fs/ext4/super.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/ext4/super.c
   4  *
   5  * Copyright (C) 1992, 1993, 1994, 1995
   6  * Remy Card ([email protected])
   7  * Laboratoire MASI - Institut Blaise Pascal
   8  * Universite Pierre et Marie Curie (Paris VI)
   9  *
  10  *  from
  11  *
  12  *  linux/fs/minix/inode.c
  13  *
  14  *  Copyright (C) 1991, 1992  Linus Torvalds
  15  *
  16  *  Big-endian to little-endian byte-swapping/bitmaps by
  17  *        David S. Miller ([email protected]), 1995
  18  */
  19
  20 #include <linux/module.h>
  21 #include <linux/string.h>
  22 #include <linux/fs.h>
  23 #include <linux/time.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/backing-dev.h>
  29 #include <linux/parser.h>
  30 #include <linux/buffer_head.h>
  31 #include <linux/exportfs.h>
  32 #include <linux/vfs.h>
  33 #include <linux/random.h>
  34 #include <linux/mount.h>
  35 #include <linux/namei.h>
  36 #include <linux/quotaops.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/dax.h>
  42 #include <linux/cleancache.h>
  43 #include <linux/uaccess.h>
  44 #include <linux/iversion.h>
  45 #include <linux/unicode.h>
  46 #include <linux/part_stat.h>
  47 #include <linux/kthread.h>
  48 #include <linux/freezer.h>
  49
  50 #include "ext4.h"
  51 #include "ext4_extents.h"       /* Needed for trace points definition */
  52 #include "ext4_jbd2.h"
  53 #include "xattr.h"
  54 #include "acl.h"
  55 #include "mballoc.h"
  56 #include "fsmap.h"
  57
  58 #define CREATE_TRACE_POINTS
  59 #include <trace/events/ext4.h>
  60
  61 static struct ext4_lazy_init *ext4_li_info;
  62 static struct mutex ext4_li_mtx;
  63 static struct ratelimit_state ext4_mount_msg_ratelimit;
  64
  65 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  66                              unsigned long journal_devnum);
  67 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  68 static int ext4_commit_super(struct super_block *sb, int sync);
  69 static int ext4_mark_recovery_complete(struct super_block *sb,
  70                                         struct ext4_super_block *es);
  71 static int ext4_clear_journal_err(struct super_block *sb,
  72                                   struct ext4_super_block *es);
  73 static int ext4_sync_fs(struct super_block *sb, int wait);
  74 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  75 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  76 static int ext4_unfreeze(struct super_block *sb);
  77 static int ext4_freeze(struct super_block *sb);
  78 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  79                        const char *dev_name, void *data);
  80 static inline int ext2_feature_set_ok(struct super_block *sb);
  81 static inline int ext3_feature_set_ok(struct super_block *sb);
  82 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  83 static void ext4_destroy_lazyinit_thread(void);
  84 static void ext4_unregister_li_request(struct super_block *sb);
  85 static void ext4_clear_request_list(void);
  86 static struct inode *ext4_get_journal_inode(struct super_block *sb,
  87                                             unsigned int journal_inum);
  88
  89 /*
  90  * Lock ordering
  91  *
  92  * Note the difference between i_mmap_sem (EXT4_I(inode)->i_mmap_sem) and
  93  * i_mmap_rwsem (inode->i_mmap_rwsem)!
  94  *
  95  * page fault path:
  96  * mmap_lock -> sb_start_pagefault -> i_mmap_sem (r) -> transaction start ->
  97  *   page lock -> i_data_sem (rw)
  98  *
  99  * buffered write path:
 100  * sb_start_write -> i_mutex -> mmap_lock
 101  * sb_start_write -> i_mutex -> transaction start -> page lock ->
 102  *   i_data_sem (rw)
 103  *
 104  * truncate:
 105  * sb_start_write -> i_mutex -> i_mmap_sem (w) -> i_mmap_rwsem (w) -> page lock
 106  * sb_start_write -> i_mutex -> i_mmap_sem (w) -> transaction start ->
 107  *   i_data_sem (rw)
 108  *
 109  * direct IO:
 110  * sb_start_write -> i_mutex -> mmap_lock
 111  * sb_start_write -> i_mutex -> transaction start -> i_data_sem (rw)
 112  *
 113  * writepages:
 114  * transaction start -> page lock(s) -> i_data_sem (rw)
 115  */
 116
 117 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
 118 static struct file_system_type ext2_fs_type = {
 119         .owner          = THIS_MODULE,
 120         .name           = "ext2",
 121         .mount          = ext4_mount,
 122         .kill_sb        = kill_block_super,
 123         .fs_flags       = FS_REQUIRES_DEV,
 124 };
 125 MODULE_ALIAS_FS("ext2");
 126 MODULE_ALIAS("ext2");
 127 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
 128 #else
 129 #define IS_EXT2_SB(sb) (0)
 130 #endif
 131
 132
 133 static struct file_system_type ext3_fs_type = {
 134         .owner          = THIS_MODULE,
 135         .name           = "ext3",
 136         .mount          = ext4_mount,
 137         .kill_sb        = kill_block_super,
 138         .fs_flags       = FS_REQUIRES_DEV,
 139 };
 140 MODULE_ALIAS_FS("ext3");
 141 MODULE_ALIAS("ext3");
 142 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 143
 144 /*
 145  * This works like sb_bread() except it uses ERR_PTR for error
 146  * returns.  Currently with sb_bread it's impossible to distinguish
 147  * between ENOMEM and EIO situations (since both result in a NULL
 148  * return.
