fs/ext4/super.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/ext4/super.c
   4  *
   5  * Copyright (C) 1992, 1993, 1994, 1995
   6  * Remy Card (card@masi.ibp.fr)
   7  * Laboratoire MASI - Institut Blaise Pascal
   8  * Universite Pierre et Marie Curie (Paris VI)
   9  *
  10  *  from
  11  *
  12  *  linux/fs/minix/inode.c
  13  *
  14  *  Copyright (C) 1991, 1992  Linus Torvalds
  15  *
  16  *  Big-endian to little-endian byte-swapping/bitmaps by
  17  *        David S. Miller (davem@caip.rutgers.edu), 1995
  18  */
  19
  20 #include <linux/module.h>
  21 #include <linux/string.h>
  22 #include <linux/fs.h>
  23 #include <linux/time.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/backing-dev.h>
  29 #include <linux/parser.h>
  30 #include <linux/buffer_head.h>
  31 #include <linux/exportfs.h>
  32 #include <linux/vfs.h>
  33 #include <linux/random.h>
  34 #include <linux/mount.h>
  35 #include <linux/namei.h>
  36 #include <linux/quotaops.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/dax.h>
  42 #include <linux/cleancache.h>
  43 #include <linux/uaccess.h>
  44 #include <linux/iversion.h>
  45 #include <linux/unicode.h>
  46 #include <linux/part_stat.h>
  47 #include <linux/kthread.h>
  48 #include <linux/freezer.h>
  49
  50 #include "ext4.h"
  51 #include "ext4_extents.h"       /* Needed for trace points definition */
  52 #include "ext4_jbd2.h"
  53 #include "xattr.h"
  54 #include "acl.h"
  55 #include "mballoc.h"
  56 #include "fsmap.h"
  57
  58 #define CREATE_TRACE_POINTS
  59 #include <trace/events/ext4.h>
  60
  61 static struct ext4_lazy_init *ext4_li_info;
  62 static struct mutex ext4_li_mtx;
  63 static struct ratelimit_state ext4_mount_msg_ratelimit;
  64
  65 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  66                              unsigned long journal_devnum);
  67 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  68 static int ext4_commit_super(struct super_block *sb, int sync);
  69 static void ext4_mark_recovery_complete(struct super_block *sb,
  70                                         struct ext4_super_block *es);
  71 static void ext4_clear_journal_err(struct super_block *sb,
  72                                    struct ext4_super_block *es);
  73 static int ext4_sync_fs(struct super_block *sb, int wait);
  74 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  75 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  76 static int ext4_unfreeze(struct super_block *sb);
  77 static int ext4_freeze(struct super_block *sb);
  78 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  79                        const char *dev_name, void *data);
  80 static inline int ext2_feature_set_ok(struct super_block *sb);
  81 static inline int ext3_feature_set_ok(struct super_block *sb);
  82 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  83 static void ext4_destroy_lazyinit_thread(void);
  84 static void ext4_unregister_li_request(struct super_block *sb);
  85 static void ext4_clear_request_list(void);
  86 static struct inode *ext4_get_journal_inode(struct super_block *sb,
  87                                             unsigned int journal_inum);
  88
  89 /*
  90  * Lock ordering
  91  *
  92  * Note the difference between i_mmap_sem (EXT4_I(inode)->i_mmap_sem) and
  93  * i_mmap_rwsem (inode->i_mmap_rwsem)!
  94  *
  95  * page fault path:
  96  * mmap_sem -> sb_start_pagefault -> i_mmap_sem (r) -> transaction start ->
  97  *   page lock -> i_data_sem (rw)
  98  *
  99  * buffered write path:
 100  * sb_start_write -> i_mutex -> mmap_sem
 101  * sb_start_write -> i_mutex -> transaction start -> page lock ->
 102  *   i_data_sem (rw)
 103  *
 104  * truncate:
 105  * sb_start_write -> i_mutex -> i_mmap_sem (w) -> i_mmap_rwsem (w) -> page lock
 106  * sb_start_write -> i_mutex -> i_mmap_sem (w) -> transaction start ->
 107  *   i_data_sem (rw)
 108  *
 109  * direct IO:
 110  * sb_start_write -> i_mutex -> mmap_sem
 111  * sb_start_write -> i_mutex -> transaction start -> i_data_sem (rw)
 112  *
 113  * writepages:
 114  * transaction start -> page lock(s) -> i_data_sem (rw)
 115  */
 116
 117 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
 118 static struct file_system_type ext2_fs_type = {
 119         .owner          = THIS_MODULE,
 120         .name           = "ext2",
 121         .mount          = ext4_mount,
 122         .kill_sb        = kill_block_super,
 123         .fs_flags       = FS_REQUIRES_DEV,
 124 };
 125 MODULE_ALIAS_FS("ext2");
 126 MODULE_ALIAS("ext2");
 127 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
 128 #else
 129 #define IS_EXT2_SB(sb) (0)
 130 #endif
 131
 132
 133 static struct file_system_type ext3_fs_type = {
 134         .owner          = THIS_MODULE,
 135         .name           = "ext3",
 136         .mount          = ext4_mount,
 137         .kill_sb        = kill_block_super,
 138         .fs_flags       = FS_REQUIRES_DEV,
 139 };
 140 MODULE_ALIAS_FS("ext3");
 141 MODULE_ALIAS("ext3");
 142 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 143
 144 /*
 145  * This works like sb_bread() except it uses ERR_PTR for error
 146  * returns.  Currently with sb_bread it's impossible to distinguish
 147  * between ENOMEM and EIO situations (since both result in a NULL
 148  * return.
 149  */
 150 struct buffer_head *
 151 ext4_sb_bread(struct super_block *sb, sector_t block, int op_flags)
 152 {
 153         struct buffer_head *bh = sb_getblk(sb, block);
 154
 155         if (bh == NULL)
 156                 return ERR_PTR(-ENOMEM);
 157         if (ext4_buffer_uptodate(bh))
 158                 return bh;
 159         ll_rw_block(REQ_OP_READ, REQ_META | op_flags, 1, &bh);
 160         wait_on_buffer(bh);
 161         if (buffer_uptodate(bh))
 162                 return bh;
 163         put_bh(bh);
 164         return ERR_PTR(-EIO);
 165 }
 166
 167 static int ext4_verify_csum_type(struct super_block *sb,
 168                                  struct ext4_super_block *es)
 169 {
 170         if (!ext4_has_feature_metadata_csum(sb))
 171                 return 1;
 172
 173         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 174 }
 175
 176 static __le32 ext4_superblock_csum(struct super_block *sb,
 177                                    struct ext4_super_block *es)
 178 {
 179         struct ext4_sb_info *sbi = EXT4_SB(sb);
 180         int offset = offsetof(struct ext4_super_block, s_checksum);
 181         __u32 csum;
 182
 183         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 184
 185         return cpu_to_le32(csum);
 186 }
 187
 188 static int ext4_superblock_csum_verify(struct super_block *sb,
 189                                        struct ext4_super_block *es)
 190 {
 191         if (!ext4_has_metadata_csum(sb))
 192                 return 1;
 193
 194         return es->s_checksum == ext4_superblock_csum(sb, es);
 195 }
 196
 197 void ext4_superblock_csum_set(struct super_block *sb)
 198 {
 199         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 200
 201         if (!ext4_has_metadata_csum(sb))
 202                 return;
 203
 204         es->s_checksum = ext4_superblock_csum(sb, es);
 205 }
 206
 207 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 208                                struct ext4_group_desc *bg)
 209 {
 210         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 211                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 212                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 213 }
 214
 215 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 216                                struct ext4_group_desc *bg)
 217 {
 218         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 219                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 220                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 221 }
 222
 223 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 224                               struct ext4_group_desc *bg)
 225 {
 226         return le32_to_cpu(bg->bg_inode_table_lo) |
 227                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 228                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 229 }
 230
 231 __u32 ext4_free_group_clusters(struct super_block *sb,
 232                                struct ext4_group_desc *bg)
 233 {
 234         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 235                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 236                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 237 }
 238
 239 __u32 ext4_free_inodes_count(struct super_block *sb,
 240                               struct ext4_group_desc *bg)
 241 {
 242         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 243                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 244                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 245 }
 246
 247 __u32 ext4_used_dirs_count(struct super_block *sb,
 248                               struct ext4_group_desc *bg)
 249 {
 250         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 251                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 252                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 253 }
 254
 255 __u32 ext4_itable_unused_count(struct super_block *sb,
 256                               struct ext4_group_desc *bg)
 257 {
 258         return le16_to_cpu(bg->bg_itable_unused_lo) |
 259                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 260                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 261 }
 262
 263 void ext4_block_bitmap_set(struct super_block *sb,
 264                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 265 {
 266         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 267         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 268                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 269 }
 270
 271 void ext4_inode_bitmap_set(struct super_block *sb,
 272                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 273 {
 274         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 275         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 276                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 277 }
 278
 279 void ext4_inode_table_set(struct super_block *sb,
 280                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 281 {
 282         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 283         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 284                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 285 }
 286
 287 void ext4_free_group_clusters_set(struct super_block *sb,
 288                                   struct ext4_group_desc *bg, __u32 count)
 289 {
 290         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 291         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 292                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 293 }
 294
 295 void ext4_free_inodes_set(struct super_block *sb,
 296                           struct ext4_group_desc *bg, __u32 count)
 297 {
 298         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 299         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 300                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 301 }
 302
 303 void ext4_used_dirs_set(struct super_block *sb,
 304                           struct ext4_group_desc *bg, __u32 count)
 305 {
 306         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 307         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 308                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 309 }
 310
 311 void ext4_itable_unused_set(struct super_block *sb,
 312                           struct ext4_group_desc *bg, __u32 count)
 313 {
 314         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 315         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 316                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 317 }
 318
 319 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi)
 320 {
 321         time64_t now = ktime_get_real_seconds();
 322
 323         now = clamp_val(now, 0, (1ull << 40) - 1);
 324
 325         *lo = cpu_to_le32(lower_32_bits(now));
 326         *hi = upper_32_bits(now);
 327 }
 328
 329 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi)
 330 {
 331         return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo);
 332 }
 333 #define ext4_update_tstamp(es, tstamp) \
 334         __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
 335 #define ext4_get_tstamp(es, tstamp) \
 336         __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
 337
 338 static void __save_error_info(struct super_block *sb, int error,
 339                               __u32 ino, __u64 block,
 340                               const char *func, unsigned int line)
 341 {
 342         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 343         int err;
 344
 345         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 346         if (bdev_read_only(sb->s_bdev))
 347                 return;
 348         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 349         ext4_update_tstamp(es, s_last_error_time);
 350         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 351         es->s_last_error_line = cpu_to_le32(line);
 352         es->s_last_error_ino = cpu_to_le32(ino);
 353         es->s_last_error_block = cpu_to_le64(block);
 354         switch (error) {
 355         case EIO:
 356                 err = EXT4_ERR_EIO;
 357                 break;
 358         case ENOMEM:
 359                 err = EXT4_ERR_ENOMEM;
 360                 break;
 361         case EFSBADCRC:
 362                 err = EXT4_ERR_EFSBADCRC;
 363                 break;
 364         case 0:
 365         case EFSCORRUPTED:
 366                 err = EXT4_ERR_EFSCORRUPTED;
 367                 break;
 368         case ENOSPC:
 369                 err = EXT4_ERR_ENOSPC;
 370                 break;
 371         case ENOKEY:
 372                 err = EXT4_ERR_ENOKEY;
 373                 break;
 374         case EROFS:
 375                 err = EXT4_ERR_EROFS;
 376                 break;
 377         case EFBIG:
 378                 err = EXT4_ERR_EFBIG;
 379                 break;
 380         case EEXIST:
 381                 err = EXT4_ERR_EEXIST;
 382                 break;
 383         case ERANGE:
 384                 err = EXT4_ERR_ERANGE;
 385                 break;
 386         case EOVERFLOW:
 387                 err = EXT4_ERR_EOVERFLOW;
 388                 break;
 389         case EBUSY:
 390                 err = EXT4_ERR_EBUSY;
 391                 break;
 392         case ENOTDIR:
 393                 err = EXT4_ERR_ENOTDIR;
 394                 break;
 395         case ENOTEMPTY:
 396                 err = EXT4_ERR_ENOTEMPTY;
 397                 break;
 398         case ESHUTDOWN:
 399                 err = EXT4_ERR_ESHUTDOWN;
 400                 break;
 401         case EFAULT:
 402                 err = EXT4_ERR_EFAULT;
 403                 break;
 404         default:
 405                 err = EXT4_ERR_UNKNOWN;
 406         }
 407         es->s_last_error_errcode = err;
 408         if (!es->s_first_error_time) {
 409                 es->s_first_error_time = es->s_last_error_time;
 410                 es->s_first_error_time_hi = es->s_last_error_time_hi;
 411                 strncpy(es->s_first_error_func, func,
 412                         sizeof(es->s_first_error_func));
 413                 es->s_first_error_line = cpu_to_le32(line);
 414                 es->s_first_error_ino = es->s_last_error_ino;
 415                 es->s_first_error_block = es->s_last_error_block;
 416                 es->s_first_error_errcode = es->s_last_error_errcode;
 417         }
 418         /*
 419          * Start the daily error reporting function if it hasn't been
 420          * started already
 421          */
 422         if (!es->s_error_count)
 423                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 424         le32_add_cpu(&es->s_error_count, 1);
 425 }
 426
 427 static void save_error_info(struct super_block *sb, int error,
 428                             __u32 ino, __u64 block,
 429                             const char *func, unsigned int line)
 430 {
 431         __save_error_info(sb, error, ino, block, func, line);
 432         if (!bdev_read_only(sb->s_bdev))
 433                 ext4_commit_super(sb, 1);
 434 }
 435
 436 /*
 437  * The del_gendisk() function uninitializes the disk-specific data
 438  * structures, including the bdi structure, without telling anyone
 439  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 440  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 441  * This is a kludge to prevent these oops until we can put in a proper
 442  * hook in del_gendisk() to inform the VFS and file system layers.
 443  */
 444 static int block_device_ejected(struct super_block *sb)
 445 {
 446         struct inode *bd_inode = sb->s_bdev->bd_inode;
 447         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 448
 449         return bdi->dev == NULL;
 450 }
 451
 452 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 453 {
 454         struct super_block              *sb = journal->j_private;
 455         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 456         int                             error = is_journal_aborted(journal);
 457         struct ext4_journal_cb_entry    *jce;
 458
 459         BUG_ON(txn->t_state == T_FINISHED);
 460
 461         ext4_process_freed_data(sb, txn->t_tid);
 462
 463         spin_lock(&sbi->s_md_lock);
 464         while (!list_empty(&txn->t_private_list)) {
 465                 jce = list_entry(txn->t_private_list.next,
 466                                  struct ext4_journal_cb_entry, jce_list);
 467                 list_del_init(&jce->jce_list);
 468                 spin_unlock(&sbi->s_md_lock);
 469                 jce->jce_func(sb, jce, error);
 470                 spin_lock(&sbi->s_md_lock);
 471         }
 472         spin_unlock(&sbi->s_md_lock);
 473 }
 474
 475 static bool system_going_down(void)
 476 {
 477         return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF
 478                 || system_state == SYSTEM_RESTART;
 479 }
 480
 481 /* Deal with the reporting of failure conditions on a filesystem such as
 482  * inconsistencies detected or read IO failures.
 483  *
 484  * On ext2, we can store the error state of the filesystem in the
 485  * superblock.  That is not possible on ext4, because we may have other
 486  * write ordering constraints on the superblock which prevent us from
 487  * writing it out straight away; and given that the journal is about to
 488  * be aborted, we can't rely on the current, or future, transactions to
 489  * write out the superblock safely.
 490  *
 491  * We'll just use the jbd2_journal_abort() error code to record an error in
 492  * the journal instead.  On recovery, the journal will complain about
 493  * that error until we've noted it down and cleared it.
 494  */
 495
 496 static void ext4_handle_error(struct super_block *sb)
 497 {
 498         if (test_opt(sb, WARN_ON_ERROR))
 499                 WARN_ON_ONCE(1);
 500
 501         if (sb_rdonly(sb))
 502                 return;
 503
 504         if (!test_opt(sb, ERRORS_CONT)) {
 505                 journal_t *journal = EXT4_SB(sb)->s_journal;
 506
 507                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 508                 if (journal)
 509                         jbd2_journal_abort(journal, -EIO);
 510         }
 511         /*
 512          * We force ERRORS_RO behavior when system is rebooting. Otherwise we
 513          * could panic during 'reboot -f' as the underlying device got already
 514          * disabled.
 515          */
 516         if (test_opt(sb, ERRORS_RO) || system_going_down()) {
 517                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 518                 /*
 519                  * Make sure updated value of ->s_mount_flags will be visible
 520                  * before ->s_flags update
 521                  */
 522                 smp_wmb();
 523                 sb->s_flags |= SB_RDONLY;
 524         } else if (test_opt(sb, ERRORS_PANIC)) {
 525                 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 #ifdef CONFIG_UNICODE
1104         utf8_unload(sbi->s_encoding);
1105 #endif
1106         kfree(sbi);
1107 }
1108
1109 static struct kmem_cache *ext4_inode_cachep;
1110
1111 /*
1112  * Called inside transaction, so use GFP_NOFS
1113  */
1114 static struct inode *ext4_alloc_inode(struct super_block *sb)
1115 {
1116         struct ext4_inode_info *ei;
1117
1118         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
1119         if (!ei)
1120                 return NULL;
1121
1122         inode_set_iversion(&ei->vfs_inode, 1);
1123         spin_lock_init(&ei->i_raw_lock);
1124         INIT_LIST_HEAD(&ei->i_prealloc_list);
1125         spin_lock_init(&ei->i_prealloc_lock);
1126         ext4_es_init_tree(&ei->i_es_tree);
1127         rwlock_init(&ei->i_es_lock);
1128         INIT_LIST_HEAD(&ei->i_es_list);
1129         ei->i_es_all_nr = 0;
1130         ei->i_es_shk_nr = 0;
1131         ei->i_es_shrink_lblk = 0;
1132         ei->i_reserved_data_blocks = 0;
1133         spin_lock_init(&(ei->i_block_reservation_lock));
1134         ext4_init_pending_tree(&ei->i_pending_tree);
1135 #ifdef CONFIG_QUOTA
1136         ei->i_reserved_quota = 0;
1137         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
1138 #endif
1139         ei->jinode = NULL;
1140         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
1141         spin_lock_init(&ei->i_completed_io_lock);
1142         ei->i_sync_tid = 0;
1143         ei->i_datasync_tid = 0;
1144         atomic_set(&ei->i_unwritten, 0);
1145         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
1146         return &ei->vfs_inode;
1147 }
1148
1149 static int ext4_drop_inode(struct inode *inode)
1150 {
1151         int drop = generic_drop_inode(inode);
1152
1153         if (!drop)
1154                 drop = fscrypt_drop_inode(inode);
1155
1156         trace_ext4_drop_inode(inode, drop);
1157         return drop;
1158 }
1159
1160 static void ext4_free_in_core_inode(struct inode *inode)
1161 {
1162         fscrypt_free_inode(inode);
1163         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1164 }
1165
1166 static void ext4_destroy_inode(struct inode *inode)
1167 {
1168         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1169                 ext4_msg(inode->i_sb, KERN_ERR,
1170                          "Inode %lu (%p): orphan list check failed!",
1171                          inode->i_ino, EXT4_I(inode));
1172                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1173                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1174                                 true);
1175                 dump_stack();
1176         }
1177 }
1178
1179 static void init_once(void *foo)
1180 {
1181         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1182
1183         INIT_LIST_HEAD(&ei->i_orphan);
1184         init_rwsem(&ei->xattr_sem);
1185         init_rwsem(&ei->i_data_sem);
1186         init_rwsem(&ei->i_mmap_sem);
1187         inode_init_once(&ei->vfs_inode);
1188 }
1189
1190 static int __init init_inodecache(void)
1191 {
1192         ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache",
1193                                 sizeof(struct ext4_inode_info), 0,
1194                                 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
1195                                         SLAB_ACCOUNT),
1196                                 offsetof(struct ext4_inode_info, i_data),
1197                                 sizeof_field(struct ext4_inode_info, i_data),
1198                                 init_once);
1199         if (ext4_inode_cachep == NULL)
1200                 return -ENOMEM;
1201         return 0;
1202 }
1203
1204 static void destroy_inodecache(void)
1205 {
1206         /*
1207          * Make sure all delayed rcu free inodes are flushed before we
1208          * destroy cache.
1209          */
1210         rcu_barrier();
1211         kmem_cache_destroy(ext4_inode_cachep);
1212 }
1213
1214 void ext4_clear_inode(struct inode *inode)
1215 {
1216         invalidate_inode_buffers(inode);
1217         clear_inode(inode);
1218         ext4_discard_preallocations(inode);
1219         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1220         dquot_drop(inode);
1221         if (EXT4_I(inode)->jinode) {
1222                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1223                                                EXT4_I(inode)->jinode);
1224                 jbd2_free_inode(EXT4_I(inode)->jinode);
1225                 EXT4_I(inode)->jinode = NULL;
1226         }
1227         fscrypt_put_encryption_info(inode);
1228         fsverity_cleanup_inode(inode);
1229 }
1230
1231 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1232                                         u64 ino, u32 generation)
1233 {
1234         struct inode *inode;
1235
1236         /*
1237          * Currently we don't know the generation for parent directory, so
1238          * a generation of 0 means "accept any"
1239          */
1240         inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE);
1241         if (IS_ERR(inode))
1242                 return ERR_CAST(inode);
1243         if (generation && inode->i_generation != generation) {
1244                 iput(inode);
1245                 return ERR_PTR(-ESTALE);
1246         }
1247
1248         return inode;
1249 }
1250
1251 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1252                                         int fh_len, int fh_type)
1253 {
1254         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1255                                     ext4_nfs_get_inode);
1256 }
1257
1258 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1259                                         int fh_len, int fh_type)
1260 {
1261         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1262                                     ext4_nfs_get_inode);
1263 }
1264
1265 static int ext4_nfs_commit_metadata(struct inode *inode)
1266 {
1267         struct writeback_control wbc = {
1268                 .sync_mode = WB_SYNC_ALL
1269         };
1270
1271         trace_ext4_nfs_commit_metadata(inode);
1272         return ext4_write_inode(inode, &wbc);
1273 }
1274
1275 /*
1276  * Try to release metadata pages (indirect blocks, directories) which are
1277  * mapped via the block device.  Since these pages could have journal heads
1278  * which would prevent try_to_free_buffers() from freeing them, we must use
1279  * jbd2 layer's try_to_free_buffers() function to release them.
1280  */
1281 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1282                                  gfp_t wait)
1283 {
1284         journal_t *journal = EXT4_SB(sb)->s_journal;
1285
1286         WARN_ON(PageChecked(page));
1287         if (!page_has_buffers(page))
1288                 return 0;
1289         if (journal)
1290                 return jbd2_journal_try_to_free_buffers(journal, page,
1291                                                 wait & ~__GFP_DIRECT_RECLAIM);
1292         return try_to_free_buffers(page);
1293 }
1294
1295 #ifdef CONFIG_FS_ENCRYPTION
1296 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1297 {
1298         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1299                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1300 }
1301
1302 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1303                                                         void *fs_data)
1304 {
1305         handle_t *handle = fs_data;
1306         int res, res2, credits, retries = 0;
1307
1308         /*
1309          * Encrypting the root directory is not allowed because e2fsck expects
1310          * lost+found to exist and be unencrypted, and encrypting the root
1311          * directory would imply encrypting the lost+found directory as well as
1312          * the filename "lost+found" itself.
