086         err = ext4_mark_inode_dirty(handle, inode);
6087         ext4_journal_stop(handle);
6088 out_unlock:
6089         inode_unlock(inode);
6090 out_put:
6091         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
6092         iput(inode);
6093         return err;
6094 out:
6095         return dquot_quota_off(sb, type);
6096 }
6097
6098 /* Read data from quotafile - avoid pagecache and such because we cannot afford
6099  * acquiring the locks... As quota files are never truncated and quota code
6100  * itself serializes the operations (and no one else should touch the files)
6101  * we don't have to be afraid of races */
6102 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
6103                                size_t len, loff_t off)
6104 {
6105         struct inode *inode = sb_dqopt(sb)->files[type];
6106         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
6107         int offset = off & (sb->s_blocksize - 1);
6108         int tocopy;
6109         size_t toread;
6110         struct buffer_head *bh;
6111         loff_t i_size = i_size_read(inode);
6112
6113         if (off > i_size)
6114                 return 0;
6115         if (off+len > i_size)
6116                 len = i_size-off;
6117         toread = len;
6118         while (toread > 0) {
6119                 tocopy = sb->s_blocksize - offset < toread ?
6120                                 sb->s_blocksize - offset : toread;
6121                 bh = ext4_bread(NULL, inode, blk, 0);
6122                 if (IS_ERR(bh))
6123                         return PTR_ERR(bh);
6124                 if (!bh)        /* A hole? */
6125                         memset(data, 0, tocopy);
6126                 else
6127                         memcpy(data, bh->b_data+offset, tocopy);
6128                 brelse(bh);
6129                 offset = 0;
6130                 toread -= tocopy;
6131                 data += tocopy;
6132                 blk++;
6133         }
6134         return len;
6135 }
6136
6137 /* Write to quotafile (we know the transaction is already started and has
6138  * enough credits) */
6139 static ssize_t ext4_quota_write(struct super_block *sb, int type,
6140                                 const char *data, size_t len, loff_t off)
6141 {
6142         struct inode *inode = sb_dqopt(sb)->files[type];
6143         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
6144         int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1);
6145         int retries = 0;
6146         struct buffer_head *bh;
6147         handle_t *handle = journal_current_handle();
6148
6149         if (EXT4_SB(sb)->s_journal && !handle) {
6150                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
6151                         " cancelled because transaction is not started",
6152                         (unsigned long long)off, (unsigned long long)len);
6153                 return -EIO;
6154         }
6155         /*
6156          * Since we account only one data block in transaction credits,
6157          * then it is impossible to cross a block boundary.
6158          */
6159         if (sb->s_blocksize - offset < len) {
6160                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
6161                         " cancelled because not block aligned",
6162                         (unsigned long long)off, (unsigned long long)len);
6163                 return -EIO;
6164         }
6165
6166         do {
6167                 bh = ext4_bread(handle, inode, blk,
6168                                 EXT4_GET_BLOCKS_CREATE |
6169                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
6170         } while (PTR_ERR(bh) == -ENOSPC &&
6171                  ext4_should_retry_alloc(inode->i_sb, &retries));
6172         if (IS_ERR(bh))
6173                 return PTR_ERR(bh);
6174         if (!bh)
6175                 goto out;
6176         BUFFER_TRACE(bh, "get write access");
6177         err = ext4_journal_get_write_access(handle, bh);
6178         if (err) {
6179                 brelse(bh);
6180                 return err;
6181         }
6182         lock_buffer(bh);
6183         memcpy(bh->b_data+offset, data, len);
6184         flush_dcache_page(bh->b_page);
6185         unlock_buffer(bh);
6186         err = ext4_handle_dirty_metadata(handle, NULL, bh);
6187         brelse(bh);
6188 out:
6189         if (inode->i_size < off + len) {
6190                 i_size_write(inode, off + len);
6191                 EXT4_I(inode)->i_disksize = inode->i_size;
6192                 err2 = ext4_mark_inode_dirty(handle, inode);
6193                 if (unlikely(err2 && !err))
6194                         err = err2;
6195         }
6196         return err ? err : len;
6197 }
6198 #endif
6199
6200 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
6201                        const char *dev_name, void *data)
6202 {
6203         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