 149  */
 150 struct buffer_head *
 151 ext4_sb_bread(struct super_block *sb, sector_t block, int op_flags)
 152 {
 153         struct buffer_head *bh = sb_getblk(sb, block);
 154
 155         if (bh == NULL)
 156                 return ERR_PTR(-ENOMEM);
 157         if (ext4_buffer_uptodate(bh))
 158                 return bh;
 159         ll_rw_block(REQ_OP_READ, REQ_META | op_flags, 1, &bh);
 160         wait_on_buffer(bh);
 161         if (buffer_uptodate(bh))
 162                 return bh;
 163         put_bh(bh);
 164         return ERR_PTR(-EIO);
 165 }
 166
 167 static int ext4_verify_csum_type(struct super_block *sb,
 168                                  struct ext4_super_block *es)
 169 {
 170         if (!ext4_has_feature_metadata_csum(sb))
 171                 return 1;
 172
 173         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 174 }
 175
 176 static __le32 ext4_superblock_csum(struct super_block *sb,
 177                                    struct ext4_super_block *es)
 178 {
 179         struct ext4_sb_info *sbi = EXT4_SB(sb);
 180         int offset = offsetof(struct ext4_super_block, s_checksum);
 181         __u32 csum;
 182
 183         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 184
 185         return cpu_to_le32(csum);
 186 }
 187
 188 static int ext4_superblock_csum_verify(struct super_block *sb,
 189                                        struct ext4_super_block *es)
 190 {
 191         if (!ext4_has_metadata_csum(sb))
 192                 return 1;
 193
 194         return es->s_checksum == ext4_superblock_csum(sb, es);
 195 }
 196
 197 void ext4_superblock_csum_set(struct super_block *sb)
 198 {
 199         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 200
 201         if (!ext4_has_metadata_csum(sb))
 202                 return;
 203
 204         es->s_checksum = ext4_superblock_csum(sb, es);
 205 }
 206
 207 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 208                                struct ext4_group_desc *bg)
 209 {
 210         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 211                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 212                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 213 }
 214
 215 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 216                                struct ext4_group_desc *bg)
 217 {
 218         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 219                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 220                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 221 }
 222
 223 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 224                               struct ext4_group_desc *bg)
 225 {
 226         return le32_to_cpu(bg->bg_inode_table_lo) |
 227                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 228                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 229 }
 230
 231 __u32 ext4_free_group_clusters(struct super_block *sb,
 232                                struct ext4_group_desc *bg)
 233 {
 234         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 235                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 236                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 237 }
 238
 239 __u32 ext4_free_inodes_count(struct super_block *sb,
 240                               struct ext4_group_desc *bg)
 241 {
 242         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 243                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 244                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 245 }
 246
 247 __u32 ext4_used_dirs_count(struct super_block *sb,
 248                               struct ext4_group_desc *bg)
 249 {
 250         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 251                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 252                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 253 }
 254
 255 __u32 ext4_itable_unused_count(struct super_block *sb,
 256                               struct ext4_group_desc *bg)
 257 {
 258         return le16_to_cpu(bg->bg_itable_unused_lo) |
 259                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 260                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 261 }
 262
 263 void ext4_block_bitmap_set(struct super_block *sb,
 264                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 265 {
 266         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 267         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 268                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 269 }
 270
 271 void ext4_inode_bitmap_set(struct super_block *sb,
 272                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 273 {
 274         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 275         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 276                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 277 }
 278
 279 void ext4_inode_table_set(struct super_block *sb,
 280                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 281 {
 282         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 283         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 284                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 285 }
 286
 287 void ext4_free_group_clusters_set(struct super_block *sb,
 288                                   struct ext4_group_desc *bg, __u32 count)
 289 {
 290         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 291         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 292                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 293 }
 294
 295 void ext4_free_inodes_set(struct super_block *sb,
 296                           struct ext4_group_desc *bg, __u32 count)