1313          */
1314         if (inode->i_ino == EXT4_ROOT_INO)
1315                 return -EPERM;
1316
1317         if (WARN_ON_ONCE(IS_DAX(inode) && i_size_read(inode)))
1318                 return -EINVAL;
1319
1320         if (ext4_test_inode_flag(inode, EXT4_INODE_DAX))
1321                 return -EOPNOTSUPP;
1322
1323         res = ext4_convert_inline_data(inode);
1324         if (res)
1325                 return res;
1326
1327         /*
1328          * If a journal handle was specified, then the encryption context is
1329          * being set on a new inode via inheritance and is part of a larger
1330          * transaction to create the inode.  Otherwise the encryption context is
1331          * being set on an existing inode in its own transaction.  Only in the
1332          * latter case should the "retry on ENOSPC" logic be used.
1333          */
1334
1335         if (handle) {
1336                 res = ext4_xattr_set_handle(handle, inode,
1337                                             EXT4_XATTR_INDEX_ENCRYPTION,
1338                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1339                                             ctx, len, 0);
1340                 if (!res) {
1341                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1342                         ext4_clear_inode_state(inode,
1343                                         EXT4_STATE_MAY_INLINE_DATA);
1344                         /*
1345                          * Update inode->i_flags - S_ENCRYPTED will be enabled,
1346                          * S_DAX may be disabled
1347                          */
1348                         ext4_set_inode_flags(inode, false);
1349                 }
1350                 return res;
1351         }
1352
1353         res = dquot_initialize(inode);
1354         if (res)
1355                 return res;
1356 retry:
1357         res = ext4_xattr_set_credits(inode, len, false /* is_create */,
1358                                      &credits);
1359         if (res)
1360                 return res;
1361
1362         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
1363         if (IS_ERR(handle))
1364                 return PTR_ERR(handle);
1365
1366         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1367                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1368                                     ctx, len, 0);
1369         if (!res) {
1370                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1371                 /*
1372                  * Update inode->i_flags - S_ENCRYPTED will be enabled,
1373                  * S_DAX may be disabled
1374                  */
1375                 ext4_set_inode_flags(inode, false);
1376                 res = ext4_mark_inode_dirty(handle, inode);
1377                 if (res)
1378                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1379         }
1380         res2 = ext4_journal_stop(handle);
1381
1382         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1383                 goto retry;
1384         if (!res)
1385                 res = res2;
1386         return res;
1387 }
1388
1389 static bool ext4_dummy_context(struct inode *inode)
1390 {
1391         return DUMMY_ENCRYPTION_ENABLED(EXT4_SB(inode->i_sb));
1392 }
1393
1394 static bool ext4_has_stable_inodes(struct super_block *sb)
1395 {
1396         return ext4_has_feature_stable_inodes(sb);
1397 }
1398
1399 static void ext4_get_ino_and_lblk_bits(struct super_block *sb,
1400                                        int *ino_bits_ret, int *lblk_bits_ret)
1401 {
1402         *ino_bits_ret = 8 * sizeof(EXT4_SB(sb)->s_es->s_inodes_count);
1403         *lblk_bits_ret = 8 * sizeof(ext4_lblk_t);
1404 }
1405
1406 static const struct fscrypt_operations ext4_cryptops = {
1407         .key_prefix             = "ext4:",
1408         .get_context            = ext4_get_context,
1409         .set_context            = ext4_set_context,
1410         .dummy_context          = ext4_dummy_context,
1411         .empty_dir              = ext4_empty_dir,
1412         .max_namelen            = EXT4_NAME_LEN,
1413         .has_stable_inodes      = ext4_has_stable_inodes,
1414         .get_ino_and_lblk_bits  = ext4_get_ino_and_lblk_bits,
1415 };
1416 #endif
1417
1418 #ifdef CONFIG_QUOTA
1419 static const char * const quotatypes[] = INITQFNAMES;
1420 #define QTYPE2NAME(t) (quotatypes[t])
1421
1422 static int ext4_write_dquot(struct dquot *dquot);
1423 static int ext4_acquire_dquot(struct dquot *dquot);
1424 static int ext4_release_dquot(struct dquot *dquot);
1425 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1426 static int ext4_write_info(struct super_block *sb, int type);
1427 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1428                          const struct path *path);
1429 static int ext4_quota_on_mount(struct super_block *sb, int type);
1430 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1431                                size_t len, loff_t off);
1432 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1433                                 const char *data, size_t len, loff_t off);
1434 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1435                              unsigned int flags);
1436 static int ext4_enable_quotas(struct super_block *sb);
1437
1438 static struct dquot **ext4_get_dquots(struct inode *inode)
1439 {
1440         return EXT4_I(inode)->i_dquot;
1441 }
1442
1443 static const struct dquot_operations ext4_quota_operations = {
1444         .get_reserved_space     = ext4_get_reserved_space,
1445         .write_dquot            = ext4_write_dquot,
1446         .acquire_dquot          = ext4_acquire_dquot,
1447         .release_dquot          = ext4_release_dquot,
1448         .mark_dirty             = ext4_mark_dquot_dirty,
1449         .write_info             = ext4_write_info,
1450         .alloc_dquot            = dquot_alloc,
1451         .destroy_dquot          = dquot_destroy,
1452         .get_projid             = ext4_get_projid,
1453         .get_inode_usage        = ext4_get_inode_usage,
1454         .get_next_id            = dquot_get_next_id,
1455 };
1456
1457 static const struct quotactl_ops ext4_qctl_operations = {
1458         .quota_on       = ext4_quota_on,
1459         .quota_off      = ext4_quota_off,
1460         .quota_sync     = dquot_quota_sync,
1461         .get_state      = dquot_get_state,
1462         .set_info       = dquot_set_dqinfo,
1463         .get_dqblk      = dquot_get_dqblk,
1464         .set_dqblk      = dquot_set_dqblk,
1465         .get_nextdqblk  = dquot_get_next_dqblk,
1466 };
1467 #endif
1468
1469 static const struct super_operations ext4_sops = {
1470         .alloc_inode    = ext4_alloc_inode,
1471         .free_inode     = ext4_free_in_core_inode,
1472         .destroy_inode  = ext4_destroy_inode,
1473         .write_inode    = ext4_write_inode,
1474         .dirty_inode    = ext4_dirty_inode,
1475         .drop_inode     = ext4_drop_inode,
1476         .evict_inode    = ext4_evict_inode,
1477         .put_super      = ext4_put_super,
1478         .sync_fs        = ext4_sync_fs,
1479         .freeze_fs      = ext4_freeze,
1480         .unfreeze_fs    = ext4_unfreeze,
1481         .statfs         = ext4_statfs,
1482         .remount_fs     = ext4_remount,
1483         .show_options   = ext4_show_options,
1484 #ifdef CONFIG_QUOTA
1485         .quota_read     = ext4_quota_read,
1486         .quota_write    = ext4_quota_write,
1487         .get_dquots     = ext4_get_dquots,
1488 #endif
1489         .bdev_try_to_free_page = bdev_try_to_free_page,
1490 };
1491
1492 static const struct export_operations ext4_export_ops = {
1493         .fh_to_dentry = ext4_fh_to_dentry,
1494         .fh_to_parent = ext4_fh_to_parent,
1495         .get_parent = ext4_get_parent,
1496         .commit_metadata = ext4_nfs_commit_metadata,
1497 };
1498
1499 enum {
1500         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1501         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1502         Opt_nouid32, Opt_debug, Opt_removed,
1503         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1504         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1505         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1506         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1507         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1508         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1509         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1510         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1511         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1512         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version,
1513         Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never,
1514         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
1515         Opt_nowarn_on_error, Opt_mblk_io_submit,
1516         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1517         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1518         Opt_inode_readahead_blks, Opt_journal_ioprio,
1519         Opt_dioread_nolock, Opt_dioread_lock,
1520         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1521         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1522 };
1523
1524 static const match_table_t tokens = {
1525         {Opt_bsd_df, "bsddf"},
1526         {Opt_minix_df, "minixdf"},
1527         {Opt_grpid, "grpid"},
1528         {Opt_grpid, "bsdgroups"},
1529         {Opt_nogrpid, "nogrpid"},
1530         {Opt_nogrpid, "sysvgroups"},
1531         {Opt_resgid, "resgid=%u"},
1532         {Opt_resuid, "resuid=%u"},
1533         {Opt_sb, "sb=%u"},
1534         {Opt_err_cont, "errors=continue"},
1535         {Opt_err_panic, "errors=panic"},
1536         {Opt_err_ro, "errors=remount-ro"},
1537         {Opt_nouid32, "nouid32"},
1538         {Opt_debug, "debug"},
1539         {Opt_removed, "oldalloc"},
1540         {Opt_removed, "orlov"},
1541         {Opt_user_xattr, "user_xattr"},
1542         {Opt_nouser_xattr, "nouser_xattr"},
1543         {Opt_acl, "acl"},
1544         {Opt_noacl, "noacl"},
1545         {Opt_noload, "norecovery"},
1546         {Opt_noload, "noload"},
1547         {Opt_removed, "nobh"},
1548         {Opt_removed, "bh"},
1549         {Opt_commit, "commit=%u"},
1550         {Opt_min_batch_time, "min_batch_time=%u"},
1551         {Opt_max_batch_time, "max_batch_time=%u"},
1552         {Opt_journal_dev, "journal_dev=%u"},
1553         {Opt_journal_path, "journal_path=%s"},
1554         {Opt_journal_checksum, "journal_checksum"},
1555         {Opt_nojournal_checksum, "nojournal_checksum"},
1556         {Opt_journal_async_commit, "journal_async_commit"},
1557         {Opt_abort, "abort"},
1558         {Opt_data_journal, "data=journal"},
1559         {Opt_data_ordered, "data=ordered"},
1560         {Opt_data_writeback, "data=writeback"},
1561         {Opt_data_err_abort, "data_err=abort"},
1562         {Opt_data_err_ignore, "data_err=ignore"},
1563         {Opt_offusrjquota, "usrjquota="},
1564         {Opt_usrjquota, "usrjquota=%s"},
1565         {Opt_offgrpjquota, "grpjquota="},
1566         {Opt_grpjquota, "grpjquota=%s"},
1567         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1568         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1569         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1570         {Opt_grpquota, "grpquota"},
1571         {Opt_noquota, "noquota"},
1572         {Opt_quota, "quota"},
1573         {Opt_usrquota, "usrquota"},
1574         {Opt_prjquota, "prjquota"},
1575         {Opt_barrier, "barrier=%u"},
1576         {Opt_barrier, "barrier"},
1577         {Opt_nobarrier, "nobarrier"},
1578         {Opt_i_version, "i_version"},
1579         {Opt_dax, "dax"},
1580         {Opt_dax_always, "dax=always"},
1581         {Opt_dax_inode, "dax=inode"},
1582         {Opt_dax_never, "dax=never"},
1583         {Opt_stripe, "stripe=%u"},
1584         {Opt_delalloc, "delalloc"},
1585         {Opt_warn_on_error, "warn_on_error"},
1586         {Opt_nowarn_on_error, "nowarn_on_error"},
1587         {Opt_lazytime, "lazytime"},
1588         {Opt_nolazytime, "nolazytime"},
1589         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1590         {Opt_nodelalloc, "nodelalloc"},
1591         {Opt_removed, "mblk_io_submit"},
1592         {Opt_removed, "nomblk_io_submit"},
1593         {Opt_block_validity, "block_validity"},
1594         {Opt_noblock_validity, "noblock_validity"},
1595         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1596         {Opt_journal_ioprio, "journal_ioprio=%u"},
1597         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1598         {Opt_auto_da_alloc, "auto_da_alloc"},
1599         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1600         {Opt_dioread_nolock, "dioread_nolock"},
1601         {Opt_dioread_lock, "nodioread_nolock"},
1602         {Opt_dioread_lock, "dioread_lock"},
1603         {Opt_discard, "discard"},
1604         {Opt_nodiscard, "nodiscard"},
1605         {Opt_init_itable, "init_itable=%u"},
1606         {Opt_init_itable, "init_itable"},
1607         {Opt_noinit_itable, "noinit_itable"},
1608         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1609         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1610         {Opt_nombcache, "nombcache"},
1611         {Opt_nombcache, "no_mbcache"},  /* for backward compatibility */
1612         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1613         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1614         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1615         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1616         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1617         {Opt_err, NULL},
1618 };
1619
1620 static ext4_fsblk_t get_sb_block(void **data)
1621 {
1622         ext4_fsblk_t    sb_block;
1623         char            *options = (char *) *data;
1624
1625         if (!options || strncmp(options, "sb=", 3) != 0)
1626                 return 1;       /* Default location */
1627
1628         options += 3;
1629         /* TODO: use simple_strtoll with >32bit ext4 */
1630         sb_block = simple_strtoul(options, &options, 0);
1631         if (*options && *options != ',') {
1632                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1633                        (char *) *data);
1634                 return 1;
1635         }
1636         if (*options == ',')
1637                 options++;
1638         *data = (void *) options;
1639
1640         return sb_block;
1641 }
1642
1643 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1644 static const char deprecated_msg[] =
1645         "Mount option \"%s\" will be removed by %s\n"
1646         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1647
1648 #ifdef CONFIG_QUOTA
1649 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1650 {
1651         struct ext4_sb_info *sbi = EXT4_SB(sb);
1652         char *qname, *old_qname = get_qf_name(sb, sbi, qtype);
1653         int ret = -1;
1654
1655         if (sb_any_quota_loaded(sb) && !old_qname) {
1656                 ext4_msg(sb, KERN_ERR,
1657                         "Cannot change journaled "
1658                         "quota options when quota turned on");
1659                 return -1;
1660         }
1661         if (ext4_has_feature_quota(sb)) {
1662                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1663                          "ignored when QUOTA feature is enabled");
1664                 return 1;
1665         }
1666         qname = match_strdup(args);
1667         if (!qname) {
1668                 ext4_msg(sb, KERN_ERR,
1669                         "Not enough memory for storing quotafile name");
1670                 return -1;
1671         }
1672         if (old_qname) {
1673                 if (strcmp(old_qname, qname) == 0)
1674                         ret = 1;
1675                 else
1676                         ext4_msg(sb, KERN_ERR,
1677                                  "%s quota file already specified",
1678                                  QTYPE2NAME(qtype));
1679                 goto errout;
1680         }
1681         if (strchr(qname, '/')) {
1682                 ext4_msg(sb, KERN_ERR,
1683                         "quotafile must be on filesystem root");
1684                 goto errout;
1685         }
1686         rcu_assign_pointer(sbi->s_qf_names[qtype], qname);
1687         set_opt(sb, QUOTA);
1688         return 1;
1689 errout:
1690         kfree(qname);
1691         return ret;
1692 }
1693
1694 static int clear_qf_name(struct super_block *sb, int qtype)
1695 {
1696
1697         struct ext4_sb_info *sbi = EXT4_SB(sb);
1698         char *old_qname = get_qf_name(sb, sbi, qtype);
1699
1700         if (sb_any_quota_loaded(sb) && old_qname) {
1701                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1702                         " when quota turned on");
1703                 return -1;
1704         }
1705         rcu_assign_pointer(sbi->s_qf_names[qtype], NULL);
1706         synchronize_rcu();
1707         kfree(old_qname);
1708         return 1;
1709 }
1710 #endif
1711
1712 #define MOPT_SET        0x0001
1713 #define MOPT_CLEAR      0x0002
1714 #define MOPT_NOSUPPORT  0x0004
1715 #define MOPT_EXPLICIT   0x0008
1716 #define MOPT_CLEAR_ERR  0x0010
1717 #define MOPT_GTE0       0x0020
1718 #ifdef CONFIG_QUOTA
1719 #define MOPT_Q          0
1720 #define MOPT_QFMT       0x0040
1721 #else
1722 #define MOPT_Q          MOPT_NOSUPPORT
1723 #define MOPT_QFMT       MOPT_NOSUPPORT
1724 #endif
1725 #define MOPT_DATAJ      0x0080
1726 #define MOPT_NO_EXT2    0x0100
1727 #define MOPT_NO_EXT3    0x0200
1728 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1729 #define MOPT_STRING     0x0400
1730 #define MOPT_SKIP       0x0800
1731
1732 static const struct mount_opts {
1733         int     token;
1734         int     mount_opt;
1735         int     flags;
1736 } ext4_mount_opts[] = {
1737         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1738         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1739         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1740         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1741         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1742         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1743         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1744          MOPT_EXT4_ONLY | MOPT_SET},
1745         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1746          MOPT_EXT4_ONLY | MOPT_CLEAR},
1747         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1748         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1749         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1750          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1751         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1752          MOPT_EXT4_ONLY | MOPT_CLEAR},
1753         {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET},
1754         {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR},
1755         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1756          MOPT_EXT4_ONLY | MOPT_CLEAR},
1757         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1758          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1759         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1760                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1761          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1762         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1763         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1764         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1765         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1766         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1767          MOPT_NO_EXT2},
1768         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1769          MOPT_NO_EXT2},
1770         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1771         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1772         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1773         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1774         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1775         {Opt_commit, 0, MOPT_GTE0},
1776         {Opt_max_batch_time, 0, MOPT_GTE0},
1777         {Opt_min_batch_time, 0, MOPT_GTE0},
1778         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1779         {Opt_init_itable, 0, MOPT_GTE0},
1780         {Opt_dax, EXT4_MOUNT_DAX_ALWAYS, MOPT_SET | MOPT_SKIP},
1781         {Opt_dax_always, EXT4_MOUNT_DAX_ALWAYS,
1782                 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP},
1783         {Opt_dax_inode, EXT4_MOUNT2_DAX_INODE,
1784                 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP},
1785         {Opt_dax_never, EXT4_MOUNT2_DAX_NEVER,
1786                 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP},
1787         {Opt_stripe, 0, MOPT_GTE0},
1788         {Opt_resuid, 0, MOPT_GTE0},
1789         {Opt_resgid, 0, MOPT_GTE0},
1790         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1791         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1792         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1793         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1794         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1795         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1796          MOPT_NO_EXT2 | MOPT_DATAJ},
1797         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1798         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1799 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1800         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1801         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1802 #else
1803         {Opt_acl, 0, MOPT_NOSUPPORT},
1804         {Opt_noacl, 0, MOPT_NOSUPPORT},
1805 #endif
1806         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1807         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1808         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1809         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1810         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1811                                                         MOPT_SET | MOPT_Q},
1812         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1813                                                         MOPT_SET | MOPT_Q},
1814         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1815                                                         MOPT_SET | MOPT_Q},
1816         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1817                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1818                                                         MOPT_CLEAR | MOPT_Q},
1819         {Opt_usrjquota, 0, MOPT_Q},
1820         {Opt_grpjquota, 0, MOPT_Q},
1821         {Opt_offusrjquota, 0, MOPT_Q},
1822         {Opt_offgrpjquota, 0, MOPT_Q},
1823         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1824         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1825         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1826         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1827         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1828         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1829         {Opt_err, 0, 0}
1830 };
1831
1832 #ifdef CONFIG_UNICODE
1833 static const struct ext4_sb_encodings {
1834         __u16 magic;
1835         char *name;
1836         char *version;
1837 } ext4_sb_encoding_map[] = {
1838         {EXT4_ENC_UTF8_12_1, "utf8", "12.1.0"},
1839 };
1840
1841 static int ext4_sb_read_encoding(const struct ext4_super_block *es,
1842                                  const struct ext4_sb_encodings **encoding,
1843                                  __u16 *flags)
1844 {
1845         __u16 magic = le16_to_cpu(es->s_encoding);
1846         int i;
1847
1848         for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++)
1849                 if (magic == ext4_sb_encoding_map[i].magic)
1850                         break;
1851
1852         if (i >= ARRAY_SIZE(ext4_sb_encoding_map))
1853                 return -EINVAL;
1854
1855         *encoding = &ext4_sb_encoding_map[i];
1856         *flags = le16_to_cpu(es->s_encoding_flags);
1857
1858         return 0;
1859 }
1860 #endif
1861
1862 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1863                             substring_t *args, unsigned long *journal_devnum,
1864                             unsigned int *journal_ioprio, int is_remount)
1865 {
1866         struct ext4_sb_info *sbi = EXT4_SB(sb);
1867         const struct mount_opts *m;
1868         kuid_t uid;
1869         kgid_t gid;
1870         int arg = 0;
1871
1872 #ifdef CONFIG_QUOTA
1873         if (token == Opt_usrjquota)
1874                 return set_qf_name(sb, USRQUOTA, &args[0]);
1875         else if (token == Opt_grpjquota)
1876                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1877         else if (token == Opt_offusrjquota)
1878                 return clear_qf_name(sb, USRQUOTA);
1879         else if (token == Opt_offgrpjquota)
1880                 return clear_qf_name(sb, GRPQUOTA);
1881 #endif
1882         switch (token) {
1883         case Opt_noacl:
1884         case Opt_nouser_xattr:
1885                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1886                 break;
1887         case Opt_sb:
1888                 return 1;       /* handled by get_sb_block() */
1889         case Opt_removed:
1890                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1891                 return 1;
1892         case Opt_abort:
1893                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1894                 return 1;
1895         case Opt_i_version:
1896                 sb->s_flags |= SB_I_VERSION;
1897                 return 1;
1898         case Opt_lazytime:
1899                 sb->s_flags |= SB_LAZYTIME;
1900                 return 1;
1901         case Opt_nolazytime:
1902                 sb->s_flags &= ~SB_LAZYTIME;
1903                 return 1;
1904         }
1905
1906         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1907                 if (token == m->token)
1908                         break;
1909
1910         if (m->token == Opt_err) {
1911                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1912                          "or missing value", opt);
1913                 return -1;
1914         }
1915
1916         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1917                 ext4_msg(sb, KERN_ERR,
1918                          "Mount option \"%s\" incompatible with ext2", opt);
1919                 return -1;
1920         }
1921         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1922                 ext4_msg(sb, KERN_ERR,
1923                          "Mount option \"%s\" incompatible with ext3", opt);
1924                 return -1;
1925         }
1926
1927         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1928                 return -1;
1929         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1930                 return -1;
1931         if (m->flags & MOPT_EXPLICIT) {
1932                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1933                         set_opt2(sb, EXPLICIT_DELALLOC);
1934                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1935                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1936                 } else
1937                         return -1;
1938         }
1939         if (m->flags & MOPT_CLEAR_ERR)
1940                 clear_opt(sb, ERRORS_MASK);
1941         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1942                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1943                          "options when quota turned on");
1944                 return -1;
1945         }
1946
1947         if (m->flags & MOPT_NOSUPPORT) {
1948                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1949         } else if (token == Opt_commit) {
1950                 if (arg == 0)
1951                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1952                 else if (arg > INT_MAX / HZ) {
1953                         ext4_msg(sb, KERN_ERR,
1954                                  "Invalid commit interval %d, "
1955                                  "must be smaller than %d",
1956                                  arg, INT_MAX / HZ);
1957                         return -1;
1958                 }
1959                 sbi->s_commit_interval = HZ * arg;
1960         } else if (token == Opt_debug_want_extra_isize) {
1961                 if ((arg & 1) ||
1962                     (arg < 4) ||
1963                     (arg > (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE))) {
1964                         ext4_msg(sb, KERN_ERR,
1965                                  "Invalid want_extra_isize %d", arg);
1966                         return -1;
1967                 }
1968                 sbi->s_want_extra_isize = arg;
1969         } else if (token == Opt_max_batch_time) {
1970                 sbi->s_max_batch_time = arg;
1971         } else if (token == Opt_min_batch_time) {
1972                 sbi->s_min_batch_time = arg;
1973         } else if (token == Opt_inode_readahead_blks) {
1974                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1975                         ext4_msg(sb, KERN_ERR,
1976                                  "EXT4-fs: inode_readahead_blks must be "
1977                                  "0 or a power of 2 smaller than 2^31");
1978                         return -1;
1979                 }
1980                 sbi->s_inode_readahead_blks = arg;
1981         } else if (token == Opt_init_itable) {
1982                 set_opt(sb, INIT_INODE_TABLE);
1983                 if (!args->from)
1984                         arg = EXT4_DEF_LI_WAIT_MULT;
1985                 sbi->s_li_wait_mult = arg;
1986         } else if (token == Opt_max_dir_size_kb) {
1987                 sbi->s_max_dir_size_kb = arg;
1988         } else if (token == Opt_stripe) {
1989                 sbi->s_stripe = arg;
1990         } else if (token == Opt_resuid) {
1991                 uid = make_kuid(current_user_ns(), arg);
1992                 if (!uid_valid(uid)) {
1993                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1994                         return -1;
1995                 }
1996                 sbi->s_resuid = uid;
1997         } else if (token == Opt_resgid) {
1998                 gid = make_kgid(current_user_ns(), arg);
1999                 if (!gid_valid(gid)) {
2000                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
2001                         return -1;
2002                 }
2003                 sbi->s_resgid = gid;
2004         } else if (token == Opt_journal_dev) {
2005                 if (is_remount) {
2006                         ext4_msg(sb, KERN_ERR,
2007                                  "Cannot specify journal on remount");
2008                         return -1;
2009                 }
2010                 *journal_devnum = arg;
2011         } else if (token == Opt_journal_path) {
2012                 char *journal_path;
2013                 struct inode *journal_inode;
2014                 struct path path;
2015                 int error;
2016
2017                 if (is_remount) {
2018                         ext4_msg(sb, KERN_ERR,
2019                                  "Cannot specify journal on remount");
2020                         return -1;
2021                 }
2022                 journal_path = match_strdup(&args[0]);
2023                 if (!journal_path) {
2024                         ext4_msg(sb, KERN_ERR, "error: could not dup "
2025                                 "journal device string");
2026                         return -1;
2027                 }
2028
2029                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
2030                 if (error) {
2031                         ext4_msg(sb, KERN_ERR, "error: could not find "
2032                                 "journal device path: error %d", error);
2033                         kfree(journal_path);