6204 }
6205
6206 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
6207 static inline void register_as_ext2(void)
6208 {
6209         int err = register_filesystem(&ext2_fs_type);
6210         if (err)
6211                 printk(KERN_WARNING
6212                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
6213 }
6214
6215 static inline void unregister_as_ext2(void)
6216 {
6217         unregister_filesystem(&ext2_fs_type);
6218 }
6219
6220 static inline int ext2_feature_set_ok(struct super_block *sb)
6221 {
6222         if (ext4_has_unknown_ext2_incompat_features(sb))
6223                 return 0;
6224         if (sb_rdonly(sb))
6225                 return 1;
6226         if (ext4_has_unknown_ext2_ro_compat_features(sb))
6227                 return 0;
6228         return 1;
6229 }
6230 #else
6231 static inline void register_as_ext2(void) { }
6232 static inline void unregister_as_ext2(void) { }
6233 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
6234 #endif
6235
6236 static inline void register_as_ext3(void)
6237 {
6238         int err = register_filesystem(&ext3_fs_type);
6239         if (err)
6240                 printk(KERN_WARNING
6241                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
6242 }
6243
6244 static inline void unregister_as_ext3(void)
6245 {
6246         unregister_filesystem(&ext3_fs_type);
6247 }
6248
6249 static inline int ext3_feature_set_ok(struct super_block *sb)
6250 {
6251         if (ext4_has_unknown_ext3_incompat_features(sb))
6252                 return 0;
6253         if (!ext4_has_feature_journal(sb))
6254                 return 0;
6255         if (sb_rdonly(sb))
6256                 return 1;
6257         if (ext4_has_unknown_ext3_ro_compat_features(sb))
6258                 return 0;
6259         return 1;
6260 }
6261
6262 static struct file_system_type ext4_fs_type = {
6263         .owner          = THIS_MODULE,
6264         .name           = "ext4",
6265         .mount          = ext4_mount,
6266         .kill_sb        = kill_block_super,
6267         .fs_flags       = FS_REQUIRES_DEV,
6268 };
6269 MODULE_ALIAS_FS("ext4");
6270
6271 /* Shared across all ext4 file systems */
6272 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
6273
6274 static int __init ext4_init_fs(void)
6275 {
6276         int i, err;
6277
6278         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
6279         ext4_li_info = NULL;
6280         mutex_init(&ext4_li_mtx);
6281
6282         /* Build-time check for flags consistency */
6283         ext4_check_flag_values();
6284
6285         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
6286                 init_waitqueue_head(&ext4__ioend_wq[i]);
6287
6288         err = ext4_init_es();
6289         if (err)
6290                 return err;
6291
6292         err = ext4_init_pending();
6293         if (err)
6294                 goto out7;
6295
6296         err = ext4_init_post_read_processing();
6297         if (err)
6298                 goto out6;
6299
6300         err = ext4_init_pageio();
6301         if (err)
6302                 goto out5;
6303
6304         err = ext4_init_system_zone();
6305         if (err)
6306                 goto out4;
6307
6308         err = ext4_init_sysfs();
6309         if (err)
6310                 goto out3;
6311
6312         err = ext4_init_mballoc();
6313         if (err)
6314                 goto out2;
6315         err = init_inodecache();
6316         if (err)
6317                 goto out1;
6318         register_as_ext3();
6319         register_as_ext2();
6320         err = register_filesystem(&ext4_fs_type);
6321         if (err)
6322                 goto out;
6323
6324         return 0;
6325 out:
6326         unregister_as_ext2();
6327         unregister_as_ext3();
6328         destroy_inodecache();
6329 out1:
6330         ext4_exit_mballoc();
6331 out2:
6332         ext4_exit_sysfs();
6333 out3:
6334         ext4_exit_system_zone();
6335 out4:
6336         ext4_exit_pageio();
6337 out5:
6338         ext4_exit_post_read_processing();
6339 out6:
6340         ext4_exit_pending();
6341 out7:
6342         ext4_exit_es();
6343
6344         return err;
6345 }
6346
6347 static void __exit ext4_exit_fs(void)
6348 {
6349         ext4_destroy_lazyinit_thread();
6350         unregister_as_ext2();
6351         unregister_as_ext3();
6352         unregister_filesystem(&ext4_fs_type);
6353         destroy_inodecache();
6354         ext4_exit_mballoc();
6355         ext4_exit_sysfs();
6356         ext4_exit_system_zone();
6357         ext4_exit_pageio();
6358         ext4_exit_post_read_processing();
6359         ext4_exit_es();
6360         ext4_exit_pending();
6361 }
6362
6363 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
6364 MODULE_DESCRIPTION("Fourth Extended Filesystem");
6365 MODULE_LICENSE("GPL");
6366 MODULE_SOFTDEP("pre: crc32c");
6367 module_init(ext4_init_fs)
6368 module_exit(ext4_exit_fs)