 297 {
 298         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 299         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 300                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 301 }
 302
 303 void ext4_used_dirs_set(struct super_block *sb,
 304                           struct ext4_group_desc *bg, __u32 count)
 305 {
 306         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 307         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 308                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 309 }
 310
 311 void ext4_itable_unused_set(struct super_block *sb,
 312                           struct ext4_group_desc *bg, __u32 count)
 313 {
 314         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 315         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 316                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 317 }
 318
 319 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi)
 320 {
 321         time64_t now = ktime_get_real_seconds();
 322
 323         now = clamp_val(now, 0, (1ull << 40) - 1);
 324
 325         *lo = cpu_to_le32(lower_32_bits(now));
 326         *hi = upper_32_bits(now);
 327 }
 328
 329 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi)
 330 {
 331         return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo);
 332 }
 333 #define ext4_update_tstamp(es, tstamp) \
 334         __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
 335 #define ext4_get_tstamp(es, tstamp) \
 336         __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
 337
 338 static void __save_error_info(struct super_block *sb, int error,
 339                               __u32 ino, __u64 block,
 340                               const char *func, unsigned int line)
 341 {
 342         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 343         int err;
 344
 345         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 346         if (bdev_read_only(sb->s_bdev))
 347                 return;
 348         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 349         ext4_update_tstamp(es, s_last_error_time);
 350         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 351         es->s_last_error_line = cpu_to_le32(line);
 352         es->s_last_error_ino = cpu_to_le32(ino);
 353         es->s_last_error_block = cpu_to_le64(block);
 354         switch (error) {
 355         case EIO:
 356                 err = EXT4_ERR_EIO;
 357                 break;
 358         case ENOMEM:
 359                 err = EXT4_ERR_ENOMEM;
 360                 break;
 361         case EFSBADCRC:
 362                 err = EXT4_ERR_EFSBADCRC;
 363                 break;
 364         case 0:
 365         case EFSCORRUPTED:
 366                 err = EXT4_ERR_EFSCORRUPTED;
 367                 break;
 368         case ENOSPC:
 369                 err = EXT4_ERR_ENOSPC;
 370                 break;
 371         case ENOKEY:
 372                 err = EXT4_ERR_ENOKEY;
 373                 break;
 374         case EROFS:
 375                 err = EXT4_ERR_EROFS;
 376                 break;
 377         case EFBIG:
 378                 err = EXT4_ERR_EFBIG;
 379                 break;
 380         case EEXIST:
 381                 err = EXT4_ERR_EEXIST;
 382                 break;
 383         case ERANGE:
 384                 err = EXT4_ERR_ERANGE;
 385                 break;
 386         case EOVERFLOW:
 387                 err = EXT4_ERR_EOVERFLOW;
 388                 break;
 389         case EBUSY:
 390                 err = EXT4_ERR_EBUSY;
 391                 break;
 392         case ENOTDIR:
 393                 err = EXT4_ERR_ENOTDIR;
 394                 break;
 395         case ENOTEMPTY:
 396                 err = EXT4_ERR_ENOTEMPTY;
 397                 break;
 398         case ESHUTDOWN:
 399                 err = EXT4_ERR_ESHUTDOWN;
 400                 break;
 401         case EFAULT:
 402                 err = EXT4_ERR_EFAULT;
 403                 break;
 404         default:
 405                 err = EXT4_ERR_UNKNOWN;
 406         }
 407         es->s_last_error_errcode = err;
 408         if (!es->s_first_error_time) {
 409                 es->s_first_error_time = es->s_last_error_time;
 410                 es->s_first_error_time_hi = es->s_last_error_time_hi;
 411                 strncpy(es->s_first_error_func, func,
 412                         sizeof(es->s_first_error_func));
 413                 es->s_first_error_line = cpu_to_le32(line);
 414                 es->s_first_error_ino = es->s_last_error_ino;
 415                 es->s_first_error_block = es->s_last_error_block;
 416                 es->s_first_error_errcode = es->s_last_error_errcode;
 417         }
 418         /*
 419          * Start the daily error reporting function if it hasn't been
 420          * started already
 421          */
 422         if (!es->s_error_count)
 423                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 424         le32_add_cpu(&es->s_error_count, 1);
 425 }
 426
 427 static void save_error_info(struct super_block *sb, int error,
 428                             __u32 ino, __u64 block,
 429                             const char *func, unsigned int line)
 430 {
 431         __save_error_info(sb, error, ino, block, func, line);
 432         if (!bdev_read_only(sb->s_bdev))
 433                 ext4_commit_super(sb, 1);
 434 }
 435
 436 /*
 437  * The del_gendisk() function uninitializes the disk-specific data
 438  * structures, including the bdi structure, without telling anyone
 439  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 440  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 441  * This is a kludge to prevent these oops until we can put in a proper
 442  * hook in del_gendisk() to inform the VFS and file system layers.