2034                         return -1;
2035                 }
2036
2037                 journal_inode = d_inode(path.dentry);
2038                 if (!S_ISBLK(journal_inode->i_mode)) {
2039                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
2040                                 "is not a block device", journal_path);
2041                         path_put(&path);
2042                         kfree(journal_path);
2043                         return -1;
2044                 }
2045
2046                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
2047                 path_put(&path);
2048                 kfree(journal_path);
2049         } else if (token == Opt_journal_ioprio) {
2050                 if (arg > 7) {
2051                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
2052                                  " (must be 0-7)");
2053                         return -1;
2054                 }
2055                 *journal_ioprio =
2056                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
2057         } else if (token == Opt_test_dummy_encryption) {
2058 #ifdef CONFIG_FS_ENCRYPTION
2059                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
2060                 ext4_msg(sb, KERN_WARNING,
2061                          "Test dummy encryption mode enabled");
2062 #else
2063                 ext4_msg(sb, KERN_WARNING,
2064                          "Test dummy encryption mount option ignored");
2065 #endif
2066         } else if (m->flags & MOPT_DATAJ) {
2067                 if (is_remount) {
2068                         if (!sbi->s_journal)
2069                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
2070                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
2071                                 ext4_msg(sb, KERN_ERR,
2072                                          "Cannot change data mode on remount");
2073                                 return -1;
2074                         }
2075                 } else {
2076                         clear_opt(sb, DATA_FLAGS);
2077                         sbi->s_mount_opt |= m->mount_opt;
2078                 }
2079 #ifdef CONFIG_QUOTA
2080         } else if (m->flags & MOPT_QFMT) {
2081                 if (sb_any_quota_loaded(sb) &&
2082                     sbi->s_jquota_fmt != m->mount_opt) {
2083                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
2084                                  "quota options when quota turned on");
2085                         return -1;
2086                 }
2087                 if (ext4_has_feature_quota(sb)) {
2088                         ext4_msg(sb, KERN_INFO,
2089                                  "Quota format mount options ignored "
2090                                  "when QUOTA feature is enabled");
2091                         return 1;
2092                 }
2093                 sbi->s_jquota_fmt = m->mount_opt;
2094 #endif
2095         } else if (token == Opt_dax || token == Opt_dax_always ||
2096                    token == Opt_dax_inode || token == Opt_dax_never) {
2097 #ifdef CONFIG_FS_DAX
2098                 switch (token) {
2099                 case Opt_dax:
2100                 case Opt_dax_always:
2101                         if (is_remount &&
2102                             (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2103                              (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) {
2104                         fail_dax_change_remount:
2105                                 ext4_msg(sb, KERN_ERR, "can't change "
2106                                          "dax mount option while remounting");
2107                                 return -1;
2108                         }
2109                         if (is_remount &&
2110                             (test_opt(sb, DATA_FLAGS) ==
2111                              EXT4_MOUNT_JOURNAL_DATA)) {
2112                                     ext4_msg(sb, KERN_ERR, "can't mount with "
2113                                              "both data=journal and dax");
2114                                     return -1;
2115                         }
2116                         ext4_msg(sb, KERN_WARNING,
2117                                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
2118                         sbi->s_mount_opt |= EXT4_MOUNT_DAX_ALWAYS;
2119                         sbi->s_mount_opt2 &= ~EXT4_MOUNT2_DAX_NEVER;
2120                         break;
2121                 case Opt_dax_never:
2122                         if (is_remount &&
2123                             (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2124                              (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS)))
2125                                 goto fail_dax_change_remount;
2126                         sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_NEVER;
2127                         sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS;
2128                         break;
2129                 case Opt_dax_inode:
2130                         if (is_remount &&
2131                             ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2132                              (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2133                              !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE)))
2134                                 goto fail_dax_change_remount;
2135                         sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS;
2136                         sbi->s_mount_opt2 &= ~EXT4_MOUNT2_DAX_NEVER;
2137                         /* Strictly for printing options */
2138                         sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_INODE;
2139                         break;
2140                 }
2141 #else
2142                 ext4_msg(sb, KERN_INFO, "dax option not supported");
2143                 sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_NEVER;
2144                 sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS;
2145                 return -1;
2146 #endif
2147         } else if (token == Opt_data_err_abort) {
2148                 sbi->s_mount_opt |= m->mount_opt;
2149         } else if (token == Opt_data_err_ignore) {
2150                 sbi->s_mount_opt &= ~m->mount_opt;
2151         } else {
2152                 if (!args->from)
2153                         arg = 1;
2154                 if (m->flags & MOPT_CLEAR)
2155                         arg = !arg;
2156                 else if (unlikely(!(m->flags & MOPT_SET))) {
2157                         ext4_msg(sb, KERN_WARNING,
2158                                  "buggy handling of option %s", opt);
2159                         WARN_ON(1);
2160                         return -1;
2161                 }
2162                 if (arg != 0)
2163                         sbi->s_mount_opt |= m->mount_opt;
2164                 else
2165                         sbi->s_mount_opt &= ~m->mount_opt;
2166         }
2167         return 1;
2168 }
2169
2170 static int parse_options(char *options, struct super_block *sb,
2171                          unsigned long *journal_devnum,
2172                          unsigned int *journal_ioprio,
2173                          int is_remount)
2174 {
2175         struct ext4_sb_info __maybe_unused *sbi = EXT4_SB(sb);
2176         char *p, __maybe_unused *usr_qf_name, __maybe_unused *grp_qf_name;
2177         substring_t args[MAX_OPT_ARGS];
2178         int token;
2179
2180         if (!options)
2181                 return 1;
2182
2183         while ((p = strsep(&options, ",")) != NULL) {
2184                 if (!*p)
2185                         continue;
2186                 /*
2187                  * Initialize args struct so we know whether arg was
2188                  * found; some options take optional arguments.
2189                  */
2190                 args[0].to = args[0].from = NULL;
2191                 token = match_token(p, tokens, args);
2192                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
2193                                      journal_ioprio, is_remount) < 0)
2194                         return 0;
2195         }
2196 #ifdef CONFIG_QUOTA
2197         /*
2198          * We do the test below only for project quotas. 'usrquota' and
2199          * 'grpquota' mount options are allowed even without quota feature
2200          * to support legacy quotas in quota files.
2201          */
2202         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
2203                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
2204                          "Cannot enable project quota enforcement.");
2205                 return 0;
2206         }
2207         usr_qf_name = get_qf_name(sb, sbi, USRQUOTA);
2208         grp_qf_name = get_qf_name(sb, sbi, GRPQUOTA);
2209         if (usr_qf_name || grp_qf_name) {
2210                 if (test_opt(sb, USRQUOTA) && usr_qf_name)
2211                         clear_opt(sb, USRQUOTA);
2212
2213                 if (test_opt(sb, GRPQUOTA) && grp_qf_name)
2214                         clear_opt(sb, GRPQUOTA);
2215
2216                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
2217                         ext4_msg(sb, KERN_ERR, "old and new quota "
2218                                         "format mixing");
2219                         return 0;
2220                 }
2221
2222                 if (!sbi->s_jquota_fmt) {
2223                         ext4_msg(sb, KERN_ERR, "journaled quota format "
2224                                         "not specified");
2225                         return 0;
2226                 }
2227         }
2228 #endif
2229         if (test_opt(sb, DIOREAD_NOLOCK)) {
2230                 int blocksize =
2231                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
2232                 if (blocksize < PAGE_SIZE)
2233                         ext4_msg(sb, KERN_WARNING, "Warning: mounting with an "
2234                                  "experimental mount option 'dioread_nolock' "
2235                                  "for blocksize < PAGE_SIZE");
2236         }
2237         return 1;
2238 }
2239
2240 static inline void ext4_show_quota_options(struct seq_file *seq,
2241                                            struct super_block *sb)
2242 {
2243 #if defined(CONFIG_QUOTA)
2244         struct ext4_sb_info *sbi = EXT4_SB(sb);
2245         char *usr_qf_name, *grp_qf_name;
2246
2247         if (sbi->s_jquota_fmt) {
2248                 char *fmtname = "";
2249
2250                 switch (sbi->s_jquota_fmt) {
2251                 case QFMT_VFS_OLD:
2252                         fmtname = "vfsold";
2253                         break;
2254                 case QFMT_VFS_V0:
2255                         fmtname = "vfsv0";
2256                         break;
2257                 case QFMT_VFS_V1:
2258                         fmtname = "vfsv1";
2259                         break;
2260                 }
2261                 seq_printf(seq, ",jqfmt=%s", fmtname);
2262         }
2263
2264         rcu_read_lock();
2265         usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]);
2266         grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]);
2267         if (usr_qf_name)
2268                 seq_show_option(seq, "usrjquota", usr_qf_name);
2269         if (grp_qf_name)
2270                 seq_show_option(seq, "grpjquota", grp_qf_name);
2271         rcu_read_unlock();
2272 #endif
2273 }
2274
2275 static const char *token2str(int token)
2276 {
2277         const struct match_token *t;
2278
2279         for (t = tokens; t->token != Opt_err; t++)
2280                 if (t->token == token && !strchr(t->pattern, '='))
2281                         break;
2282         return t->pattern;
2283 }
2284
2285 /*
2286  * Show an option if
2287  *  - it's set to a non-default value OR
2288  *  - if the per-sb default is different from the global default
2289  */
2290 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2291                               int nodefs)
2292 {
2293         struct ext4_sb_info *sbi = EXT4_SB(sb);
2294         struct ext4_super_block *es = sbi->s_es;
2295         int def_errors, def_mount_opt = sbi->s_def_mount_opt;
2296         const struct mount_opts *m;
2297         char sep = nodefs ? '\n' : ',';
2298
2299 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2300 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2301
2302         if (sbi->s_sb_block != 1)
2303                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2304
2305         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2306                 int want_set = m->flags & MOPT_SET;
2307                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2308                     (m->flags & MOPT_CLEAR_ERR) || m->flags & MOPT_SKIP)
2309                         continue;
2310                 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2311                         continue; /* skip if same as the default */
2312                 if ((want_set &&
2313                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2314                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2315                         continue; /* select Opt_noFoo vs Opt_Foo */
2316                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2317         }
2318
2319         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2320             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2321                 SEQ_OPTS_PRINT("resuid=%u",
2322                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2323         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2324             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2325                 SEQ_OPTS_PRINT("resgid=%u",
2326                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2327         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2328         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2329                 SEQ_OPTS_PUTS("errors=remount-ro");
2330         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2331                 SEQ_OPTS_PUTS("errors=continue");
2332         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2333                 SEQ_OPTS_PUTS("errors=panic");
2334         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2335                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2336         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2337                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2338         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2339                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2340         if (sb->s_flags & SB_I_VERSION)
2341                 SEQ_OPTS_PUTS("i_version");
2342         if (nodefs || sbi->s_stripe)
2343                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2344         if (nodefs || EXT4_MOUNT_DATA_FLAGS &
2345                         (sbi->s_mount_opt ^ def_mount_opt)) {
2346                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2347                         SEQ_OPTS_PUTS("data=journal");
2348                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2349                         SEQ_OPTS_PUTS("data=ordered");
2350                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2351                         SEQ_OPTS_PUTS("data=writeback");
2352         }
2353         if (nodefs ||
2354             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2355                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2356                                sbi->s_inode_readahead_blks);
2357
2358         if (test_opt(sb, INIT_INODE_TABLE) && (nodefs ||
2359                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2360                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2361         if (nodefs || sbi->s_max_dir_size_kb)
2362                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2363         if (test_opt(sb, DATA_ERR_ABORT))
2364                 SEQ_OPTS_PUTS("data_err=abort");
2365         if (DUMMY_ENCRYPTION_ENABLED(sbi))
2366                 SEQ_OPTS_PUTS("test_dummy_encryption");
2367
2368         if (test_opt(sb, DAX_ALWAYS)) {
2369                 if (IS_EXT2_SB(sb))
2370                         SEQ_OPTS_PUTS("dax");
2371                 else
2372                         SEQ_OPTS_PUTS("dax=always");
2373         } else if (test_opt2(sb, DAX_NEVER)) {
2374                 SEQ_OPTS_PUTS("dax=never");
2375         } else if (test_opt2(sb, DAX_INODE)) {
2376                 SEQ_OPTS_PUTS("dax=inode");
2377         }
2378
2379         ext4_show_quota_options(seq, sb);
2380         return 0;
2381 }
2382
2383 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2384 {
2385         return _ext4_show_options(seq, root->d_sb, 0);
2386 }
2387
2388 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2389 {
2390         struct super_block *sb = seq->private;
2391         int rc;
2392
2393         seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
2394         rc = _ext4_show_options(seq, sb, 1);
2395         seq_puts(seq, "\n");
2396         return rc;
2397 }
2398
2399 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2400                             int read_only)
2401 {
2402         struct ext4_sb_info *sbi = EXT4_SB(sb);
2403         int err = 0;
2404
2405         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2406                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2407                          "forcing read-only mode");
2408                 err = -EROFS;
2409                 goto done;
2410         }
2411         if (read_only)
2412                 goto done;
2413         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2414                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2415                          "running e2fsck is recommended");
2416         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2417                 ext4_msg(sb, KERN_WARNING,
2418                          "warning: mounting fs with errors, "
2419                          "running e2fsck is recommended");
2420         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2421                  le16_to_cpu(es->s_mnt_count) >=
2422                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2423                 ext4_msg(sb, KERN_WARNING,
2424                          "warning: maximal mount count reached, "
2425                          "running e2fsck is recommended");
2426         else if (le32_to_cpu(es->s_checkinterval) &&
2427                  (ext4_get_tstamp(es, s_lastcheck) +
2428                   le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds()))
2429                 ext4_msg(sb, KERN_WARNING,
2430                          "warning: checktime reached, "
2431                          "running e2fsck is recommended");
2432         if (!sbi->s_journal)
2433                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2434         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2435                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2436         le16_add_cpu(&es->s_mnt_count, 1);
2437         ext4_update_tstamp(es, s_mtime);
2438         if (sbi->s_journal)
2439                 ext4_set_feature_journal_needs_recovery(sb);
2440
2441         err = ext4_commit_super(sb, 1);
2442 done:
2443         if (test_opt(sb, DEBUG))
2444                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2445                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2446                         sb->s_blocksize,
2447                         sbi->s_groups_count,
2448                         EXT4_BLOCKS_PER_GROUP(sb),
2449                         EXT4_INODES_PER_GROUP(sb),
2450                         sbi->s_mount_opt, sbi->s_mount_opt2);
2451
2452         cleancache_init_fs(sb);
2453         return err;
2454 }
2455
2456 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2457 {
2458         struct ext4_sb_info *sbi = EXT4_SB(sb);
2459         struct flex_groups **old_groups, **new_groups;
2460         int size, i, j;
2461
2462         if (!sbi->s_log_groups_per_flex)
2463                 return 0;
2464
2465         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2466         if (size <= sbi->s_flex_groups_allocated)
2467                 return 0;
2468
2469         new_groups = kvzalloc(roundup_pow_of_two(size *
2470                               sizeof(*sbi->s_flex_groups)), GFP_KERNEL);
2471         if (!new_groups) {
2472                 ext4_msg(sb, KERN_ERR,
2473                          "not enough memory for %d flex group pointers", size);
2474                 return -ENOMEM;
2475         }
2476         for (i = sbi->s_flex_groups_allocated; i < size; i++) {
2477                 new_groups[i] = kvzalloc(roundup_pow_of_two(
2478                                          sizeof(struct flex_groups)),
2479                                          GFP_KERNEL);
2480                 if (!new_groups[i]) {
2481                         for (j = sbi->s_flex_groups_allocated; j < i; j++)
2482                                 kvfree(new_groups[j]);
2483                         kvfree(new_groups);
2484                         ext4_msg(sb, KERN_ERR,
2485                                  "not enough memory for %d flex groups", size);
2486                         return -ENOMEM;
2487                 }
2488         }
2489         rcu_read_lock();
2490         old_groups = rcu_dereference(sbi->s_flex_groups);
2491         if (old_groups)
2492                 memcpy(new_groups, old_groups,
2493                        (sbi->s_flex_groups_allocated *
2494                         sizeof(struct flex_groups *)));
2495         rcu_read_unlock();
2496         rcu_assign_pointer(sbi->s_flex_groups, new_groups);
2497         sbi->s_flex_groups_allocated = size;
2498         if (old_groups)
2499                 ext4_kvfree_array_rcu(old_groups);
2500         return 0;
2501 }
2502
2503 static int ext4_fill_flex_info(struct super_block *sb)
2504 {
2505         struct ext4_sb_info *sbi = EXT4_SB(sb);
2506         struct ext4_group_desc *gdp = NULL;
2507         struct flex_groups *fg;
2508         ext4_group_t flex_group;
2509         int i, err;
2510
2511         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2512         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2513                 sbi->s_log_groups_per_flex = 0;
2514                 return 1;
2515         }
2516
2517         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2518         if (err)
2519                 goto failed;
2520
2521         for (i = 0; i < sbi->s_groups_count; i++) {
2522                 gdp = ext4_get_group_desc(sb, i, NULL);
2523
2524                 flex_group = ext4_flex_group(sbi, i);
2525                 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
2526                 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes);
2527                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2528                              &fg->free_clusters);
2529                 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs);
2530         }
2531
2532         return 1;
2533 failed:
2534         return 0;
2535 }
2536
2537 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2538                                    struct ext4_group_desc *gdp)
2539 {
2540         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2541         __u16 crc = 0;
2542         __le32 le_group = cpu_to_le32(block_group);
2543         struct ext4_sb_info *sbi = EXT4_SB(sb);
2544
2545         if (ext4_has_metadata_csum(sbi->s_sb)) {
2546                 /* Use new metadata_csum algorithm */
2547                 __u32 csum32;
2548                 __u16 dummy_csum = 0;
2549
2550                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2551                                      sizeof(le_group));
2552                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2553                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2554                                      sizeof(dummy_csum));
2555                 offset += sizeof(dummy_csum);
2556                 if (offset < sbi->s_desc_size)
2557                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2558                                              sbi->s_desc_size - offset);
2559
2560                 crc = csum32 & 0xFFFF;
2561                 goto out;
2562         }
2563
2564         /* old crc16 code */
2565         if (!ext4_has_feature_gdt_csum(sb))
2566                 return 0;
2567
2568         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2569         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2570         crc = crc16(crc, (__u8 *)gdp, offset);
2571         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2572         /* for checksum of struct ext4_group_desc do the rest...*/
2573         if (ext4_has_feature_64bit(sb) &&
2574             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2575                 crc = crc16(crc, (__u8 *)gdp + offset,
2576                             le16_to_cpu(sbi->s_es->s_desc_size) -
2577                                 offset);
2578
2579 out:
2580         return cpu_to_le16(crc);
2581 }
2582
2583 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2584                                 struct ext4_group_desc *gdp)
2585 {
2586         if (ext4_has_group_desc_csum(sb) &&
2587             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2588                 return 0;
2589
2590         return 1;
2591 }
2592
2593 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2594                               struct ext4_group_desc *gdp)
2595 {
2596         if (!ext4_has_group_desc_csum(sb))
2597                 return;
2598         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2599 }
2600
2601 /* Called at mount-time, super-block is locked */
2602 static int ext4_check_descriptors(struct super_block *sb,
2603                                   ext4_fsblk_t sb_block,
2604                                   ext4_group_t *first_not_zeroed)
2605 {
2606         struct ext4_sb_info *sbi = EXT4_SB(sb);
2607         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2608         ext4_fsblk_t last_block;
2609         ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0);
2610         ext4_fsblk_t block_bitmap;
2611         ext4_fsblk_t inode_bitmap;
2612         ext4_fsblk_t inode_table;
2613         int flexbg_flag = 0;
2614         ext4_group_t i, grp = sbi->s_groups_count;
2615
2616         if (ext4_has_feature_flex_bg(sb))
2617                 flexbg_flag = 1;
2618
2619         ext4_debug("Checking group descriptors");
2620
2621         for (i = 0; i < sbi->s_groups_count; i++) {
2622                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2623
2624                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2625                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2626                 else
2627                         last_block = first_block +
2628                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2629
2630                 if ((grp == sbi->s_groups_count) &&
2631                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2632                         grp = i;
2633
2634                 block_bitmap = ext4_block_bitmap(sb, gdp);
2635                 if (block_bitmap == sb_block) {
2636                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2637                                  "Block bitmap for group %u overlaps "
2638                                  "superblock", i);
2639                         if (!sb_rdonly(sb))
2640                                 return 0;
2641                 }
2642                 if (block_bitmap >= sb_block + 1 &&
2643                     block_bitmap <= last_bg_block) {
2644                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2645                                  "Block bitmap for group %u overlaps "
2646                                  "block group descriptors", i);
2647                         if (!sb_rdonly(sb))
2648                                 return 0;
2649                 }
2650                 if (block_bitmap < first_block || block_bitmap > last_block) {
2651                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2652                                "Block bitmap for group %u not in group "
2653                                "(block %llu)!", i, block_bitmap);
2654                         return 0;
2655                 }
2656                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2657                 if (inode_bitmap == sb_block) {
2658                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2659                                  "Inode bitmap for group %u overlaps "
2660                                  "superblock", i);
2661                         if (!sb_rdonly(sb))
2662                                 return 0;
2663                 }
2664                 if (inode_bitmap >= sb_block + 1 &&
2665                     inode_bitmap <= last_bg_block) {
2666                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2667                                  "Inode bitmap for group %u overlaps "
2668                                  "block group descriptors", i);
2669                         if (!sb_rdonly(sb))
2670                                 return 0;
2671                 }
2672                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2673                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2674                                "Inode bitmap for group %u not in group "
2675                                "(block %llu)!", i, inode_bitmap);
2676                         return 0;
2677                 }
2678                 inode_table = ext4_inode_table(sb, gdp);
2679                 if (inode_table == sb_block) {
2680                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2681                                  "Inode table for group %u overlaps "
2682                                  "superblock", i);
2683                         if (!sb_rdonly(sb))
2684                                 return 0;
2685                 }
2686                 if (inode_table >= sb_block + 1 &&
2687                     inode_table <= last_bg_block) {
2688                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2689                                  "Inode table for group %u overlaps "
2690                                  "block group descriptors", i);
2691                         if (!sb_rdonly(sb))
2692                                 return 0;
2693                 }
2694                 if (inode_table < first_block ||
2695                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2696                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2697                                "Inode table for group %u not in group "
2698                                "(block %llu)!", i, inode_table);
2699                         return 0;
2700                 }
2701                 ext4_lock_group(sb, i);
2702                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2703                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2704                                  "Checksum for group %u failed (%u!=%u)",
2705                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2706                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2707                         if (!sb_rdonly(sb)) {
2708                                 ext4_unlock_group(sb, i);
2709                                 return 0;
2710                         }
2711                 }
2712                 ext4_unlock_group(sb, i);
2713                 if (!flexbg_flag)
2714                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2715         }
2716         if (NULL != first_not_zeroed)
2717                 *first_not_zeroed = grp;
2718         return 1;
2719 }
2720
2721 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2722  * the superblock) which were deleted from all directories, but held open by
2723  * a process at the time of a crash.  We walk the list and try to delete these
2724  * inodes at recovery time (only with a read-write filesystem).
2725  *
2726  * In order to keep the orphan inode chain consistent during traversal (in
2727  * case of crash during recovery), we link each inode into the superblock
2728  * orphan list_head and handle it the same way as an inode deletion during
2729  * normal operation (which journals the operations for us).
2730  *
2731  * We only do an iget() and an iput() on each inode, which is very safe if we
2732  * accidentally point at an in-use or already deleted inode.  The worst that
2733  * can happen in this case is that we get a "bit already cleared" message from
2734  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2735  * e2fsck was run on this filesystem, and it must have already done the orphan
2736  * inode cleanup for us, so we can safely abort without any further action.
2737  */
2738 static void ext4_orphan_cleanup(struct super_block *sb,
2739                                 struct ext4_super_block *es)
2740 {
2741         unsigned int s_flags = sb->s_flags;
2742         int ret, nr_orphans = 0, nr_truncates = 0;
2743 #ifdef CONFIG_QUOTA
2744         int quota_update = 0;
2745         int i;
2746 #endif
2747         if (!es->s_last_orphan) {
2748                 jbd_debug(4, "no orphan inodes to clean up\n");
2749                 return;
2750         }
2751
2752         if (bdev_read_only(sb->s_bdev)) {
2753                 ext4_msg(sb, KERN_ERR, "write access "
2754                         "unavailable, skipping orphan cleanup");
2755                 return;
2756         }
2757
2758         /* Check if feature set would not allow a r/w mount */
2759         if (!ext4_feature_set_ok(sb, 0)) {
2760                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2761                          "unknown ROCOMPAT features");
2762                 return;
2763         }
2764
2765         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2766                 /* don't clear list on RO mount w/ errors */
2767                 if (es->s_last_orphan && !(s_flags & SB_RDONLY)) {
2768                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2769                                   "clearing orphan list.\n");
2770                         es->s_last_orphan = 0;
2771                 }
2772                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2773                 return;
2774         }
2775
2776         if (s_flags & SB_RDONLY) {
2777                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2778                 sb->s_flags &= ~SB_RDONLY;
2779         }
2780 #ifdef CONFIG_QUOTA
2781         /* Needed for iput() to work correctly and not trash data */
2782         sb->s_flags |= SB_ACTIVE;
2783
2784         /*
2785          * Turn on quotas which were not enabled for read-only mounts if
2786          * filesystem has quota feature, so that they are updated correctly.
2787          */
2788         if (ext4_has_feature_quota(sb) && (s_flags & SB_RDONLY)) {
2789                 int ret = ext4_enable_quotas(sb);
2790
2791                 if (!ret)
2792                         quota_update = 1;
2793                 else
2794                         ext4_msg(sb, KERN_ERR,
2795                                 "Cannot turn on quotas: error %d", ret);
2796         }
2797
2798         /* Turn on journaled quotas used for old sytle */
2799         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2800                 if (EXT4_SB(sb)->s_qf_names[i]) {
2801                         int ret = ext4_quota_on_mount(sb, i);
2802
2803                         if (!ret)
2804                                 quota_update = 1;
2805                         else
2806                                 ext4_msg(sb, KERN_ERR,
2807                                         "Cannot turn on journaled "
2808                                         "quota: type %d: error %d", i, ret);
2809                 }
2810         }
2811 #endif
2812
2813         while (es->s_last_orphan) {
2814                 struct inode *inode;
2815
2816                 /*
2817                  * We may have encountered an error during cleanup; if
2818                  * so, skip the rest.
2819                  */
2820                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2821                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2822                         es->s_last_orphan = 0;
2823                         break;
2824                 }
2825
2826                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2827                 if (IS_ERR(inode)) {
2828                         es->s_last_orphan = 0;
2829                         break;
2830                 }
2831
2832                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2833                 dquot_initialize(inode);
2834                 if (inode->i_nlink) {
2835                         if (test_opt(sb, DEBUG))
2836                                 ext4_msg(sb, KERN_DEBUG,
2837                                         "%s: truncating inode %lu to %lld bytes",
2838                                         __func__, inode->i_ino, inode->i_size);
2839                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2840                                   inode->i_ino, inode->i_size);
2841                         inode_lock(inode);
2842                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2843                         ret = ext4_truncate(inode);
2844                         if (ret)
2845                                 ext4_std_error(inode->i_sb, ret);
2846                         inode_unlock(inode);
2847                         nr_truncates++;
2848                 } else {
2849                         if (test_opt(sb, DEBUG))
2850                                 ext4_msg(sb, KERN_DEBUG,
2851                                         "%s: deleting unreferenced inode %lu",
2852                                         __func__, inode->i_ino);
2853                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2854                                   inode->i_ino);
2855                         nr_orphans++;
2856                 }
2857                 iput(inode);  /* The delete magic happens here! */
2858         }
2859
2860 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2861
2862         if (nr_orphans)
2863                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2864                        PLURAL(nr_orphans));
2865         if (nr_truncates)
2866                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2867                        PLURAL(nr_truncates));
2868 #ifdef CONFIG_QUOTA
2869         /* Turn off quotas if they were enabled for orphan cleanup */
2870         if (quota_update) {
2871                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2872                         if (sb_dqopt(sb)->files[i])
2873                                 dquot_quota_off(sb, i);
2874                 }
2875         }
2876 #endif
2877         sb->s_flags = s_flags; /* Restore SB_RDONLY status */
2878 }
2879
2880 /*
2881  * Maximal extent format file size.
2882  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2883  * extent format containers, within a sector_t, and within i_blocks
2884  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2885  * so that won't be a limiting factor.
2886  *
2887  * However there is other limiting factor. We do store extents in the form
2888  * of starting block and length, hence the resulting length of the extent
2889  * covering maximum file size must fit into on-disk format containers as
2890  * well. Given that length is always by 1 unit bigger than max unit (because
2891  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2892  *
2893  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2894  */
2895 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2896 {
2897         loff_t res;
2898         loff_t upper_limit = MAX_LFS_FILESIZE;
2899
2900         BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64));
2901
2902         if (!has_huge_files) {
2903                 upper_limit = (1LL << 32) - 1;
2904
2905                 /* total blocks in file system block size */
2906                 upper_limit >>= (blkbits - 9);
2907                 upper_limit <<= blkbits;
2908         }
2909
2910         /*
2911          * 32-bit extent-start container, ee_block. We lower the maxbytes
2912          * by one fs block, so ee_len can cover the extent of maximum file
2913          * size
2914          */
2915         res = (1LL << 32) - 1;
2916         res <<= blkbits;
2917
2918         /* Sanity check against vm- & vfs- imposed limits */
2919         if (res > upper_limit)
2920                 res = upper_limit;
2921
2922         return res;
2923 }
2924
2925 /*
2926  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2927  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2928  * We need to be 1 filesystem block less than the 2^48 sector limit.
2929  */
2930 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2931 {
2932         loff_t res = EXT4_NDIR_BLOCKS;
2933         int meta_blocks;
2934         loff_t upper_limit;
2935         /* This is calculated to be the largest file size for a dense, block
2936          * mapped file such that the file's total number of 512-byte sectors,
2937          * including data and all indirect blocks, does not exceed (2^48 - 1).
2938          *
2939          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2940          * number of 512-byte sectors of the file.
2941          */
2942
2943         if (!has_huge_files) {
2944                 /*
2945                  * !has_huge_files or implies that the inode i_block field
2946                  * represents total file blocks in 2^32 512-byte sectors ==
2947                  * size of vfs inode i_blocks * 8
2948                  */
2949                 upper_limit = (1LL << 32) - 1;
2950
2951                 /* total blocks in file system block size */
2952                 upper_limit >>= (bits - 9);
2953
2954         } else {
2955                 /*
2956                  * We use 48 bit ext4_inode i_blocks
2957                  * With EXT4_HUGE_FILE_FL set the i_blocks
2958                  * represent total number of blocks in
2959                  * file system block size
2960                  */
2961                 upper_limit = (1LL << 48) - 1;
2962
2963         }
2964
2965         /* indirect blocks */
2966         meta_blocks = 1;
2967         /* double indirect blocks */
2968         meta_blocks += 1 + (1LL << (bits-2));
2969         /* tripple indirect blocks */
2970         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2971
2972         upper_limit -= meta_blocks;
2973         upper_limit <<= bits;
2974
2975         res += 1LL << (bits-2);
2976         res += 1LL << (2*(bits-2));
2977         res += 1LL << (3*(bits-2));
2978         res <<= bits;
2979         if (res > upper_limit)
2980                 res = upper_limit;
2981
2982         if (res > MAX_LFS_FILESIZE)
2983                 res = MAX_LFS_FILESIZE;
2984
2985         return res;
2986 }
2987
2988 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2989                                    ext4_fsblk_t logical_sb_block, int nr)
2990 {
2991         struct ext4_sb_info *sbi = EXT4_SB(sb);
2992         ext4_group_t bg, first_meta_bg;
2993         int has_super = 0;
2994
2995         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2996
2997         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2998                 return logical_sb_block + nr + 1;
2999         bg = sbi->s_desc_per_block * nr;
3000         if (ext4_bg_has_super(sb, bg))
3001                 has_super = 1;
3002
3003         /*
3004          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
3005          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
3006          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
3007          * compensate.
3008          */
3009         if (sb->s_blocksize == 1024 && nr == 0 &&
3010             le32_to_cpu(sbi->s_es->s_first_data_block) == 0)
3011                 has_super++;
3012
3013         return (has_super + ext4_group_first_block_no(sb, bg));
3014 }
3015
3016 /**
3017  * ext4_get_stripe_size: Get the stripe size.
3018  * @sbi: In memory super block info
3019  *
3020  * If we have specified it via mount option, then
3021  * use the mount option value. If the value specified at mount time is
3022  * greater than the blocks per group use the super block value.
3023  * If the super block value is greater than blocks per group return 0.
3024  * Allocator needs it be less than blocks per group.
3025  *
3026  */
3027 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
3028 {
3029         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
3030         unsigned long stripe_width =
3031                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
3032         int ret;
3033
3034         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
3035                 ret = sbi->s_stripe;
3036         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
3037                 ret = stripe_width;
3038         else if (stride && stride <= sbi->s_blocks_per_group)
3039                 ret = stride;
3040         else
3041                 ret = 0;
3042
3043         /*
3044          * If the stripe width is 1, this makes no sense and
3045          * we set it to 0 to turn off stripe handling code.
3046          */
3047         if (ret <= 1)
3048                 ret = 0;
3049
3050         return ret;
3051 }
3052
3053 /*
3054  * Check whether this filesystem can be mounted based on
3055  * the features present and the RDONLY/RDWR mount requested.
3056  * Returns 1 if this filesystem can be mounted as requested,
3057  * 0 if it cannot be.
3058  */
3059 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
3060 {
3061         if (ext4_has_unknown_ext4_incompat_features(sb)) {
3062                 ext4_msg(sb, KERN_ERR,
3063                         "Couldn't mount because of "
3064                         "unsupported optional features (%x)",
3065                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
3066                         ~EXT4_FEATURE_INCOMPAT_SUPP));
3067                 return 0;
3068         }
3069
3070 #ifndef CONFIG_UNICODE
3071         if (ext4_has_feature_casefold(sb)) {
3072                 ext4_msg(sb, KERN_ERR,
3073                          "Filesystem with casefold feature cannot be "
3074                          "mounted without CONFIG_UNICODE");
3075                 return 0;
3076         }
3077 #endif
3078
3079         if (readonly)
3080                 return 1;
3081
3082         if (ext4_has_feature_readonly(sb)) {
3083                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
3084                 sb->s_flags |= SB_RDONLY;
3085                 return 1;
3086         }
3087
3088         /* Check that feature set is OK for a read-write mount */
3089         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
3090                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
3091                          "unsupported optional features (%x)",
3092                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
3093                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
3094                 return 0;
3095         }
3096         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
3097                 ext4_msg(sb, KERN_ERR,
3098                          "Can't support bigalloc feature without "
3099                          "extents feature\n");
3100                 return 0;
3101         }
3102
3103 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2)
3104         if (!readonly && (ext4_has_feature_quota(sb) ||
3105                           ext4_has_feature_project(sb))) {
3106                 ext4_msg(sb, KERN_ERR,
3107                          "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2");
3108                 return 0;
3109         }
3110 #endif  /* CONFIG_QUOTA */
3111         return 1;
3112 }
3113
3114 /*
3115  * This function is called once a day if we have errors logged
3116  * on the file system
3117  */
3118 static void print_daily_error_info(struct timer_list *t)
3119 {
3120         struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report);
3121         struct super_block *sb = sbi->s_sb;
3122         struct ext4_super_block *es = sbi->s_es;
3123
3124         if (es->s_error_count)
3125                 /* fsck newer than v1.41.13 is needed to clean this condition. */
3126                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
3127                          le32_to_cpu(es->s_error_count));
3128         if (es->s_first_error_time) {
3129                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d",
3130                        sb->s_id,
3131                        ext4_get_tstamp(es, s_first_error_time),
3132                        (int) sizeof(es->s_first_error_func),
3133                        es->s_first_error_func,
3134                        le32_to_cpu(es->s_first_error_line));
3135                 if (es->s_first_error_ino)
3136                         printk(KERN_CONT ": inode %u",
3137                                le32_to_cpu(es->s_first_error_ino));
3138                 if (es->s_first_error_block)
3139                         printk(KERN_CONT ": block %llu", (unsigned long long)
3140                                le64_to_cpu(es->s_first_error_block));
3141                 printk(KERN_CONT "\n");
3142         }
3143         if (es->s_last_error_time) {
3144                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d",
3145                        sb->s_id,
3146                        ext4_get_tstamp(es, s_last_error_time),
3147                        (int) sizeof(es->s_last_error_func),
3148                        es->s_last_error_func,
3149                        le32_to_cpu(es->s_last_error_line));
3150                 if (es->s_last_error_ino)
3151                         printk(KERN_CONT ": inode %u",
3152                                le32_to_cpu(es->s_last_error_ino));
3153                 if (es->s_last_error_block)
3154                         printk(KERN_CONT ": block %llu", (unsigned long long)
3155                                le64_to_cpu(es->s_last_error_block));
3156                 printk(KERN_CONT "\n");
3157         }
3158         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
3159 }
3160
3161 /* Find next suitable group and run ext4_init_inode_table */
3162 static int ext4_run_li_request(struct ext4_li_request *elr)
3163 {
3164         struct ext4_group_desc *gdp = NULL;
3165         ext4_group_t group, ngroups;
3166         struct super_block *sb;
3167         unsigned long timeout = 0;
3168         int ret = 0;
3169
3170         sb = elr->lr_super;
3171         ngroups = EXT4_SB(sb)->s_groups_count;
3172
3173         for (group = elr->lr_next_group; group < ngroups; group++) {
3174                 gdp = ext4_get_group_desc(sb, group, NULL);
3175                 if (!gdp) {
3176                         ret = 1;
3177                         break;
3178                 }
3179
3180                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3181                         break;
3182         }
3183
3184         if (group >= ngroups)
3185                 ret = 1;
3186
3187         if (!ret) {
3188                 timeout = jiffies;
3189                 ret = ext4_init_inode_table(sb, group,
3190                                             elr->lr_timeout ? 0 : 1);
3191                 if (elr->lr_timeout == 0) {
3192                         timeout = (jiffies - timeout) *
3193                                   elr->lr_sbi->s_li_wait_mult;
3194                         elr->lr_timeout = timeout;
3195                 }
3196                 elr->lr_next_sched = jiffies + elr->lr_timeout;
3197                 elr->lr_next_group = group + 1;
3198         }
3199         return ret;
3200 }
3201
3202 /*
3203  * Remove lr_request from the list_request and free the
3204  * request structure. Should be called with li_list_mtx held
3205  */
3206 static void ext4_remove_li_request(struct ext4_li_request *elr)
3207 {
3208         struct ext4_sb_info *sbi;
3209
3210         if (!elr)
3211                 return;
3212
3213         sbi = elr->lr_sbi;
3214
3215         list_del(&elr->lr_request);
3216         sbi->s_li_request = NULL;
3217         kfree(elr);
3218 }
3219
3220 static void ext4_unregister_li_request(struct super_block *sb)
3221 {
3222         mutex_lock(&ext4_li_mtx);
3223         if (!ext4_li_info) {
3224                 mutex_unlock(&ext4_li_mtx);
3225                 return;
3226         }
3227
3228         mutex_lock(&ext4_li_info->li_list_mtx);
3229         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
3230         mutex_unlock(&ext4_li_info->li_list_mtx);
3231         mutex_unlock(&ext4_li_mtx);
3232 }
3233
3234 static struct task_struct *ext4_lazyinit_task;
3235
3236 /*
3237  * This is the function where ext4lazyinit thread lives. It walks
3238  * through the request list searching for next scheduled filesystem.
3239  * When such a fs is found, run the lazy initialization request
3240  * (ext4_rn_li_request) and keep track of the time spend in this
3241  * function. Based on that time we compute next schedule time of
3242  * the request. When walking through the list is complete, compute
3243  * next waking time and put itself into sleep.
3244  */
3245 static int ext4_lazyinit_thread(void *arg)
3246 {
3247         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
3248         struct list_head *pos, *n;
3249         struct ext4_li_request *elr;
3250         unsigned long next_wakeup, cur;
3251
3252         BUG_ON(NULL == eli);
3253
3254 cont_thread:
3255         while (true) {
3256                 next_wakeup = MAX_JIFFY_OFFSET;
3257
3258                 mutex_lock(&eli->li_list_mtx);
3259                 if (list_empty(&eli->li_request_list)) {
3260                         mutex_unlock(&eli->li_list_mtx);
3261                         goto exit_thread;
3262                 }
3263                 list_for_each_safe(pos, n, &eli->li_request_list) {
3264                         int err = 0;
3265                         int progress = 0;
3266                         elr = list_entry(pos, struct ext4_li_request,
3267                                          lr_request);
3268
3269                         if (time_before(jiffies, elr->lr_next_sched)) {
3270                                 if (time_before(elr->lr_next_sched, next_wakeup))
3271                                         next_wakeup = elr->lr_next_sched;
3272                                 continue;
3273                         }
3274                         if (down_read_trylock(&elr->lr_super->s_umount)) {
3275                                 if (sb_start_write_trylock(elr->lr_super)) {
3276                                         progress = 1;
3277                                         /*
3278                                          * We hold sb->s_umount, sb can not
3279                                          * be removed from the list, it is
3280                                          * now safe to drop li_list_mtx
3281                                          */
3282                                         mutex_unlock(&eli->li_list_mtx);
3283                                         err = ext4_run_li_request(elr);
3284                                         sb_end_write(elr->lr_super);
3285                                         mutex_lock(&eli->li_list_mtx);
3286                                         n = pos->next;
3287                                 }
3288                                 up_read((&elr->lr_super->s_umount));
3289                         }
3290                         /* error, remove the lazy_init job */
3291                         if (err) {
3292                                 ext4_remove_li_request(elr);
3293                                 continue;
3294                         }
3295                         if (!progress) {
3296                                 elr->lr_next_sched = jiffies +
3297                                         (prandom_u32()
3298                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3299                         }
3300                         if (time_before(elr->lr_next_sched, next_wakeup))
3301                                 next_wakeup = elr->lr_next_sched;
3302                 }
3303                 mutex_unlock(&eli->li_list_mtx);
3304
3305                 try_to_freeze();
3306
3307                 cur = jiffies;
3308                 if ((time_after_eq(cur, next_wakeup)) ||
3309                     (MAX_JIFFY_OFFSET == next_wakeup)) {
3310                         cond_resched();
3311                         continue;
3312                 }
3313
3314                 schedule_timeout_interruptible(next_wakeup - cur);
3315
3316                 if (kthread_should_stop()) {
3317                         ext4_clear_request_list();
3318                         goto exit_thread;
3319                 }
3320         }
3321
3322 exit_thread:
3323         /*
3324          * It looks like the request list is empty, but we need
3325          * to check it under the li_list_mtx lock, to prevent any
3326          * additions into it, and of course we should lock ext4_li_mtx
3327          * to atomically free the list and ext4_li_info, because at
3328          * this point another ext4 filesystem could be registering
3329          * new one.
3330          */
3331         mutex_lock(&ext4_li_mtx);
3332         mutex_lock(&eli->li_list_mtx);
3333         if (!list_empty(&eli->li_request_list)) {
3334                 mutex_unlock(&eli->li_list_mtx);
3335                 mutex_unlock(&ext4_li_mtx);
3336                 goto cont_thread;
3337         }
3338         mutex_unlock(&eli->li_list_mtx);
3339         kfree(ext4_li_info);
3340         ext4_li_info = NULL;
3341         mutex_unlock(&ext4_li_mtx);
3342
3343         return 0;
3344 }
3345
3346 static void ext4_clear_request_list(void)
3347 {
3348         struct list_head *pos, *n;
3349         struct ext4_li_request *elr;
3350
3351         mutex_lock(&ext4_li_info->li_list_mtx);
3352         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3353                 elr = list_entry(pos, struct ext4_li_request,
3354                                  lr_request);
3355                 ext4_remove_li_request(elr);
3356         }
3357         mutex_unlock(&ext4_li_info->li_list_mtx);
3358 }
3359
3360 static int ext4_run_lazyinit_thread(void)
3361 {
3362         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3363                                          ext4_li_info, "ext4lazyinit");
3364         if (IS_ERR(ext4_lazyinit_task)) {
3365                 int err = PTR_ERR(ext4_lazyinit_task);
3366                 ext4_clear_request_list();
3367                 kfree(ext4_li_info);
3368                 ext4_li_info = NULL;
3369                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3370                                  "initialization thread\n",
3371                                  err);
3372                 return err;
3373         }
3374         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3375         return 0;
3376 }
3377
3378 /*
3379  * Check whether it make sense to run itable init. thread or not.
3380  * If there is at least one uninitialized inode table, return
3381  * corresponding group number, else the loop goes through all
3382  * groups and return total number of groups.