 443  */
 444 static int block_device_ejected(struct super_block *sb)
 445 {
 446         struct inode *bd_inode = sb->s_bdev->bd_inode;
 447         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 448
 449         return bdi->dev == NULL;
 450 }
 451
 452 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 453 {
 454         struct super_block              *sb = journal->j_private;
 455         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 456         int                             error = is_journal_aborted(journal);
 457         struct ext4_journal_cb_entry    *jce;
 458
 459         BUG_ON(txn->t_state == T_FINISHED);
 460
 461         ext4_process_freed_data(sb, txn->t_tid);
 462
 463         spin_lock(&sbi->s_md_lock);
 464         while (!list_empty(&txn->t_private_list)) {
 465                 jce = list_entry(txn->t_private_list.next,
 466                                  struct ext4_journal_cb_entry, jce_list);
 467                 list_del_init(&jce->jce_list);
 468                 spin_unlock(&sbi->s_md_lock);
 469                 jce->jce_func(sb, jce, error);
 470                 spin_lock(&sbi->s_md_lock);
 471         }
 472         spin_unlock(&sbi->s_md_lock);
 473 }
 474
 475 static bool system_going_down(void)
 476 {
 477         return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF
 478                 || system_state == SYSTEM_RESTART;
 479 }
 480
 481 /* Deal with the reporting of failure conditions on a filesystem such as
 482  * inconsistencies detected or read IO failures.
 483  *
 484  * On ext2, we can store the error state of the filesystem in the
 485  * superblock.  That is not possible on ext4, because we may have other
 486  * write ordering constraints on the superblock which prevent us from
 487  * writing it out straight away; and given that the journal is about to
 488  * be aborted, we can't rely on the current, or future, transactions to
 489  * write out the superblock safely.
 490  *
 491  * We'll just use the jbd2_journal_abort() error code to record an error in
 492  * the journal instead.  On recovery, the journal will complain about
 493  * that error until we've noted it down and cleared it.
 494  */
 495
 496 static void ext4_handle_error(struct super_block *sb)
 497 {
 498         if (test_opt(sb, WARN_ON_ERROR))
 499                 WARN_ON_ONCE(1);
 500
 501         if (sb_rdonly(sb))
 502                 return;
 503
 504         if (!test_opt(sb, ERRORS_CONT)) {
 505                 journal_t *journal = EXT4_SB(sb)->s_journal;
 506
 507                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 508                 if (journal)
 509                         jbd2_journal_abort(journal, -EIO);
 510         }
 511         /*
 512          * We force ERRORS_RO behavior when system is rebooting. Otherwise we
 513          * could panic during 'reboot -f' as the underlying device got already
 514          * disabled.
 515          */
 516         if (test_opt(sb, ERRORS_RO) || system_going_down()) {
 517                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 518                 /*
 519                  * Make sure updated value of ->s_mount_flags will be visible
 520                  * before ->s_flags update
 521                  */
 522                 smp_wmb();
 523                 sb->s_flags |= SB_RDONLY;
 524         } else if (test_opt(sb, ERRORS_PANIC)) {
 525                 panic("EXT4-fs (device %s): panic forced after error\n",
 526                         sb->s_id);
 527         }
 528 }
 529
 530 #define ext4_error_ratelimit(sb)                                        \
 531                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 532                              "EXT4-fs error")
 533
 534 void __ext4_error(struct super_block *sb, const char *function,
 535                   unsigned int line, int error, __u64 block,
 536                   const char *fmt, ...)
 537 {
 538         struct va_format vaf;
 539         va_list args;
 540
 541         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 542                 return;
 543
 544         trace_ext4_error(sb, function, line);
 545         if (ext4_error_ratelimit(sb)) {
 546                 va_start(args, fmt);
 547                 vaf.fmt = fmt;
 548                 vaf.va = &args;
 549                 printk(KERN_CRIT
 550                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 551                        sb->s_id, function, line, current->comm, &vaf);
 552                 va_end(args);
 553         }
 554         save_error_info(sb, error, 0, block, function, line);
 555         ext4_handle_error(sb);
 556 }
 557
 558 void __ext4_error_inode(struct inode *inode, const char *function,
 559                         unsigned int line, ext4_fsblk_t block, int error,
 560                         const char *fmt, ...)