3383  */
3384 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3385 {
3386         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3387         struct ext4_group_desc *gdp = NULL;
3388
3389         if (!ext4_has_group_desc_csum(sb))
3390                 return ngroups;
3391
3392         for (group = 0; group < ngroups; group++) {
3393                 gdp = ext4_get_group_desc(sb, group, NULL);
3394                 if (!gdp)
3395                         continue;
3396
3397                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3398                         break;
3399         }
3400
3401         return group;
3402 }
3403
3404 static int ext4_li_info_new(void)
3405 {
3406         struct ext4_lazy_init *eli = NULL;
3407
3408         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3409         if (!eli)
3410                 return -ENOMEM;
3411
3412         INIT_LIST_HEAD(&eli->li_request_list);
3413         mutex_init(&eli->li_list_mtx);
3414
3415         eli->li_state |= EXT4_LAZYINIT_QUIT;
3416
3417         ext4_li_info = eli;
3418
3419         return 0;
3420 }
3421
3422 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3423                                             ext4_group_t start)
3424 {
3425         struct ext4_sb_info *sbi = EXT4_SB(sb);
3426         struct ext4_li_request *elr;
3427
3428         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3429         if (!elr)
3430                 return NULL;
3431
3432         elr->lr_super = sb;
3433         elr->lr_sbi = sbi;
3434         elr->lr_next_group = start;
3435
3436         /*
3437          * Randomize first schedule time of the request to
3438          * spread the inode table initialization requests
3439          * better.
3440          */
3441         elr->lr_next_sched = jiffies + (prandom_u32() %
3442                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3443         return elr;
3444 }
3445
3446 int ext4_register_li_request(struct super_block *sb,
3447                              ext4_group_t first_not_zeroed)
3448 {
3449         struct ext4_sb_info *sbi = EXT4_SB(sb);
3450         struct ext4_li_request *elr = NULL;
3451         ext4_group_t ngroups = sbi->s_groups_count;
3452         int ret = 0;
3453
3454         mutex_lock(&ext4_li_mtx);
3455         if (sbi->s_li_request != NULL) {
3456                 /*
3457                  * Reset timeout so it can be computed again, because
3458                  * s_li_wait_mult might have changed.
3459                  */
3460                 sbi->s_li_request->lr_timeout = 0;
3461                 goto out;
3462         }
3463
3464         if (first_not_zeroed == ngroups || sb_rdonly(sb) ||
3465             !test_opt(sb, INIT_INODE_TABLE))
3466                 goto out;
3467
3468         elr = ext4_li_request_new(sb, first_not_zeroed);
3469         if (!elr) {
3470                 ret = -ENOMEM;
3471                 goto out;
3472         }
3473
3474         if (NULL == ext4_li_info) {
3475                 ret = ext4_li_info_new();
3476                 if (ret)
3477                         goto out;
3478         }
3479
3480         mutex_lock(&ext4_li_info->li_list_mtx);
3481         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3482         mutex_unlock(&ext4_li_info->li_list_mtx);
3483
3484         sbi->s_li_request = elr;
3485         /*
3486          * set elr to NULL here since it has been inserted to
3487          * the request_list and the removal and free of it is
3488          * handled by ext4_clear_request_list from now on.
3489          */
3490         elr = NULL;
3491
3492         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3493                 ret = ext4_run_lazyinit_thread();
3494                 if (ret)
3495                         goto out;
3496         }
3497 out:
3498         mutex_unlock(&ext4_li_mtx);
3499         if (ret)
3500                 kfree(elr);
3501         return ret;
3502 }
3503
3504 /*
3505  * We do not need to lock anything since this is called on
3506  * module unload.
3507  */
3508 static void ext4_destroy_lazyinit_thread(void)
3509 {
3510         /*
3511          * If thread exited earlier
3512          * there's nothing to be done.
3513          */
3514         if (!ext4_li_info || !ext4_lazyinit_task)
3515                 return;
3516
3517         kthread_stop(ext4_lazyinit_task);
3518 }
3519
3520 static int set_journal_csum_feature_set(struct super_block *sb)
3521 {
3522         int ret = 1;
3523         int compat, incompat;
3524         struct ext4_sb_info *sbi = EXT4_SB(sb);
3525
3526         if (ext4_has_metadata_csum(sb)) {
3527                 /* journal checksum v3 */
3528                 compat = 0;
3529                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3530         } else {
3531                 /* journal checksum v1 */
3532                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3533                 incompat = 0;
3534         }
3535
3536         jbd2_journal_clear_features(sbi->s_journal,
3537                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3538                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3539                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3540         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3541                 ret = jbd2_journal_set_features(sbi->s_journal,
3542                                 compat, 0,
3543                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3544                                 incompat);
3545         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3546                 ret = jbd2_journal_set_features(sbi->s_journal,
3547                                 compat, 0,
3548                                 incompat);
3549                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3550                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3551         } else {
3552                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3553                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3554         }
3555
3556         return ret;
3557 }
3558
3559 /*
3560  * Note: calculating the overhead so we can be compatible with
3561  * historical BSD practice is quite difficult in the face of
3562  * clusters/bigalloc.  This is because multiple metadata blocks from
3563  * different block group can end up in the same allocation cluster.
3564  * Calculating the exact overhead in the face of clustered allocation
3565  * requires either O(all block bitmaps) in memory or O(number of block
3566  * groups**2) in time.  We will still calculate the superblock for
3567  * older file systems --- and if we come across with a bigalloc file
3568  * system with zero in s_overhead_clusters the estimate will be close to
3569  * correct especially for very large cluster sizes --- but for newer
3570  * file systems, it's better to calculate this figure once at mkfs
3571  * time, and store it in the superblock.  If the superblock value is
3572  * present (even for non-bigalloc file systems), we will use it.
3573  */
3574 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3575                           char *buf)
3576 {
3577         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3578         struct ext4_group_desc  *gdp;
3579         ext4_fsblk_t            first_block, last_block, b;
3580         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3581         int                     s, j, count = 0;
3582
3583         if (!ext4_has_feature_bigalloc(sb))
3584                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3585                         sbi->s_itb_per_group + 2);
3586
3587         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3588                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3589         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3590         for (i = 0; i < ngroups; i++) {
3591                 gdp = ext4_get_group_desc(sb, i, NULL);
3592                 b = ext4_block_bitmap(sb, gdp);
3593                 if (b >= first_block && b <= last_block) {
3594                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3595                         count++;
3596                 }
3597                 b = ext4_inode_bitmap(sb, gdp);
3598                 if (b >= first_block && b <= last_block) {
3599                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3600                         count++;
3601                 }
3602                 b = ext4_inode_table(sb, gdp);
3603                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3604                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3605                                 int c = EXT4_B2C(sbi, b - first_block);
3606                                 ext4_set_bit(c, buf);
3607                                 count++;
3608                         }
3609                 if (i != grp)
3610                         continue;
3611                 s = 0;
3612                 if (ext4_bg_has_super(sb, grp)) {
3613                         ext4_set_bit(s++, buf);
3614                         count++;
3615                 }
3616                 j = ext4_bg_num_gdb(sb, grp);
3617                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3618                         ext4_error(sb, "Invalid number of block group "
3619                                    "descriptor blocks: %d", j);
3620                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3621                 }
3622                 count += j;
3623                 for (; j > 0; j--)
3624                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3625         }
3626         if (!count)
3627                 return 0;
3628         return EXT4_CLUSTERS_PER_GROUP(sb) -
3629                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3630 }
3631
3632 /*
3633  * Compute the overhead and stash it in sbi->s_overhead
3634  */
3635 int ext4_calculate_overhead(struct super_block *sb)
3636 {
3637         struct ext4_sb_info *sbi = EXT4_SB(sb);
3638         struct ext4_super_block *es = sbi->s_es;
3639         struct inode *j_inode;
3640         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3641         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3642         ext4_fsblk_t overhead = 0;
3643         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3644
3645         if (!buf)
3646                 return -ENOMEM;
3647
3648         /*
3649          * Compute the overhead (FS structures).  This is constant
3650          * for a given filesystem unless the number of block groups
3651          * changes so we cache the previous value until it does.
3652          */
3653
3654         /*
3655          * All of the blocks before first_data_block are overhead
3656          */
3657         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3658
3659         /*
3660          * Add the overhead found in each block group
3661          */
3662         for (i = 0; i < ngroups; i++) {
3663                 int blks;
3664
3665                 blks = count_overhead(sb, i, buf);
3666                 overhead += blks;
3667                 if (blks)
3668                         memset(buf, 0, PAGE_SIZE);
3669                 cond_resched();
3670         }
3671
3672         /*
3673          * Add the internal journal blocks whether the journal has been
3674          * loaded or not
3675          */
3676         if (sbi->s_journal && !sbi->journal_bdev)
3677                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3678         else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) {
3679                 /* j_inum for internal journal is non-zero */
3680                 j_inode = ext4_get_journal_inode(sb, j_inum);
3681                 if (j_inode) {
3682                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3683                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
3684                         iput(j_inode);
3685                 } else {
3686                         ext4_msg(sb, KERN_ERR, "can't get journal size");
3687                 }
3688         }
3689         sbi->s_overhead = overhead;
3690         smp_wmb();
3691         free_page((unsigned long) buf);
3692         return 0;
3693 }
3694
3695 static void ext4_set_resv_clusters(struct super_block *sb)
3696 {
3697         ext4_fsblk_t resv_clusters;
3698         struct ext4_sb_info *sbi = EXT4_SB(sb);
3699
3700         /*
3701          * There's no need to reserve anything when we aren't using extents.
3702          * The space estimates are exact, there are no unwritten extents,
3703          * hole punching doesn't need new metadata... This is needed especially
3704          * to keep ext2/3 backward compatibility.
3705          */
3706         if (!ext4_has_feature_extents(sb))
3707                 return;
3708         /*
3709          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3710          * This should cover the situations where we can not afford to run
3711          * out of space like for example punch hole, or converting
3712          * unwritten extents in delalloc path. In most cases such
3713          * allocation would require 1, or 2 blocks, higher numbers are
3714          * very rare.
3715          */
3716         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3717                          sbi->s_cluster_bits);
3718
3719         do_div(resv_clusters, 50);
3720         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3721
3722         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3723 }
3724
3725 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3726 {
3727         struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev);
3728         char *orig_data = kstrdup(data, GFP_KERNEL);
3729         struct buffer_head *bh, **group_desc;
3730         struct ext4_super_block *es = NULL;
3731         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3732         struct flex_groups **flex_groups;
3733         ext4_fsblk_t block;
3734         ext4_fsblk_t sb_block = get_sb_block(&data);
3735         ext4_fsblk_t logical_sb_block;
3736         unsigned long offset = 0;
3737         unsigned long journal_devnum = 0;
3738         unsigned long def_mount_opts;
3739         struct inode *root;
3740         const char *descr;
3741         int ret = -ENOMEM;
3742         int blocksize, clustersize;
3743         unsigned int db_count;
3744         unsigned int i;
3745         int needs_recovery, has_huge_files;
3746         __u64 blocks_count;
3747         int err = 0;
3748         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3749         ext4_group_t first_not_zeroed;
3750
3751         if ((data && !orig_data) || !sbi)
3752                 goto out_free_base;
3753
3754         sbi->s_daxdev = dax_dev;
3755         sbi->s_blockgroup_lock =
3756                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3757         if (!sbi->s_blockgroup_lock)
3758                 goto out_free_base;
3759
3760         sb->s_fs_info = sbi;
3761         sbi->s_sb = sb;
3762         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3763         sbi->s_sb_block = sb_block;
3764         if (sb->s_bdev->bd_part)
3765                 sbi->s_sectors_written_start =
3766                         part_stat_read(sb->s_bdev->bd_part, sectors[STAT_WRITE]);
3767
3768         /* Cleanup superblock name */
3769         strreplace(sb->s_id, '/', '!');
3770
3771         /* -EINVAL is default */
3772         ret = -EINVAL;
3773         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3774         if (!blocksize) {
3775                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3776                 goto out_fail;
3777         }
3778
3779         /*
3780          * The ext4 superblock will not be buffer aligned for other than 1kB
3781          * block sizes.  We need to calculate the offset from buffer start.
3782          */
3783         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3784                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3785                 offset = do_div(logical_sb_block, blocksize);
3786         } else {
3787                 logical_sb_block = sb_block;
3788         }
3789
3790         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3791                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3792                 goto out_fail;
3793         }
3794         /*
3795          * Note: s_es must be initialized as soon as possible because
3796          *       some ext4 macro-instructions depend on its value
3797          */
3798         es = (struct ext4_super_block *) (bh->b_data + offset);
3799         sbi->s_es = es;
3800         sb->s_magic = le16_to_cpu(es->s_magic);
3801         if (sb->s_magic != EXT4_SUPER_MAGIC)
3802                 goto cantfind_ext4;
3803         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3804
3805         /* Warn if metadata_csum and gdt_csum are both set. */
3806         if (ext4_has_feature_metadata_csum(sb) &&
3807             ext4_has_feature_gdt_csum(sb))
3808                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3809                              "redundant flags; please run fsck.");
3810
3811         /* Check for a known checksum algorithm */
3812         if (!ext4_verify_csum_type(sb, es)) {
3813                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3814                          "unknown checksum algorithm.");
3815                 silent = 1;
3816                 goto cantfind_ext4;
3817         }
3818
3819         /* Load the checksum driver */
3820         sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3821         if (IS_ERR(sbi->s_chksum_driver)) {
3822                 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3823                 ret = PTR_ERR(sbi->s_chksum_driver);
3824                 sbi->s_chksum_driver = NULL;
3825                 goto failed_mount;
3826         }
3827
3828         /* Check superblock checksum */
3829         if (!ext4_superblock_csum_verify(sb, es)) {
3830                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3831                          "invalid superblock checksum.  Run e2fsck?");
3832                 silent = 1;
3833                 ret = -EFSBADCRC;
3834                 goto cantfind_ext4;
3835         }
3836
3837         /* Precompute checksum seed for all metadata */
3838         if (ext4_has_feature_csum_seed(sb))
3839                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3840         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
3841                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3842                                                sizeof(es->s_uuid));
3843
3844         /* Set defaults before we parse the mount options */
3845         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3846         set_opt(sb, INIT_INODE_TABLE);
3847         if (def_mount_opts & EXT4_DEFM_DEBUG)
3848                 set_opt(sb, DEBUG);
3849         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3850                 set_opt(sb, GRPID);
3851         if (def_mount_opts & EXT4_DEFM_UID16)
3852                 set_opt(sb, NO_UID32);
3853         /* xattr user namespace & acls are now defaulted on */
3854         set_opt(sb, XATTR_USER);
3855 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3856         set_opt(sb, POSIX_ACL);
3857 #endif
3858         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3859         if (ext4_has_metadata_csum(sb))
3860                 set_opt(sb, JOURNAL_CHECKSUM);
3861
3862         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3863                 set_opt(sb, JOURNAL_DATA);
3864         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3865                 set_opt(sb, ORDERED_DATA);
3866         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3867                 set_opt(sb, WRITEBACK_DATA);
3868
3869         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3870                 set_opt(sb, ERRORS_PANIC);
3871         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3872                 set_opt(sb, ERRORS_CONT);
3873         else
3874                 set_opt(sb, ERRORS_RO);
3875         /* block_validity enabled by default; disable with noblock_validity */
3876         set_opt(sb, BLOCK_VALIDITY);
3877         if (def_mount_opts & EXT4_DEFM_DISCARD)
3878                 set_opt(sb, DISCARD);
3879
3880         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3881         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3882         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3883         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3884         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3885
3886         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3887                 set_opt(sb, BARRIER);
3888
3889         /*
3890          * enable delayed allocation by default
3891          * Use -o nodelalloc to turn it off
3892          */
3893         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3894             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3895                 set_opt(sb, DELALLOC);
3896
3897         /*
3898          * set default s_li_wait_mult for lazyinit, for the case there is
3899          * no mount option specified.
3900          */
3901         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3902
3903         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3904
3905         if (blocksize == PAGE_SIZE)
3906                 set_opt(sb, DIOREAD_NOLOCK);
3907
3908         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3909             blocksize > EXT4_MAX_BLOCK_SIZE) {
3910                 ext4_msg(sb, KERN_ERR,
3911                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3912                          blocksize, le32_to_cpu(es->s_log_block_size));
3913                 goto failed_mount;
3914         }
3915
3916         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3917                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3918                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3919         } else {
3920                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3921                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3922                 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) {
3923                         ext4_msg(sb, KERN_ERR, "invalid first ino: %u",
3924                                  sbi->s_first_ino);
3925                         goto failed_mount;
3926                 }
3927                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3928                     (!is_power_of_2(sbi->s_inode_size)) ||
3929                     (sbi->s_inode_size > blocksize)) {
3930                         ext4_msg(sb, KERN_ERR,
3931                                "unsupported inode size: %d",
3932                                sbi->s_inode_size);
3933                         ext4_msg(sb, KERN_ERR, "blocksize: %d", blocksize);
3934                         goto failed_mount;
3935                 }
3936                 /*
3937                  * i_atime_extra is the last extra field available for
3938                  * [acm]times in struct ext4_inode. Checking for that
3939                  * field should suffice to ensure we have extra space
3940                  * for all three.
3941                  */
3942                 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) +
3943                         sizeof(((struct ext4_inode *)0)->i_atime_extra)) {
3944                         sb->s_time_gran = 1;
3945                         sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX;
3946                 } else {
3947                         sb->s_time_gran = NSEC_PER_SEC;
3948                         sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX;
3949                 }
3950                 sb->s_time_min = EXT4_TIMESTAMP_MIN;
3951         }
3952         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3953                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3954                         EXT4_GOOD_OLD_INODE_SIZE;
3955                 if (ext4_has_feature_extra_isize(sb)) {
3956                         unsigned v, max = (sbi->s_inode_size -
3957                                            EXT4_GOOD_OLD_INODE_SIZE);
3958
3959                         v = le16_to_cpu(es->s_want_extra_isize);
3960                         if (v > max) {
3961                                 ext4_msg(sb, KERN_ERR,
3962                                          "bad s_want_extra_isize: %d", v);
3963                                 goto failed_mount;
3964                         }
3965                         if (sbi->s_want_extra_isize < v)
3966                                 sbi->s_want_extra_isize = v;
3967
3968                         v = le16_to_cpu(es->s_min_extra_isize);
3969                         if (v > max) {
3970                                 ext4_msg(sb, KERN_ERR,
3971                                          "bad s_min_extra_isize: %d", v);
3972                                 goto failed_mount;
3973                         }
3974                         if (sbi->s_want_extra_isize < v)
3975                                 sbi->s_want_extra_isize = v;
3976                 }
3977         }
3978
3979         if (sbi->s_es->s_mount_opts[0]) {
3980                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3981                                               sizeof(sbi->s_es->s_mount_opts),
3982                                               GFP_KERNEL);
3983                 if (!s_mount_opts)
3984                         goto failed_mount;
3985                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3986                                    &journal_ioprio, 0)) {
3987                         ext4_msg(sb, KERN_WARNING,
3988                                  "failed to parse options in superblock: %s",
3989                                  s_mount_opts);
3990                 }
3991                 kfree(s_mount_opts);
3992         }
3993         sbi->s_def_mount_opt = sbi->s_mount_opt;
3994         if (!parse_options((char *) data, sb, &journal_devnum,
3995                            &journal_ioprio, 0))
3996                 goto failed_mount;
3997
3998 #ifdef CONFIG_UNICODE
3999         if (ext4_has_feature_casefold(sb) && !sbi->s_encoding) {
4000                 const struct ext4_sb_encodings *encoding_info;
4001                 struct unicode_map *encoding;
4002                 __u16 encoding_flags;
4003
4004                 if (ext4_has_feature_encrypt(sb)) {
4005                         ext4_msg(sb, KERN_ERR,
4006                                  "Can't mount with encoding and encryption");
4007                         goto failed_mount;
4008                 }
4009
4010                 if (ext4_sb_read_encoding(es, &encoding_info,
4011                                           &encoding_flags)) {
4012                         ext4_msg(sb, KERN_ERR,
4013                                  "Encoding requested by superblock is unknown");
4014                         goto failed_mount;
4015                 }
4016
4017                 encoding = utf8_load(encoding_info->version);
4018                 if (IS_ERR(encoding)) {
4019                         ext4_msg(sb, KERN_ERR,
4020                                  "can't mount with superblock charset: %s-%s "
4021                                  "not supported by the kernel. flags: 0x%x.",
4022                                  encoding_info->name, encoding_info->version,
4023                                  encoding_flags);
4024                         goto failed_mount;
4025                 }
4026                 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: "
4027                          "%s-%s with flags 0x%hx", encoding_info->name,
4028                          encoding_info->version?:"\b", encoding_flags);
4029
4030                 sbi->s_encoding = encoding;
4031                 sbi->s_encoding_flags = encoding_flags;
4032         }
4033 #endif
4034
4035         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4036                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with data=journal disables delayed allocation, dioread_nolock, and O_DIRECT support!\n");
4037                 /* can't mount with both data=journal and dioread_nolock. */
4038                 clear_opt(sb, DIOREAD_NOLOCK);
4039                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4040                         ext4_msg(sb, KERN_ERR, "can't mount with "
4041                                  "both data=journal and delalloc");
4042                         goto failed_mount;
4043                 }
4044                 if (test_opt(sb, DAX_ALWAYS)) {
4045                         ext4_msg(sb, KERN_ERR, "can't mount with "
4046                                  "both data=journal and dax");
4047                         goto failed_mount;
4048                 }
4049                 if (ext4_has_feature_encrypt(sb)) {
4050                         ext4_msg(sb, KERN_WARNING,
4051                                  "encrypted files will use data=ordered "
4052                                  "instead of data journaling mode");
4053                 }
4054                 if (test_opt(sb, DELALLOC))
4055                         clear_opt(sb, DELALLOC);
4056         } else {
4057                 sb->s_iflags |= SB_I_CGROUPWB;
4058         }
4059
4060         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
4061                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
4062
4063         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
4064             (ext4_has_compat_features(sb) ||
4065              ext4_has_ro_compat_features(sb) ||
4066              ext4_has_incompat_features(sb)))
4067                 ext4_msg(sb, KERN_WARNING,
4068                        "feature flags set on rev 0 fs, "
4069                        "running e2fsck is recommended");
4070
4071         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
4072                 set_opt2(sb, HURD_COMPAT);
4073                 if (ext4_has_feature_64bit(sb)) {
4074                         ext4_msg(sb, KERN_ERR,
4075                                  "The Hurd can't support 64-bit file systems");
4076                         goto failed_mount;
4077                 }
4078
4079                 /*
4080                  * ea_inode feature uses l_i_version field which is not
4081                  * available in HURD_COMPAT mode.
4082                  */
4083                 if (ext4_has_feature_ea_inode(sb)) {
4084                         ext4_msg(sb, KERN_ERR,
4085                                  "ea_inode feature is not supported for Hurd");
4086                         goto failed_mount;
4087                 }
4088         }
4089
4090         if (IS_EXT2_SB(sb)) {
4091                 if (ext2_feature_set_ok(sb))
4092                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
4093                                  "using the ext4 subsystem");
4094                 else {
4095                         /*
4096                          * If we're probing be silent, if this looks like
4097                          * it's actually an ext[34] filesystem.
4098                          */
4099                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4100                                 goto failed_mount;
4101                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
4102                                  "to feature incompatibilities");
4103                         goto failed_mount;
4104                 }
4105         }
4106
4107         if (IS_EXT3_SB(sb)) {
4108                 if (ext3_feature_set_ok(sb))
4109                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
4110                                  "using the ext4 subsystem");
4111                 else {
4112                         /*
4113                          * If we're probing be silent, if this looks like
4114                          * it's actually an ext4 filesystem.
4115                          */
4116                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4117                                 goto failed_mount;
4118                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
4119                                  "to feature incompatibilities");
4120                         goto failed_mount;
4121                 }
4122         }
4123
4124         /*
4125          * Check feature flags regardless of the revision level, since we
4126          * previously didn't change the revision level when setting the flags,
4127          * so there is a chance incompat flags are set on a rev 0 filesystem.
4128          */
4129         if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
4130                 goto failed_mount;
4131
4132         if (le32_to_cpu(es->s_log_block_size) >
4133             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4134                 ext4_msg(sb, KERN_ERR,
4135                          "Invalid log block size: %u",
4136                          le32_to_cpu(es->s_log_block_size));
4137                 goto failed_mount;
4138         }
4139         if (le32_to_cpu(es->s_log_cluster_size) >
4140             (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4141                 ext4_msg(sb, KERN_ERR,
4142                          "Invalid log cluster size: %u",
4143                          le32_to_cpu(es->s_log_cluster_size));
4144                 goto failed_mount;
4145         }
4146
4147         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
4148                 ext4_msg(sb, KERN_ERR,
4149                          "Number of reserved GDT blocks insanely large: %d",
4150                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
4151                 goto failed_mount;
4152         }
4153
4154         if (bdev_dax_supported(sb->s_bdev, blocksize))
4155                 set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags);
4156
4157         if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) {
4158                 if (ext4_has_feature_inline_data(sb)) {
4159                         ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
4160                                         " that may contain inline data");
4161                         goto failed_mount;
4162                 }
4163                 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) {
4164                         ext4_msg(sb, KERN_ERR,
4165                                 "DAX unsupported by block device.");
4166                         goto failed_mount;
4167                 }
4168         }
4169
4170         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
4171                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
4172                          es->s_encryption_level);
4173                 goto failed_mount;
4174         }
4175
4176         if (sb->s_blocksize != blocksize) {
4177                 /* Validate the filesystem blocksize */
4178                 if (!sb_set_blocksize(sb, blocksize)) {
4179                         ext4_msg(sb, KERN_ERR, "bad block size %d",
4180                                         blocksize);
4181                         goto failed_mount;
4182                 }
4183
4184                 brelse(bh);
4185                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
4186                 offset = do_div(logical_sb_block, blocksize);
4187                 bh = sb_bread_unmovable(sb, logical_sb_block);
4188                 if (!bh) {
4189                         ext4_msg(sb, KERN_ERR,
4190                                "Can't read superblock on 2nd try");
4191                         goto failed_mount;
4192                 }
4193                 es = (struct ext4_super_block *)(bh->b_data + offset);
4194                 sbi->s_es = es;
4195                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
4196                         ext4_msg(sb, KERN_ERR,
4197                                "Magic mismatch, very weird!");
4198                         goto failed_mount;
4199                 }
4200         }
4201
4202         has_huge_files = ext4_has_feature_huge_file(sb);
4203         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
4204                                                       has_huge_files);
4205         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
4206
4207         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
4208         if (ext4_has_feature_64bit(sb)) {
4209                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
4210                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
4211                     !is_power_of_2(sbi->s_desc_size)) {
4212                         ext4_msg(sb, KERN_ERR,
4213                                "unsupported descriptor size %lu",
4214                                sbi->s_desc_size);
4215                         goto failed_mount;
4216                 }
4217         } else
4218                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
4219
4220         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
4221         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
4222
4223         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
4224         if (sbi->s_inodes_per_block == 0)
4225                 goto cantfind_ext4;
4226         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
4227             sbi->s_inodes_per_group > blocksize * 8) {
4228                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
4229                          sbi->s_inodes_per_group);
4230                 goto failed_mount;
4231         }
4232         sbi->s_itb_per_group = sbi->s_inodes_per_group /
4233                                         sbi->s_inodes_per_block;
4234         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
4235         sbi->s_sbh = bh;
4236         sbi->s_mount_state = le16_to_cpu(es->s_state);
4237         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
4238         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
4239
4240         for (i = 0; i < 4; i++)
4241                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
4242         sbi->s_def_hash_version = es->s_def_hash_version;
4243         if (ext4_has_feature_dir_index(sb)) {
4244                 i = le32_to_cpu(es->s_flags);
4245                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
4246                         sbi->s_hash_unsigned = 3;
4247                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
4248 #ifdef __CHAR_UNSIGNED__
4249                         if (!sb_rdonly(sb))
4250                                 es->s_flags |=
4251                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
4252                         sbi->s_hash_unsigned = 3;
4253 #else
4254                         if (!sb_rdonly(sb))
4255                                 es->s_flags |=
4256                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
4257 #endif
4258                 }
4259         }
4260
4261         /* Handle clustersize */
4262         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
4263         if (ext4_has_feature_bigalloc(sb)) {
4264                 if (clustersize < blocksize) {
4265                         ext4_msg(sb, KERN_ERR,
4266                                  "cluster size (%d) smaller than "
4267                                  "block size (%d)", clustersize, blocksize);
4268                         goto failed_mount;
4269                 }
4270                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
4271                         le32_to_cpu(es->s_log_block_size);
4272                 sbi->s_clusters_per_group =
4273                         le32_to_cpu(es->s_clusters_per_group);
4274                 if (sbi->s_clusters_per_group > blocksize * 8) {
4275                         ext4_msg(sb, KERN_ERR,
4276                                  "#clusters per group too big: %lu",
4277                                  sbi->s_clusters_per_group);
4278                         goto failed_mount;
4279                 }
4280                 if (sbi->s_blocks_per_group !=
4281                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
4282                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
4283                                  "clusters per group (%lu) inconsistent",
4284                                  sbi->s_blocks_per_group,
4285                                  sbi->s_clusters_per_group);
4286                         goto failed_mount;
4287                 }
4288         } else {
4289                 if (clustersize != blocksize) {
4290                         ext4_msg(sb, KERN_ERR,
4291                                  "fragment/cluster size (%d) != "
4292                                  "block size (%d)", clustersize, blocksize);
4293                         goto failed_mount;
4294                 }
4295                 if (sbi->s_blocks_per_group > blocksize * 8) {
4296                         ext4_msg(sb, KERN_ERR,
4297                                  "#blocks per group too big: %lu",
4298                                  sbi->s_blocks_per_group);
4299                         goto failed_mount;
4300                 }
4301                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
4302                 sbi->s_cluster_bits = 0;
4303         }
4304         sbi->s_cluster_ratio = clustersize / blocksize;
4305
4306         /* Do we have standard group size of clustersize * 8 blocks ? */
4307         if (sbi->s_blocks_per_group == clustersize << 3)
4308                 set_opt2(sb, STD_GROUP_SIZE);
4309
4310         /*
4311          * Test whether we have more sectors than will fit in sector_t,
4312          * and whether the max offset is addressable by the page cache.
4313          */
4314         err = generic_check_addressable(sb->s_blocksize_bits,
4315                                         ext4_blocks_count(es));
4316         if (err) {
4317                 ext4_msg(sb, KERN_ERR, "filesystem"
4318                          " too large to mount safely on this system");
4319                 goto failed_mount;
4320         }
4321
4322         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
4323                 goto cantfind_ext4;
4324
4325         /* check blocks count against device size */
4326         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
4327         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
4328                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
4329                        "exceeds size of device (%llu blocks)",
4330                        ext4_blocks_count(es), blocks_count);
4331                 goto failed_mount;
4332         }
4333
4334         /*
4335          * It makes no sense for the first data block to be beyond the end
4336          * of the filesystem.
4337          */
4338         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
4339                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4340                          "block %u is beyond end of filesystem (%llu)",
4341                          le32_to_cpu(es->s_first_data_block),
4342                          ext4_blocks_count(es));
4343                 goto failed_mount;
4344         }
4345         if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) &&
4346             (sbi->s_cluster_ratio == 1)) {
4347                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4348                          "block is 0 with a 1k block and cluster size");
4349                 goto failed_mount;
4350         }
4351
4352         blocks_count = (ext4_blocks_count(es) -
4353                         le32_to_cpu(es->s_first_data_block) +
4354                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
4355         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
4356         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
4357                 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu "
4358                        "(block count %llu, first data block %u, "
4359                        "blocks per group %lu)", blocks_count,
4360                        ext4_blocks_count(es),
4361                        le32_to_cpu(es->s_first_data_block),
4362                        EXT4_BLOCKS_PER_GROUP(sb));
4363                 goto failed_mount;
4364         }
4365         sbi->s_groups_count = blocks_count;
4366         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
4367                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
4368         if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) !=
4369             le32_to_cpu(es->s_inodes_count)) {
4370                 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu",
4371                          le32_to_cpu(es->s_inodes_count),
4372                          ((u64)sbi->s_groups_count * sbi->s_inodes_per_group));
4373                 ret = -EINVAL;
4374                 goto failed_mount;
4375         }
4376         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
4377                    EXT4_DESC_PER_BLOCK(sb);
4378         if (ext4_has_feature_meta_bg(sb)) {
4379                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
4380                         ext4_msg(sb, KERN_WARNING,
4381                                  "first meta block group too large: %u "
4382                                  "(group descriptor block count %u)",
4383                                  le32_to_cpu(es->s_first_meta_bg), db_count);
4384                         goto failed_mount;
4385                 }
4386         }
4387         rcu_assign_pointer(sbi->s_group_desc,
4388                            kvmalloc_array(db_count,
4389                                           sizeof(struct buffer_head *),
4390                                           GFP_KERNEL));
4391         if (sbi->s_group_desc == NULL) {
4392                 ext4_msg(sb, KERN_ERR, "not enough memory");
4393                 ret = -ENOMEM;
4394                 goto failed_mount;
4395         }
4396
4397         bgl_lock_init(sbi->s_blockgroup_lock);
4398
4399         /* Pre-read the descriptors into the buffer cache */
4400         for (i = 0; i < db_count; i++) {
4401                 block = descriptor_loc(sb, logical_sb_block, i);
4402                 sb_breadahead_unmovable(sb, block);
4403         }
4404
4405         for (i = 0; i < db_count; i++) {
4406                 struct buffer_head *bh;
4407
4408                 block = descriptor_loc(sb, logical_sb_block, i);
4409                 bh = sb_bread_unmovable(sb, block);
4410                 if (!bh) {
4411                         ext4_msg(sb, KERN_ERR,
4412                                "can't read group descriptor %d", i);
4413                         db_count = i;
4414                         goto failed_mount2;
4415                 }
4416                 rcu_read_lock();
4417                 rcu_dereference(sbi->s_group_desc)[i] = bh;
4418                 rcu_read_unlock();
4419         }
4420         sbi->s_gdb_count = db_count;
4421         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
4422                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
4423                 ret = -EFSCORRUPTED;
4424                 goto failed_mount2;
4425         }
4426
4427         timer_setup(&sbi->s_err_report, print_daily_error_info, 0);
4428
4429         /* Register extent status tree shrinker */
4430         if (ext4_es_register_shrinker(sbi))
4431                 goto failed_mount3;
4432
4433         sbi->s_stripe = ext4_get_stripe_size(sbi);
4434         sbi->s_extent_max_zeroout_kb = 32;
4435
4436         /*
4437          * set up enough so that it can read an inode
4438          */
4439         sb->s_op = &ext4_sops;
4440         sb->s_export_op = &ext4_export_ops;
4441         sb->s_xattr = ext4_xattr_handlers;
4442 #ifdef CONFIG_FS_ENCRYPTION
4443         sb->s_cop = &ext4_cryptops;
4444 #endif
4445 #ifdef CONFIG_FS_VERITY
4446         sb->s_vop = &ext4_verityops;
4447 #endif
4448 #ifdef CONFIG_QUOTA
4449         sb->dq_op = &ext4_quota_operations;
4450         if (ext4_has_feature_quota(sb))
4451                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4452         else
4453                 sb->s_qcop = &ext4_qctl_operations;
4454         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
4455 #endif
4456         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
4457
4458         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
4459         mutex_init(&sbi->s_orphan_lock);
4460
4461         sb->s_root = NULL;
4462
4463         needs_recovery = (es->s_last_orphan != 0 ||
4464                           ext4_has_feature_journal_needs_recovery(sb));
4465
4466         if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb))
4467                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
4468                         goto failed_mount3a;
4469
4470         /*
4471          * The first inode we look at is the journal inode.  Don't try
4472          * root first: it may be modified in the journal!
4473          */
4474         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
4475                 err = ext4_load_journal(sb, es, journal_devnum);
4476                 if (err)
4477                         goto failed_mount3a;
4478         } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
4479                    ext4_has_feature_journal_needs_recovery(sb)) {
4480                 ext4_msg(sb, KERN_ERR, "required journal recovery "
4481                        "suppressed and not mounted read-only");
4482                 goto failed_mount_wq;
4483         } else {
4484                 /* Nojournal mode, all journal mount options are illegal */
4485                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
4486                         ext4_msg(sb, KERN_ERR, "can't mount with "
4487                                  "journal_checksum, fs mounted w/o journal");
4488                         goto failed_mount_wq;
4489                 }
4490                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4491                         ext4_msg(sb, KERN_ERR, "can't mount with "
4492                                  "journal_async_commit, fs mounted w/o journal");
4493                         goto failed_mount_wq;
4494                 }
4495                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
4496                         ext4_msg(sb, KERN_ERR, "can't mount with "
4497                                  "commit=%lu, fs mounted w/o journal",
4498                                  sbi->s_commit_interval / HZ);
4499                         goto failed_mount_wq;
4500                 }
4501                 if (EXT4_MOUNT_DATA_FLAGS &
4502                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
4503                         ext4_msg(sb, KERN_ERR, "can't mount with "
4504                                  "data=, fs mounted w/o journal");
4505                         goto failed_mount_wq;
4506                 }
4507                 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM;
4508                 clear_opt(sb, JOURNAL_CHECKSUM);
4509                 clear_opt(sb, DATA_FLAGS);
4510                 sbi->s_journal = NULL;
4511                 needs_recovery = 0;
4512                 goto no_journal;
4513         }
4514
4515         if (ext4_has_feature_64bit(sb) &&
4516             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4517                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4518                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4519                 goto failed_mount_wq;
4520         }
4521
4522         if (!set_journal_csum_feature_set(sb)) {
4523                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4524                          "feature set");
4525                 goto failed_mount_wq;
4526         }
4527
4528         /* We have now updated the journal if required, so we can
4529          * validate the data journaling mode. */
4530         switch (test_opt(sb, DATA_FLAGS)) {
4531         case 0:
4532                 /* No mode set, assume a default based on the journal
4533                  * capabilities: ORDERED_DATA if the journal can
4534                  * cope, else JOURNAL_DATA
4535                  */
4536                 if (jbd2_journal_check_available_features
4537                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4538                         set_opt(sb, ORDERED_DATA);
4539                         sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
4540                 } else {
4541                         set_opt(sb, JOURNAL_DATA);
4542                         sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
4543                 }
4544                 break;
4545
4546         case EXT4_MOUNT_ORDERED_DATA:
4547         case EXT4_MOUNT_WRITEBACK_DATA:
4548                 if (!jbd2_journal_check_available_features
4549                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4550                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4551                                "requested data journaling mode");
4552                         goto failed_mount_wq;
4553                 }
4554         default:
4555                 break;
4556         }
4557
4558         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4559             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4560                 ext4_msg(sb, KERN_ERR, "can't mount with "
4561                         "journal_async_commit in data=ordered mode");
4562                 goto failed_mount_wq;
4563         }
4564
4565         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4566
4567         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4568
4569 no_journal:
4570         if (!test_opt(sb, NO_MBCACHE)) {
4571                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
4572                 if (!sbi->s_ea_block_cache) {
4573                         ext4_msg(sb, KERN_ERR,
4574                                  "Failed to create ea_block_cache");
4575                         goto failed_mount_wq;
4576                 }
4577
4578                 if (ext4_has_feature_ea_inode(sb)) {
4579                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
4580                         if (!sbi->s_ea_inode_cache) {
4581                                 ext4_msg(sb, KERN_ERR,
4582                                          "Failed to create ea_inode_cache");
4583                                 goto failed_mount_wq;
4584                         }
4585                 }
4586         }
4587
4588         if (ext4_has_feature_verity(sb) && blocksize != PAGE_SIZE) {
4589                 ext4_msg(sb, KERN_ERR, "Unsupported blocksize for fs-verity");
4590                 goto failed_mount_wq;
4591         }
4592
4593         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) &&
4594             !ext4_has_feature_encrypt(sb)) {
4595                 ext4_set_feature_encrypt(sb);
4596                 ext4_commit_super(sb, 1);
4597         }
4598
4599         /*
4600          * Get the # of file system overhead blocks from the
4601          * superblock if present.
4602          */
4603         if (es->s_overhead_clusters)
4604                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4605         else {
4606                 err = ext4_calculate_overhead(sb);
4607                 if (err)
4608                         goto failed_mount_wq;
4609         }
4610
4611         /*
4612          * The maximum number of concurrent works can be high and
4613          * concurrency isn't really necessary.  Limit it to 1.
4614          */
4615         EXT4_SB(sb)->rsv_conversion_wq =
4616                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4617         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4618                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4619                 ret = -ENOMEM;
4620                 goto failed_mount4;
4621         }
4622
4623         /*
4624          * The jbd2_journal_load will have done any necessary log recovery,
4625          * so we can safely mount the rest of the filesystem now.
4626          */
4627
4628         root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL);
4629         if (IS_ERR(root)) {
4630                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4631                 ret = PTR_ERR(root);
4632                 root = NULL;
4633                 goto failed_mount4;
4634         }
4635         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4636                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4637                 iput(root);
4638                 goto failed_mount4;
4639         }
4640
4641 #ifdef CONFIG_UNICODE
4642         if (sbi->s_encoding)
4643                 sb->s_d_op = &ext4_dentry_ops;
4644 #endif
4645
4646         sb->s_root = d_make_root(root);
4647         if (!sb->s_root) {
4648                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4649                 ret = -ENOMEM;
4650                 goto failed_mount4;
4651         }
4652
4653         ret = ext4_setup_super(sb, es, sb_rdonly(sb));
4654         if (ret == -EROFS) {
4655                 sb->s_flags |= SB_RDONLY;
4656                 ret = 0;
4657         } else if (ret)
4658                 goto failed_mount4a;
4659
4660         ext4_set_resv_clusters(sb);
4661
4662         err = ext4_setup_system_zone(sb);
4663         if (err) {
4664                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4665                          "zone (%d)", err);
4666                 goto failed_mount4a;
4667         }
4668
4669         ext4_ext_init(sb);
4670         err = ext4_mb_init(sb);
4671         if (err) {
4672                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4673                          err);
4674                 goto failed_mount5;
4675         }
4676
4677         block = ext4_count_free_clusters(sb);
4678         ext4_free_blocks_count_set(sbi->s_es,
4679                                    EXT4_C2B(sbi, block));
4680         ext4_superblock_csum_set(sb);
4681         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4682                                   GFP_KERNEL);
4683         if (!err) {
4684                 unsigned long freei = ext4_count_free_inodes(sb);
4685                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4686                 ext4_superblock_csum_set(sb);
4687                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4688                                           GFP_KERNEL);
4689         }
4690         if (!err)
4691                 err = percpu_counter_init(&sbi->s_dirs_counter,
4692                                           ext4_count_dirs(sb), GFP_KERNEL);
4693         if (!err)
4694                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4695                                           GFP_KERNEL);
4696         if (!err)
4697                 err = percpu_init_rwsem(&sbi->s_writepages_rwsem);
4698
4699         if (err) {
4700                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4701                 goto failed_mount6;
4702         }
4703
4704         if (ext4_has_feature_flex_bg(sb))
4705                 if (!ext4_fill_flex_info(sb)) {
4706                         ext4_msg(sb, KERN_ERR,
4707                                "unable to initialize "
4708                                "flex_bg meta info!");
4709                         goto failed_mount6;
4710                 }
4711
4712         err = ext4_register_li_request(sb, first_not_zeroed);
4713         if (err)
4714                 goto failed_mount6;
4715
4716         err = ext4_register_sysfs(sb);
4717         if (err)
4718                 goto failed_mount7;
4719
4720 #ifdef CONFIG_QUOTA
4721         /* Enable quota usage during mount. */
4722         if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
4723                 err = ext4_enable_quotas(sb);
4724                 if (err)
4725                         goto failed_mount8;
4726         }
4727 #endif  /* CONFIG_QUOTA */
4728
4729         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4730         ext4_orphan_cleanup(sb, es);
4731         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4732         if (needs_recovery) {
4733                 ext4_msg(sb, KERN_INFO, "recovery complete");
4734                 ext4_mark_recovery_complete(sb, es);
4735         }
4736         if (EXT4_SB(sb)->s_journal) {
4737                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4738                         descr = " journalled data mode";
4739                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4740                         descr = " ordered data mode";
4741                 else
4742                         descr = " writeback data mode";
4743         } else
4744                 descr = "out journal";
4745
4746         if (test_opt(sb, DISCARD)) {
4747                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4748                 if (!blk_queue_discard(q))
4749                         ext4_msg(sb, KERN_WARNING,
4750                                  "mounting with \"discard\" option, but "
4751                                  "the device does not support discard");
4752         }
4753
4754         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4755                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4756                          "Opts: %.*s%s%s", descr,
4757                          (int) sizeof(sbi->s_es->s_mount_opts),
4758                          sbi->s_es->s_mount_opts,
4759                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4760
4761         if (es->s_error_count)
4762                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4763
4764         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4765         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4766         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4767         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4768
4769         kfree(orig_data);
4770         return 0;
4771
4772 cantfind_ext4:
4773         if (!silent)
4774                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4775         goto failed_mount;
4776
4777 #ifdef CONFIG_QUOTA
4778 failed_mount8:
4779         ext4_unregister_sysfs(sb);
4780 #endif
4781 failed_mount7:
4782         ext4_unregister_li_request(sb);
4783 failed_mount6:
4784         ext4_mb_release(sb);
4785         rcu_read_lock();
4786         flex_groups = rcu_dereference(sbi->s_flex_groups);
4787         if (flex_groups) {
4788                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
4789                         kvfree(flex_groups[i]);
4790                 kvfree(flex_groups);
4791         }
4792         rcu_read_unlock();
4793         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4794         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4795         percpu_counter_destroy(&sbi->s_dirs_counter);
4796         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4797         percpu_free_rwsem(&sbi->s_writepages_rwsem);
4798 failed_mount5:
4799         ext4_ext_release(sb);
4800         ext4_release_system_zone(sb);
4801 failed_mount4a:
4802         dput(sb->s_root);
4803         sb->s_root = NULL;
4804 failed_mount4:
4805         ext4_msg(sb, KERN_ERR, "mount failed");
4806         if (EXT4_SB(sb)->rsv_conversion_wq)
4807                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4808 failed_mount_wq:
4809         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
4810         sbi->s_ea_inode_cache = NULL;
4811
4812         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
4813         sbi->s_ea_block_cache = NULL;
4814
4815         if (sbi->s_journal) {
4816                 jbd2_journal_destroy(sbi->s_journal);
4817                 sbi->s_journal = NULL;
4818         }
4819 failed_mount3a:
4820         ext4_es_unregister_shrinker(sbi);
4821 failed_mount3:
4822         del_timer_sync(&sbi->s_err_report);
4823         if (sbi->s_mmp_tsk)
4824                 kthread_stop(sbi->s_mmp_tsk);
4825 failed_mount2:
4826         rcu_read_lock();
4827         group_desc = rcu_dereference(sbi->s_group_desc);
4828         for (i = 0; i < db_count; i++)
4829                 brelse(group_desc[i]);
4830         kvfree(group_desc);
4831         rcu_read_unlock();
4832 failed_mount:
4833         if (sbi->s_chksum_driver)
4834                 crypto_free_shash(sbi->s_chksum_driver);
4835
4836 #ifdef CONFIG_UNICODE
4837         utf8_unload(sbi->s_encoding);
4838 #endif
4839
4840 #ifdef CONFIG_QUOTA
4841         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4842                 kfree(get_qf_name(sb, sbi, i));
4843 #endif
4844         ext4_blkdev_remove(sbi);
4845         brelse(bh);
4846 out_fail:
4847         sb->s_fs_info = NULL;
4848         kfree(sbi->s_blockgroup_lock);
4849 out_free_base:
4850         kfree(sbi);
4851         kfree(orig_data);
4852         fs_put_dax(dax_dev);
4853         return err ? err : ret;
4854 }
4855
4856 /*
4857  * Setup any per-fs journal parameters now.  We'll do this both on
4858  * initial mount, once the journal has been initialised but before we've
4859  * done any recovery; and again on any subsequent remount.
4860  */
4861 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4862 {
4863         struct ext4_sb_info *sbi = EXT4_SB(sb);
4864
4865         journal->j_commit_interval = sbi->s_commit_interval;
4866         journal->j_min_batch_time = sbi->s_min_batch_time;
4867         journal->j_max_batch_time = sbi->s_max_batch_time;
4868
4869         write_lock(&journal->j_state_lock);
4870         if (test_opt(sb, BARRIER))
4871                 journal->j_flags |= JBD2_BARRIER;
4872         else
4873                 journal->j_flags &= ~JBD2_BARRIER;
4874         if (test_opt(sb, DATA_ERR_ABORT))
4875                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4876         else
4877                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4878         write_unlock(&journal->j_state_lock);
4879 }
4880
4881 static struct inode *ext4_get_journal_inode(struct super_block *sb,
4882                                              unsigned int journal_inum)
4883 {
4884         struct inode *journal_inode;
4885
4886         /*
4887          * Test for the existence of a valid inode on disk.  Bad things
4888          * happen if we iget() an unused inode, as the subsequent iput()
4889          * will try to delete it.
4890          */
4891         journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL);
4892         if (IS_ERR(journal_inode)) {
4893                 ext4_msg(sb, KERN_ERR, "no journal found");
4894                 return NULL;
4895         }
4896         if (!journal_inode->i_nlink) {
4897                 make_bad_inode(journal_inode);
4898                 iput(journal_inode);
4899                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4900                 return NULL;
4901         }
4902
4903         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4904                   journal_inode, journal_inode->i_size);
4905         if (!S_ISREG(journal_inode->i_mode)) {
4906                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4907                 iput(journal_inode);
4908                 return NULL;
4909         }
4910         return journal_inode;
4911 }
4912
4913 static journal_t *ext4_get_journal(struct super_block *sb,
4914                                    unsigned int journal_inum)
4915 {
4916         struct inode *journal_inode;
4917         journal_t *journal;
4918
4919         BUG_ON(!ext4_has_feature_journal(sb));
4920
4921         journal_inode = ext4_get_journal_inode(sb, journal_inum);
4922         if (!journal_inode)
4923                 return NULL;
4924
4925         journal = jbd2_journal_init_inode(journal_inode);
4926         if (!journal) {
4927                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4928                 iput(journal_inode);
4929                 return NULL;
4930         }
4931         journal->j_private = sb;
4932         ext4_init_journal_params(sb, journal);
4933         return journal;
4934 }
4935
4936 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4937                                        dev_t j_dev)
4938 {
4939         struct buffer_head *bh;
4940         journal_t *journal;
4941         ext4_fsblk_t start;
4942         ext4_fsblk_t len;
4943         int hblock, blocksize;
4944         ext4_fsblk_t sb_block;
4945         unsigned long offset;
4946         struct ext4_super_block *es;
4947         struct block_device *bdev;
4948
4949         BUG_ON(!ext4_has_feature_journal(sb));
4950
4951         bdev = ext4_blkdev_get(j_dev, sb);
4952         if (bdev == NULL)
4953                 return NULL;
4954
4955         blocksize = sb->s_blocksize;
4956         hblock = bdev_logical_block_size(bdev);
4957         if (blocksize < hblock) {
4958                 ext4_msg(sb, KERN_ERR,
4959                         "blocksize too small for journal device");
4960                 goto out_bdev;
4961         }
4962
4963         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4964         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4965         set_blocksize(bdev, blocksize);
4966         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4967                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4968                        "external journal");
4969                 goto out_bdev;
4970         }
4971
4972         es = (struct ext4_super_block *) (bh->b_data + offset);
4973         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4974             !(le32_to_cpu(es->s_feature_incompat) &
4975               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4976                 ext4_msg(sb, KERN_ERR, "external journal has "
4977                                         "bad superblock");
4978                 brelse(bh);
4979                 goto out_bdev;
4980         }
4981
4982         if ((le32_to_cpu(es->s_feature_ro_compat) &
4983              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4984             es->s_checksum != ext4_superblock_csum(sb, es)) {
4985                 ext4_msg(sb, KERN_ERR, "external journal has "
4986                                        "corrupt superblock");
4987                 brelse(bh);
4988                 goto out_bdev;
4989         }
4990
4991         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4992                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4993                 brelse(bh);
4994                 goto out_bdev;
4995         }
4996
4997         len = ext4_blocks_count(es);
4998         start = sb_block + 1;
4999         brelse(bh);     /* we're done with the superblock */
5000
5001         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
5002                                         start, len, blocksize);
5003         if (!journal) {
5004                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
5005                 goto out_bdev;
5006         }
5007         journal->j_private = sb;
5008         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
5009         wait_on_buffer(journal->j_sb_buffer);
5010         if (!buffer_uptodate(journal->j_sb_buffer)) {
5011                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
5012                 goto out_journal;
5013         }
5014         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
5015                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
5016                                         "user (unsupported) - %d",
5017                         be32_to_cpu(journal->j_superblock->s_nr_users));
5018                 goto out_journal;
5019         }
5020         EXT4_SB(sb)->journal_bdev = bdev;
5021         ext4_init_journal_params(sb, journal);
5022         return journal;
5023
5024 out_journal:
5025         jbd2_journal_destroy(journal);
5026 out_bdev:
5027         ext4_blkdev_put(bdev);
5028         return NULL;
5029 }
5030
5031 static int ext4_load_journal(struct super_block *sb,
5032                              struct ext4_super_block *es,
5033                              unsigned long journal_devnum)
5034 {
5035         journal_t *journal;
5036         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
5037         dev_t journal_dev;
5038         int err = 0;
5039         int really_read_only;
5040
5041         BUG_ON(!ext4_has_feature_journal(sb));
5042
5043         if (journal_devnum &&
5044             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5045                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
5046                         "numbers have changed");
5047                 journal_dev = new_decode_dev(journal_devnum);
5048         } else
5049                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
5050
5051         really_read_only = bdev_read_only(sb->s_bdev);
5052
5053         /*
5054          * Are we loading a blank journal or performing recovery after a
5055          * crash?  For recovery, we need to check in advance whether we
5056          * can get read-write access to the device.
5057          */
5058         if (ext4_has_feature_journal_needs_recovery(sb)) {
5059                 if (sb_rdonly(sb)) {
5060                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
5061                                         "required on readonly filesystem");
5062                         if (really_read_only) {
5063                                 ext4_msg(sb, KERN_ERR, "write access "
5064                                         "unavailable, cannot proceed "
5065                                         "(try mounting with noload)");
5066                                 return -EROFS;
5067                         }
5068                         ext4_msg(sb, KERN_INFO, "write access will "
5069                                "be enabled during recovery");
5070                 }
5071         }
5072
5073         if (journal_inum && journal_dev) {
5074                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
5075                        "and inode journals!");
5076                 return -EINVAL;
5077         }
5078
5079         if (journal_inum) {
5080                 if (!(journal = ext4_get_journal(sb, journal_inum)))
5081                         return -EINVAL;
5082         } else {
5083                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
5084                         return -EINVAL;
5085         }
5086
5087         if (!(journal->j_flags & JBD2_BARRIER))
5088                 ext4_msg(sb, KERN_INFO, "barriers disabled");
5089
5090         if (!ext4_has_feature_journal_needs_recovery(sb))
5091                 err = jbd2_journal_wipe(journal, !really_read_only);
5092         if (!err) {
5093                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
5094                 if (save)
5095                         memcpy(save, ((char *) es) +
5096                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
5097                 err = jbd2_journal_load(journal);
5098                 if (save)
5099                         memcpy(((char *) es) + EXT4_S_ERR_START,
5100                                save, EXT4_S_ERR_LEN);
5101                 kfree(save);
5102         }
5103
5104         if (err) {
5105                 ext4_msg(sb, KERN_ERR, "error loading journal");
5106                 jbd2_journal_destroy(journal);
5107                 return err;
5108         }
5109
5110         EXT4_SB(sb)->s_journal = journal;
5111         ext4_clear_journal_err(sb, es);
5112
5113         if (!really_read_only && journal_devnum &&
5114             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5115                 es->s_journal_dev = cpu_to_le32(journal_devnum);
5116
5117                 /* Make sure we flush the recovery flag to disk. */
5118                 ext4_commit_super(sb, 1);
5119         }
5120
5121         return 0;
5122 }
5123
5124 static int ext4_commit_super(struct super_block *sb, int sync)
5125 {
5126         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
5127         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
5128         int error = 0;
5129
5130         if (!sbh || block_device_ejected(sb))
5131                 return error;
5132
5133         /*
5134          * The superblock bh should be mapped, but it might not be if the
5135          * device was hot-removed. Not much we can do but fail the I/O.
5136          */
5137         if (!buffer_mapped(sbh))
5138                 return error;
5139
5140         /*
5141          * If the file system is mounted read-only, don't update the
5142          * superblock write time.  This avoids updating the superblock
5143          * write time when we are mounting the root file system
5144          * read/only but we need to replay the journal; at that point,
5145          * for people who are east of GMT and who make their clock
5146          * tick in localtime for Windows bug-for-bug compatibility,
5147          * the clock is set in the future, and this will cause e2fsck
5148          * to complain and force a full file system check.
5149          */
5150         if (!(sb->s_flags & SB_RDONLY))
5151                 ext4_update_tstamp(es, s_wtime);
5152         if (sb->s_bdev->bd_part)
5153                 es->s_kbytes_written =
5154                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
5155                             ((part_stat_read(sb->s_bdev->bd_part,
5156                                              sectors[STAT_WRITE]) -
5157                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
5158         else
5159                 es->s_kbytes_written =
5160                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
5161         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
5162                 ext4_free_blocks_count_set(es,
5163                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
5164                                 &EXT4_SB(sb)->s_freeclusters_counter)));
5165         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
5166                 es->s_free_inodes_count =
5167                         cpu_to_le32(percpu_counter_sum_positive(
5168                                 &EXT4_SB(sb)->s_freeinodes_counter));
5169         BUFFER_TRACE(sbh, "marking dirty");
5170         ext4_superblock_csum_set(sb);
5171         if (sync)
5172                 lock_buffer(sbh);
5173         if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
5174                 /*
5175                  * Oh, dear.  A previous attempt to write the
5176                  * superblock failed.  This could happen because the
5177                  * USB device was yanked out.  Or it could happen to
5178                  * be a transient write error and maybe the block will
5179                  * be remapped.  Nothing we can do but to retry the
5180                  * write and hope for the best.
5181                  */
5182                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
5183                        "superblock detected");
5184                 clear_buffer_write_io_error(sbh);
5185                 set_buffer_uptodate(sbh);
5186         }
5187         mark_buffer_dirty(sbh);
5188         if (sync) {
5189                 unlock_buffer(sbh);
5190                 error = __sync_dirty_buffer(sbh,
5191                         REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
5192                 if (buffer_write_io_error(sbh)) {
5193                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
5194                                "superblock");
5195                         clear_buffer_write_io_error(sbh);
5196                         set_buffer_uptodate(sbh);
5197                 }
5198         }
5199         return error;
5200 }
5201
5202 /*
5203  * Have we just finished recovery?  If so, and if we are mounting (or
5204  * remounting) the filesystem readonly, then we will end up with a
5205  * consistent fs on disk.  Record that fact.
5206  */
5207 static void ext4_mark_recovery_complete(struct super_block *sb,
5208                                         struct ext4_super_block *es)
5209 {
5210         journal_t *journal = EXT4_SB(sb)->s_journal;
5211
5212         if (!ext4_has_feature_journal(sb)) {
5213                 BUG_ON(journal != NULL);
5214                 return;
5215         }
5216         jbd2_journal_lock_updates(journal);
5217         if (jbd2_journal_flush(journal) < 0)
5218                 goto out;
5219
5220         if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) {
5221                 ext4_clear_feature_journal_needs_recovery(sb);
5222                 ext4_commit_super(sb, 1);
5223         }
5224
5225 out:
5226         jbd2_journal_unlock_updates(journal);
5227 }
5228
5229 /*
5230  * If we are mounting (or read-write remounting) a filesystem whose journal
5231  * has recorded an error from a previous lifetime, move that error to the
5232  * main filesystem now.
5233  */
5234 static void ext4_clear_journal_err(struct super_block *sb,
5235                                    struct ext4_super_block *es)
5236 {
5237         journal_t *journal;
5238         int j_errno;
5239         const char *errstr;
5240
5241         BUG_ON(!ext4_has_feature_journal(sb));
5242
5243         journal = EXT4_SB(sb)->s_journal;
5244
5245         /*
5246          * Now check for any error status which may have been recorded in the
5247          * journal by a prior ext4_error() or ext4_abort()
5248          */
5249
5250         j_errno = jbd2_journal_errno(journal);
5251         if (j_errno) {
5252                 char nbuf[16];
5253
5254                 errstr = ext4_decode_error(sb, j_errno, nbuf);
5255                 ext4_warning(sb, "Filesystem error recorded "
5256                              "from previous mount: %s", errstr);
5257                 ext4_warning(sb, "Marking fs in need of filesystem check.");
5258
5259                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
5260                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
5261                 ext4_commit_super(sb, 1);
5262
5263                 jbd2_journal_clear_err(journal);
5264                 jbd2_journal_update_sb_errno(journal);
5265         }
5266 }
5267
5268 /*
5269  * Force the running and committing transactions to commit,
5270  * and wait on the commit.
5271  */
5272 int ext4_force_commit(struct super_block *sb)
5273 {
5274         journal_t *journal;
5275
5276         if (sb_rdonly(sb))
5277                 return 0;
5278
5279         journal = EXT4_SB(sb)->s_journal;
5280         return ext4_journal_force_commit(journal);
5281 }
5282
5283 static int ext4_sync_fs(struct super_block *sb, int wait)
5284 {
5285         int ret = 0;
5286         tid_t target;
5287         bool needs_barrier = false;
5288         struct ext4_sb_info *sbi = EXT4_SB(sb);
5289
5290         if (unlikely(ext4_forced_shutdown(sbi)))
5291                 return 0;
5292
5293         trace_ext4_sync_fs(sb, wait);
5294         flush_workqueue(sbi->rsv_conversion_wq);
5295         /*
5296          * Writeback quota in non-journalled quota case - journalled quota has
5297          * no dirty dquots
5298          */
5299         dquot_writeback_dquots(sb, -1);
5300         /*
5301          * Data writeback is possible w/o journal transaction, so barrier must
5302          * being sent at the end of the function. But we can skip it if
5303          * transaction_commit will do it for us.
5304          */
5305         if (sbi->s_journal) {
5306                 target = jbd2_get_latest_transaction(sbi->s_journal);
5307                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
5308                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
5309                         needs_barrier = true;
5310
5311                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
5312                         if (wait)
5313                                 ret = jbd2_log_wait_commit(sbi->s_journal,
5314                                                            target);
5315                 }
5316         } else if (wait && test_opt(sb, BARRIER))
5317                 needs_barrier = true;
5318         if (needs_barrier) {
5319                 int err;
5320                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
5321                 if (!ret)
5322                         ret = err;
5323         }
5324
5325         return ret;
5326 }
5327
5328 /*
5329  * LVM calls this function before a (read-only) snapshot is created.  This
5330  * gives us a chance to flush the journal completely and mark the fs clean.
5331  *
5332  * Note that only this function cannot bring a filesystem to be in a clean
5333  * state independently. It relies on upper layer to stop all data & metadata
5334  * modifications.
5335  */
5336 static int ext4_freeze(struct super_block *sb)
5337 {
5338         int error = 0;
5339         journal_t *journal;
5340
5341         if (sb_rdonly(sb))
5342                 return 0;
5343
5344         journal = EXT4_SB(sb)->s_journal;
5345
5346         if (journal) {
5347                 /* Now we set up the journal barrier. */
5348                 jbd2_journal_lock_updates(journal);
5349
5350                 /*
5351                  * Don't clear the needs_recovery flag if we failed to
5352                  * flush the journal.
5353                  */
5354                 error = jbd2_journal_flush(journal);
5355                 if (error < 0)
5356                         goto out;
5357
5358                 /* Journal blocked and flushed, clear needs_recovery flag. */
5359                 ext4_clear_feature_journal_needs_recovery(sb);
5360         }
5361
5362         error = ext4_commit_super(sb, 1);
5363 out:
5364         if (journal)
5365                 /* we rely on upper layer to stop further updates */
5366                 jbd2_journal_unlock_updates(journal);
5367         return error;
5368 }
5369
5370 /*
5371  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
5372  * flag here, even though the filesystem is not technically dirty yet.
5373  */
5374 static int ext4_unfreeze(struct super_block *sb)
5375 {
5376         if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
5377                 return 0;
5378
5379         if (EXT4_SB(sb)->s_journal) {
5380                 /* Reset the needs_recovery flag before the fs is unlocked. */
5381                 ext4_set_feature_journal_needs_recovery(sb);
5382         }
5383
5384         ext4_commit_super(sb, 1);
5385         return 0;
5386 }
5387
5388 /*
5389  * Structure to save mount options for ext4_remount's benefit
5390  */
5391 struct ext4_mount_options {
5392         unsigned long s_mount_opt;
5393         unsigned long s_mount_opt2;
5394         kuid_t s_resuid;
5395         kgid_t s_resgid;
5396         unsigned long s_commit_interval;
5397         u32 s_min_batch_time, s_max_batch_time;
5398 #ifdef CONFIG_QUOTA
5399         int s_jquota_fmt;
5400         char *s_qf_names[EXT4_MAXQUOTAS];
5401 #endif
5402 };
5403
5404 static int ext4_remount(struct super_block *sb, int *flags, char *data)
5405 {
5406         struct ext4_super_block *es;
5407         struct ext4_sb_info *sbi = EXT4_SB(sb);
5408         unsigned long old_sb_flags;
5409         struct ext4_mount_options old_opts;
5410         int enable_quota = 0;
5411         ext4_group_t g;
5412         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
5413         int err = 0;
5414 #ifdef CONFIG_QUOTA
5415         int i, j;
5416         char *to_free[EXT4_MAXQUOTAS];
5417 #endif
5418         char *orig_data = kstrdup(data, GFP_KERNEL);
5419
5420         if (data && !orig_data)
5421                 return -ENOMEM;
5422
5423         /* Store the original options */
5424         old_sb_flags = sb->s_flags;
5425         old_opts.s_mount_opt = sbi->s_mount_opt;
5426         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
5427         old_opts.s_resuid = sbi->s_resuid;
5428         old_opts.s_resgid = sbi->s_resgid;
5429         old_opts.s_commit_interval = sbi->s_commit_interval;
5430         old_opts.s_min_batch_time = sbi->s_min_batch_time;
5431         old_opts.s_max_batch_time = sbi->s_max_batch_time;
5432 #ifdef CONFIG_QUOTA
5433         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
5434         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5435                 if (sbi->s_qf_names[i]) {
5436                         char *qf_name = get_qf_name(sb, sbi, i);
5437
5438                         old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL);
5439                         if (!old_opts.s_qf_names[i]) {
5440                                 for (j = 0; j < i; j++)
5441                                         kfree(old_opts.s_qf_names[j]);
5442                                 kfree(orig_data);
5443                                 return -ENOMEM;
5444                         }
5445                 } else
5446                         old_opts.s_qf_names[i] = NULL;
5447 #endif
5448         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
5449                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
5450
5451         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
5452                 err = -EINVAL;
5453                 goto restore_opts;
5454         }
5455
5456         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
5457             test_opt(sb, JOURNAL_CHECKSUM)) {
5458                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
5459                          "during remount not supported; ignoring");
5460                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
5461         }
5462
5463         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
5464                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
5465                         ext4_msg(sb, KERN_ERR, "can't mount with "
5466                                  "both data=journal and delalloc");
5467                         err = -EINVAL;
5468                         goto restore_opts;
5469                 }
5470                 if (test_opt(sb, DIOREAD_NOLOCK)) {
5471                         ext4_msg(sb, KERN_ERR, "can't mount with "
5472                                  "both data=journal and dioread_nolock");
5473                         err = -EINVAL;
5474                         goto restore_opts;
5475                 }
5476         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
5477                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5478                         ext4_msg(sb, KERN_ERR, "can't mount with "
5479                                 "journal_async_commit in data=ordered mode");
5480                         err = -EINVAL;
5481                         goto restore_opts;
5482                 }
5483         }
5484
5485         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
5486                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
5487                 err = -EINVAL;
5488                 goto restore_opts;
5489         }
5490
5491         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
5492                 ext4_abort(sb, EXT4_ERR_ESHUTDOWN, "Abort forced by user");
5493
5494         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
5495                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
5496
5497         es = sbi->s_es;
5498
5499         if (sbi->s_journal) {
5500                 ext4_init_journal_params(sb, sbi->s_journal);
5501                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
5502         }
5503
5504         if (*flags & SB_LAZYTIME)
5505                 sb->s_flags |= SB_LAZYTIME;
5506
5507         if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) {
5508                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
5509                         err = -EROFS;
5510                         goto restore_opts;
5511                 }
5512
5513                 if (*flags & SB_RDONLY) {
5514                         err = sync_filesystem(sb);
5515                         if (err < 0)
5516                                 goto restore_opts;
5517                         err = dquot_suspend(sb, -1);
5518                         if (err < 0)
5519                                 goto restore_opts;
5520
5521                         /*
5522                          * First of all, the unconditional stuff we have to do
5523                          * to disable replay of the journal when we next remount
5524                          */
5525                         sb->s_flags |= SB_RDONLY;
5526
5527                         /*
5528                          * OK, test if we are remounting a valid rw partition
5529                          * readonly, and if so set the rdonly flag and then
5530                          * mark the partition as valid again.
5531                          */
5532                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5533                             (sbi->s_mount_state & EXT4_VALID_FS))
5534                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5535
5536                         if (sbi->s_journal)
5537                                 ext4_mark_recovery_complete(sb, es);
5538                         if (sbi->s_mmp_tsk)
5539                                 kthread_stop(sbi->s_mmp_tsk);
5540                 } else {
5541                         /* Make sure we can mount this feature set readwrite */
5542                         if (ext4_has_feature_readonly(sb) ||
5543                             !ext4_feature_set_ok(sb, 0)) {
5544                                 err = -EROFS;
5545                                 goto restore_opts;
5546                         }
5547                         /*
5548                          * Make sure the group descriptor checksums
5549                          * are sane.  If they aren't, refuse to remount r/w.
5550                          */
5551                         for (g = 0; g < sbi->s_groups_count; g++) {
5552                                 struct ext4_group_desc *gdp =
5553                                         ext4_get_group_desc(sb, g, NULL);
5554
5555                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5556                                         ext4_msg(sb, KERN_ERR,
5557                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5558                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
5559                                                le16_to_cpu(gdp->bg_checksum));
5560                                         err = -EFSBADCRC;
5561                                         goto restore_opts;
5562                                 }
5563                         }
5564
5565                         /*
5566                          * If we have an unprocessed orphan list hanging
5567                          * around from a previously readonly bdev mount,
5568                          * require a full umount/remount for now.
5569                          */
5570                         if (es->s_last_orphan) {
5571                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5572                                        "remount RDWR because of unprocessed "
5573                                        "orphan inode list.  Please "
5574                                        "umount/remount instead");
5575                                 err = -EINVAL;
5576                                 goto restore_opts;
5577                         }
5578
5579                         /*
5580                          * Mounting a RDONLY partition read-write, so reread
5581                          * and store the current valid flag.  (It may have
5582                          * been changed by e2fsck since we originally mounted
5583                          * the partition.)
5584                          */
5585                         if (sbi->s_journal)
5586                                 ext4_clear_journal_err(sb, es);
5587                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5588
5589                         err = ext4_setup_super(sb, es, 0);
5590                         if (err)
5591                                 goto restore_opts;
5592
5593                         sb->s_flags &= ~SB_RDONLY;
5594                         if (ext4_has_feature_mmp(sb))
5595                                 if (ext4_multi_mount_protect(sb,
5596                                                 le64_to_cpu(es->s_mmp_block))) {
5597                                         err = -EROFS;
5598                                         goto restore_opts;
5599                                 }
5600                         enable_quota = 1;
5601                 }
5602         }
5603
5604         /*
5605          * Reinitialize lazy itable initialization thread based on
5606          * current settings
5607          */
5608         if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
5609                 ext4_unregister_li_request(sb);
5610         else {
5611                 ext4_group_t first_not_zeroed;
5612                 first_not_zeroed = ext4_has_uninit_itable(sb);
5613                 ext4_register_li_request(sb, first_not_zeroed);
5614         }
5615
5616         ext4_setup_system_zone(sb);
5617         if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) {
5618                 err = ext4_commit_super(sb, 1);
5619                 if (err)
5620                         goto restore_opts;
5621         }
5622
5623 #ifdef CONFIG_QUOTA
5624         /* Release old quota file names */
5625         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5626                 kfree(old_opts.s_qf_names[i]);
5627         if (enable_quota) {
5628                 if (sb_any_quota_suspended(sb))
5629                         dquot_resume(sb, -1);
5630                 else if (ext4_has_feature_quota(sb)) {
5631                         err = ext4_enable_quotas(sb);
5632                         if (err)
5633                                 goto restore_opts;
5634                 }
5635         }
5636 #endif
5637
5638         *flags = (*flags & ~SB_LAZYTIME) | (sb->s_flags & SB_LAZYTIME);
5639         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5640         kfree(orig_data);
5641         return 0;
5642
5643 restore_opts:
5644         sb->s_flags = old_sb_flags;
5645         sbi->s_mount_opt = old_opts.s_mount_opt;
5646         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5647         sbi->s_resuid = old_opts.s_resuid;
5648         sbi->s_resgid = old_opts.s_resgid;
5649         sbi->s_commit_interval = old_opts.s_commit_interval;
5650         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5651         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5652 #ifdef CONFIG_QUOTA
5653         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5654         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5655                 to_free[i] = get_qf_name(sb, sbi, i);
5656                 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]);
5657         }
5658         synchronize_rcu();
5659         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5660                 kfree(to_free[i]);
5661 #endif
5662         kfree(orig_data);
5663         return err;
5664 }
5665
5666 #ifdef CONFIG_QUOTA
5667 static int ext4_statfs_project(struct super_block *sb,
5668                                kprojid_t projid, struct kstatfs *buf)
5669 {
5670         struct kqid qid;
5671         struct dquot *dquot;
5672         u64 limit;
5673         u64 curblock;
5674
5675         qid = make_kqid_projid(projid);
5676         dquot = dqget(sb, qid);
5677         if (IS_ERR(dquot))
5678                 return PTR_ERR(dquot);
5679         spin_lock(&dquot->dq_dqb_lock);
5680
5681         limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit,
5682                              dquot->dq_dqb.dqb_bhardlimit);
5683         limit >>= sb->s_blocksize_bits;
5684
5685         if (limit && buf->f_blocks > limit) {
5686                 curblock = (dquot->dq_dqb.dqb_curspace +
5687                             dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits;
5688                 buf->f_blocks = limit;
5689                 buf->f_bfree = buf->f_bavail =
5690                         (buf->f_blocks > curblock) ?
5691                          (buf->f_blocks - curblock) : 0;
5692         }
5693
5694         limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit,
5695                              dquot->dq_dqb.dqb_ihardlimit);
5696         if (limit && buf->f_files > limit) {
5697                 buf->f_files = limit;
5698                 buf->f_ffree =
5699                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
5700                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
5701         }
5702
5703         spin_unlock(&dquot->dq_dqb_lock);
5704         dqput(dquot);
5705         return 0;
5706 }
5707 #endif
5708
5709 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5710 {
5711         struct super_block *sb = dentry->d_sb;
5712         struct ext4_sb_info *sbi = EXT4_SB(sb);
5713         struct ext4_super_block *es = sbi->s_es;
5714         ext4_fsblk_t overhead = 0, resv_blocks;
5715         u64 fsid;
5716         s64 bfree;
5717         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5718
5719         if (!test_opt(sb, MINIX_DF))
5720                 overhead = sbi->s_overhead;
5721
5722         buf->f_type = EXT4_SUPER_MAGIC;
5723         buf->f_bsize = sb->s_blocksize;
5724         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5725         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5726                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5727         /* prevent underflow in case that few free space is available */
5728         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5729         buf->f_bavail = buf->f_bfree -
5730                         (ext4_r_blocks_count(es) + resv_blocks);
5731         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5732                 buf->f_bavail = 0;
5733         buf->f_files = le32_to_cpu(es->s_inodes_count);
5734         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5735         buf->f_namelen = EXT4_NAME_LEN;
5736         fsid = le64_to_cpup((void *)es->s_uuid) ^
5737                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5738         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5739         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5740
5741 #ifdef CONFIG_QUOTA
5742         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
5743             sb_has_quota_limits_enabled(sb, PRJQUOTA))
5744                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
5745 #endif
5746         return 0;
5747 }
5748
5749
5750 #ifdef CONFIG_QUOTA
5751
5752 /*
5753  * Helper functions so that transaction is started before we acquire dqio_sem
5754  * to keep correct lock ordering of transaction > dqio_sem
5755  */
5756 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5757 {
5758         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5759 }
5760
5761 static int ext4_write_dquot(struct dquot *dquot)
5762 {
5763         int ret, err;
5764         handle_t *handle;
5765         struct inode *inode;
5766
5767         inode = dquot_to_inode(dquot);
5768         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5769                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5770         if (IS_ERR(handle))
5771                 return PTR_ERR(handle);
5772         ret = dquot_commit(dquot);
5773         err = ext4_journal_stop(handle);
5774         if (!ret)
5775                 ret = err;
5776         return ret;
5777 }
5778
5779 static int ext4_acquire_dquot(struct dquot *dquot)
5780 {
5781         int ret, err;
5782         handle_t *handle;
5783
5784         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5785                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5786         if (IS_ERR(handle))
5787                 return PTR_ERR(handle);
5788         ret = dquot_acquire(dquot);
5789         err = ext4_journal_stop(handle);
5790         if (!ret)
5791                 ret = err;
5792         return ret;
5793 }
5794
5795 static int ext4_release_dquot(struct dquot *dquot)
5796 {
5797         int ret, err;
5798         handle_t *handle;
5799
5800         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5801                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5802         if (IS_ERR(handle)) {
5803                 /* Release dquot anyway to avoid endless cycle in dqput() */
5804                 dquot_release(dquot);
5805                 return PTR_ERR(handle);
5806         }
5807         ret = dquot_release(dquot);
5808         err = ext4_journal_stop(handle);
5809         if (!ret)
5810                 ret = err;
5811         return ret;
5812 }
5813
5814 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5815 {
5816         struct super_block *sb = dquot->dq_sb;
5817         struct ext4_sb_info *sbi = EXT4_SB(sb);
5818
5819         /* Are we journaling quotas? */
5820         if (ext4_has_feature_quota(sb) ||
5821             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5822                 dquot_mark_dquot_dirty(dquot);
5823                 return ext4_write_dquot(dquot);
5824         } else {
5825                 return dquot_mark_dquot_dirty(dquot);
5826         }
5827 }
5828
5829 static int ext4_write_info(struct super_block *sb, int type)
5830 {
5831         int ret, err;
5832         handle_t *handle;
5833
5834         /* Data block + inode block */
5835         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5836         if (IS_ERR(handle))
5837                 return PTR_ERR(handle);
5838         ret = dquot_commit_info(sb, type);
5839         err = ext4_journal_stop(handle);
5840         if (!ret)
5841                 ret = err;
5842         return ret;
5843 }
5844
5845 /*
5846  * Turn on quotas during mount time - we need to find
5847  * the quota file and such...
5848  */
5849 static int ext4_quota_on_mount(struct super_block *sb, int type)
5850 {
5851         return dquot_quota_on_mount(sb, get_qf_name(sb, EXT4_SB(sb), type),
5852                                         EXT4_SB(sb)->s_jquota_fmt, type);
5853 }
5854
5855 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5856 {
5857         struct ext4_inode_info *ei = EXT4_I(inode);
5858
5859         /* The first argument of lockdep_set_subclass has to be
5860          * *exactly* the same as the argument to init_rwsem() --- in
5861          * this case, in init_once() --- or lockdep gets unhappy
5862          * because the name of the lock is set using the
5863          * stringification of the argument to init_rwsem().
5864          */
5865         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5866         lockdep_set_subclass(&ei->i_data_sem, subclass);
5867 }
5868
5869 /*
5870  * Standard function to be called on quota_on
5871  */
5872 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5873                          const struct path *path)
5874 {
5875         int err;
5876
5877         if (!test_opt(sb, QUOTA))
5878                 return -EINVAL;
5879
5880         /* Quotafile not on the same filesystem? */
5881         if (path->dentry->d_sb != sb)
5882                 return -EXDEV;
5883         /* Journaling quota? */
5884         if (EXT4_SB(sb)->s_qf_names[type]) {
5885                 /* Quotafile not in fs root? */
5886                 if (path->dentry->d_parent != sb->s_root)
5887                         ext4_msg(sb, KERN_WARNING,
5888                                 "Quota file not on filesystem root. "
5889                                 "Journaled quota will not work");
5890                 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
5891         } else {
5892                 /*
5893                  * Clear the flag just in case mount options changed since
5894                  * last time.
5895                  */
5896                 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
5897         }
5898
5899         /*
5900          * When we journal data on quota file, we have to flush journal to see
5901          * all updates to the file when we bypass pagecache...
5902          */
5903         if (EXT4_SB(sb)->s_journal &&
5904             ext4_should_journal_data(d_inode(path->dentry))) {
5905                 /*
5906                  * We don't need to lock updates but journal_flush() could
5907                  * otherwise be livelocked...
5908                  */
5909                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5910                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5911                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5912                 if (err)
5913                         return err;
5914         }
5915
5916         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5917         err = dquot_quota_on(sb, type, format_id, path);
5918         if (err) {
5919                 lockdep_set_quota_inode(path->dentry->d_inode,
5920                                              I_DATA_SEM_NORMAL);
5921         } else {
5922                 struct inode *inode = d_inode(path->dentry);
5923                 handle_t *handle;
5924
5925                 /*
5926                  * Set inode flags to prevent userspace from messing with quota
5927                  * files. If this fails, we return success anyway since quotas
5928                  * are already enabled and this is not a hard failure.
5929                  */
5930                 inode_lock(inode);
5931                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5932                 if (IS_ERR(handle))
5933                         goto unlock_inode;
5934                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
5935                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
5936                                 S_NOATIME | S_IMMUTABLE);
5937                 err = ext4_mark_inode_dirty(handle, inode);
5938                 ext4_journal_stop(handle);
5939         unlock_inode:
5940                 inode_unlock(inode);
5941         }
5942         return err;
5943 }
5944
5945 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5946                              unsigned int flags)
5947 {
5948         int err;
5949         struct inode *qf_inode;
5950         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5951                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5952                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5953                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5954         };
5955
5956         BUG_ON(!ext4_has_feature_quota(sb));
5957
5958         if (!qf_inums[type])
5959                 return -EPERM;
5960
5961         qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL);
5962         if (IS_ERR(qf_inode)) {
5963                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5964                 return PTR_ERR(qf_inode);
5965         }
5966
5967         /* Don't account quota for quota files to avoid recursion */
5968         qf_inode->i_flags |= S_NOQUOTA;
5969         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
5970         err = dquot_load_quota_inode(qf_inode, type, format_id, flags);
5971         if (err)
5972                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
5973         iput(qf_inode);
5974
5975         return err;
5976 }
5977
5978 /* Enable usage tracking for all quota types. */
5979 static int ext4_enable_quotas(struct super_block *sb)
5980 {
5981         int type, err = 0;
5982         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5983                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5984                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5985                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5986         };
5987         bool quota_mopt[EXT4_MAXQUOTAS] = {
5988                 test_opt(sb, USRQUOTA),
5989                 test_opt(sb, GRPQUOTA),
5990                 test_opt(sb, PRJQUOTA),
5991         };
5992
5993         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
5994         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5995                 if (qf_inums[type]) {
5996                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5997                                 DQUOT_USAGE_ENABLED |
5998                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
5999                         if (err) {
6000                                 ext4_warning(sb,
6001                                         "Failed to enable quota tracking "
6002                                         "(type=%d, err=%d). Please run "
6003                                         "e2fsck to fix.", type, err);
6004                                 for (type--; type >= 0; type--)
6005                                         dquot_quota_off(sb, type);
6006
6007                                 return err;
6008                         }
6009                 }
6010         }
6011         return 0;
6012 }
6013
6014 static int ext4_quota_off(struct super_block *sb, int type)
6015 {
6016         struct inode *inode = sb_dqopt(sb)->files[type];
6017         handle_t *handle;
6018         int err;
6019
6020         /* Force all delayed allocation blocks to be allocated.
6021          * Caller already holds s_umount sem */
6022         if (test_opt(sb, DELALLOC))
6023                 sync_filesystem(sb);
6024
6025         if (!inode || !igrab(inode))
6026                 goto out;
6027
6028         err = dquot_quota_off(sb, type);
6029         if (err || ext4_has_feature_quota(sb))
6030                 goto out_put;
6031
6032         inode_lock(inode);
6033         /*
6034          * Update modification times of quota files when userspace can
6035          * start looking at them. If we fail, we return success anyway since
6036          * this is not a hard failure and quotas are already disabled.
6037          */
6038         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
6039         if (IS_ERR(handle)) {
6040                 err = PTR_ERR(handle);
6041                 goto out_unlock;
6042         }
6043         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
6044         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
6045         inode->i_mtime = inode->i_ctime = current_time(inode);
6046         err = ext4_mark_inode_dirty(handle, inode);
6047         ext4_journal_stop(handle);
6048 out_unlock:
6049         inode_unlock(inode);
6050 out_put:
6051         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
6052         iput(inode);
6053         return err;
6054 out:
6055         return dquot_quota_off(sb, type);
6056 }
6057
6058 /* Read data from quotafile - avoid pagecache and such because we cannot afford
6059  * acquiring the locks... As quota files are never truncated and quota code
6060  * itself serializes the operations (and no one else should touch the files)
6061  * we don't have to be afraid of races */
6062 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
6063                                size_t len, loff_t off)
6064 {
6065         struct inode *inode = sb_dqopt(sb)->files[type];
6066         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
6067         int offset = off & (sb->s_blocksize - 1);
6068         int tocopy;
6069         size_t toread;
6070         struct buffer_head *bh;
6071         loff_t i_size = i_size_read(inode);
6072
6073         if (off > i_size)
6074                 return 0;
6075         if (off+len > i_size)
6076                 len = i_size-off;
6077         toread = len;
6078         while (toread > 0) {
6079                 tocopy = sb->s_blocksize - offset < toread ?
6080                                 sb->s_blocksize - offset : toread;
6081                 bh = ext4_bread(NULL, inode, blk, 0);
6082                 if (IS_ERR(bh))
6083                         return PTR_ERR(bh);
6084                 if (!bh)        /* A hole? */
6085                         memset(data, 0, tocopy);
6086                 else
6087                         memcpy(data, bh->b_data+offset, tocopy);
6088                 brelse(bh);
6089                 offset = 0;
6090                 toread -= tocopy;
6091                 data += tocopy;
6092                 blk++;
6093         }
6094         return len;
6095 }
6096
6097 /* Write to quotafile (we know the transaction is already started and has
6098  * enough credits) */
6099 static ssize_t ext4_quota_write(struct super_block *sb, int type,
6100                                 const char *data, size_t len, loff_t off)
6101 {
6102         struct inode *inode = sb_dqopt(sb)->files[type];
6103         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
6104         int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1);
6105         int retries = 0;
6106         struct buffer_head *bh;
6107         handle_t *handle = journal_current_handle();
6108
6109         if (EXT4_SB(sb)->s_journal && !handle) {
6110                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
6111                         " cancelled because transaction is not started",
6112                         (unsigned long long)off, (unsigned long long)len);
6113                 return -EIO;
6114         }
6115         /*
6116          * Since we account only one data block in transaction credits,
6117          * then it is impossible to cross a block boundary.
6118          */
6119         if (sb->s_blocksize - offset < len) {
6120                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
6121                         " cancelled because not block aligned",
6122                         (unsigned long long)off, (unsigned long long)len);
6123                 return -EIO;
6124         }
6125
6126         do {
6127                 bh = ext4_bread(handle, inode, blk,
6128                                 EXT4_GET_BLOCKS_CREATE |
6129                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
6130         } while (PTR_ERR(bh) == -ENOSPC &&
6131                  ext4_should_retry_alloc(inode->i_sb, &retries));
6132         if (IS_ERR(bh))
6133                 return PTR_ERR(bh);
6134         if (!bh)
6135                 goto out;
6136         BUFFER_TRACE(bh, "get write access");
6137         err = ext4_journal_get_write_access(handle, bh);
6138         if (err) {
6139                 brelse(bh);
6140                 return err;
6141         }
6142         lock_buffer(bh);
6143         memcpy(bh->b_data+offset, data, len);
6144         flush_dcache_page(bh->b_page);
6145         unlock_buffer(bh);
6146         err = ext4_handle_dirty_metadata(handle, NULL, bh);
6147         brelse(bh);
6148 out:
6149         if (inode->i_size < off + len) {
6150                 i_size_write(inode, off + len);
6151                 EXT4_I(inode)->i_disksize = inode->i_size;
6152                 err2 = ext4_mark_inode_dirty(handle, inode);
6153                 if (unlikely(err2 && !err))
6154                         err = err2;
6155         }
6156         return err ? err : len;
6157 }
6158 #endif
6159
6160 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
6161                        const char *dev_name, void *data)
6162 {
6163         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
6164 }
6165
6166 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
6167 static inline void register_as_ext2(void)
6168 {
6169         int err = register_filesystem(&ext2_fs_type);
6170         if (err)
6171                 printk(KERN_WARNING
6172                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
6173 }
6174
6175 static inline void unregister_as_ext2(void)
6176 {
6177         unregister_filesystem(&ext2_fs_type);
6178 }
6179
6180 static inline int ext2_feature_set_ok(struct super_block *sb)
6181 {
6182         if (ext4_has_unknown_ext2_incompat_features(sb))
6183                 return 0;
6184         if (sb_rdonly(sb))
6185                 return 1;
6186         if (ext4_has_unknown_ext2_ro_compat_features(sb))
6187                 return 0;
6188         return 1;
6189 }
6190 #else
6191 static inline void register_as_ext2(void) { }
6192 static inline void unregister_as_ext2(void) { }
6193 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
6194 #endif
6195
6196 static inline void register_as_ext3(void)
6197 {
6198         int err = register_filesystem(&ext3_fs_type);
6199         if (err)
6200                 printk(KERN_WARNING
6201                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
6202 }
6203
6204 static inline void unregister_as_ext3(void)
6205 {
6206         unregister_filesystem(&ext3_fs_type);
6207 }
6208
6209 static inline int ext3_feature_set_ok(struct super_block *sb)
6210 {
6211         if (ext4_has_unknown_ext3_incompat_features(sb))
6212                 return 0;
6213         if (!ext4_has_feature_journal(sb))
6214                 return 0;
6215         if (sb_rdonly(sb))
6216                 return 1;
6217         if (ext4_has_unknown_ext3_ro_compat_features(sb))
6218                 return 0;
6219         return 1;
6220 }
6221
6222 static struct file_system_type ext4_fs_type = {
6223         .owner          = THIS_MODULE,
6224         .name           = "ext4",
6225         .mount          = ext4_mount,
6226         .kill_sb        = kill_block_super,
6227         .fs_flags       = FS_REQUIRES_DEV,
6228 };
6229 MODULE_ALIAS_FS("ext4");
6230
6231 /* Shared across all ext4 file systems */
6232 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
6233
6234 static int __init ext4_init_fs(void)
6235 {
6236         int i, err;
6237
6238         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
6239         ext4_li_info = NULL;
6240         mutex_init(&ext4_li_mtx);
6241
6242         /* Build-time check for flags consistency */
6243         ext4_check_flag_values();
6244
6245         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
6246                 init_waitqueue_head(&ext4__ioend_wq[i]);
6247
6248         err = ext4_init_es();
6249         if (err)
6250                 return err;
6251
6252         err = ext4_init_pending();
6253         if (err)
6254                 goto out7;
6255
6256         err = ext4_init_post_read_processing();
6257         if (err)
6258                 goto out6;
6259
6260         err = ext4_init_pageio();
6261         if (err)
6262                 goto out5;
6263
6264         err = ext4_init_system_zone();
6265         if (err)
6266                 goto out4;
6267
6268         err = ext4_init_sysfs();
6269         if (err)
6270                 goto out3;
6271
6272         err = ext4_init_mballoc();
6273         if (err)
6274                 goto out2;
6275         err = init_inodecache();
6276         if (err)
6277                 goto out1;
6278         register_as_ext3();
6279         register_as_ext2();
6280         err = register_filesystem(&ext4_fs_type);
6281         if (err)
6282                 goto out;
6283
6284         return 0;
6285 out:
6286         unregister_as_ext2();
6287         unregister_as_ext3();
6288         destroy_inodecache();
6289 out1:
6290         ext4_exit_mballoc();
6291 out2:
6292         ext4_exit_sysfs();
6293 out3:
6294         ext4_exit_system_zone();
6295 out4:
6296         ext4_exit_pageio();
6297 out5:
6298         ext4_exit_post_read_processing();
6299 out6:
6300         ext4_exit_pending();
6301 out7:
6302         ext4_exit_es();
6303
6304         return err;
6305 }
6306
6307 static void __exit ext4_exit_fs(void)
6308 {
6309         ext4_destroy_lazyinit_thread();
6310         unregister_as_ext2();
6311         unregister_as_ext3();
6312         unregister_filesystem(&ext4_fs_type);
6313         destroy_inodecache();
6314         ext4_exit_mballoc();
6315         ext4_exit_sysfs();
6316         ext4_exit_system_zone();
6317         ext4_exit_pageio();
6318         ext4_exit_post_read_processing();
6319         ext4_exit_es();
6320         ext4_exit_pending();
6321 }
6322
6323 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
6324 MODULE_DESCRIPTION("Fourth Extended Filesystem");
6325 MODULE_LICENSE("GPL");
6326 MODULE_SOFTDEP("pre: crc32c");
6327 module_init(ext4_init_fs)
6328 module_exit(ext4_exit_fs)