 561 {
 562         va_list args;
 563         struct va_format vaf;
 564
 565         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 566                 return;
 567
 568         trace_ext4_error(inode->i_sb, function, line);
 569         if (ext4_error_ratelimit(inode->i_sb)) {
 570                 va_start(args, fmt);
 571                 vaf.fmt = fmt;
 572                 vaf.va = &args;
 573                 if (block)
 574                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 575                                "inode #%lu: block %llu: comm %s: %pV\n",
 576                                inode->i_sb->s_id, function, line, inode->i_ino,
 577                                block, current->comm, &vaf);
 578                 else
 579                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 580                                "inode #%lu: comm %s: %pV\n",
 581                                inode->i_sb->s_id, function, line, inode->i_ino,
 582                                current->comm, &vaf);
 583                 va_end(args);
 584         }
 585         save_error_info(inode->i_sb, error, inode->i_ino, block,
 586                         function, line);
 587         ext4_handle_error(inode->i_sb);
 588 }
 589
 590 void __ext4_error_file(struct file *file, const char *function,
 591                        unsigned int line, ext4_fsblk_t block,
 592                        const char *fmt, ...)
 593 {
 594         va_list args;
 595         struct va_format vaf;
 596         struct inode *inode = file_inode(file);
 597         char pathname[80], *path;
 598
 599         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 600                 return;
 601
 602         trace_ext4_error(inode->i_sb, function, line);
 603         if (ext4_error_ratelimit(inode->i_sb)) {
 604                 path = file_path(file, pathname, sizeof(pathname));
 605                 if (IS_ERR(path))
 606                         path = "(unknown)";
 607                 va_start(args, fmt);
 608                 vaf.fmt = fmt;
 609                 vaf.va = &args;
 610                 if (block)
 611                         printk(KERN_CRIT
 612                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 613                                "block %llu: comm %s: path %s: %pV\n",
 614                                inode->i_sb->s_id, function, line, inode->i_ino,
 615                                block, current->comm, path, &vaf);
 616                 else
 617                         printk(KERN_CRIT
 618                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 619                                "comm %s: path %s: %pV\n",
 620                                inode->i_sb->s_id, function, line, inode->i_ino,
 621                                current->comm, path, &vaf);
 622                 va_end(args);
 623         }
 624         save_error_info(inode->i_sb, EFSCORRUPTED, inode->i_ino, block,
 625                         function, line);
 626         ext4_handle_error(inode->i_sb);
 627 }
 628
 629 const char *ext4_decode_error(struct super_block *sb, int errno,
 630                               char nbuf[16])
 631 {
 632         char *errstr = NULL;
 633
 634         switch (errno) {
 635         case -EFSCORRUPTED:
 636                 errstr = "Corrupt filesystem";
 637                 break;
 638         case -EFSBADCRC:
 639                 errstr = "Filesystem failed CRC";
 640                 break;
 641         case -EIO:
 642                 errstr = "IO failure";
 643                 break;
 644         case -ENOMEM:
 645                 errstr = "Out of memory";
 646                 break;
 647         case -EROFS:
 648                 if (!sb || (EXT4_SB(sb)->s_journal &&
 649                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 650                         errstr = "Journal has aborted";
 651                 else
 652                         errstr = "Readonly filesystem";
 653                 break;
 654         default:
 655                 /* If the caller passed in an extra buffer for unknown
 656                  * errors, textualise them now.  Else we just return
 657                  * NULL. */
 658                 if (nbuf) {
 659                         /* Check for truncated error codes... */
 660                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 661                                 errstr = nbuf;
 662                 }
 663                 break;
 664         }
 665
 666         return errstr;
 667 }
 668
 669 /* __ext4_std_error decodes expected errors from journaling functions
 670  * automatically and invokes the appropriate error response.  */
 671
 672 void __ext4_std_error(struct super_block *sb, const char *function,
 673                       unsigned int line, int errno)
 674 {
 675         char nbuf[16];
 676         const char *errstr;
 677
 678         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 679                 return;
 680
 681         /* Special case: if the error is EROFS, and we're not already
 682          * inside a transaction, then there's really no point in logging
 683          * an error. */
 684         if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb))
 685                 return;
 686
 687         if (ext4_error_ratelimit(sb)) {
 688                 errstr = ext4_decode_error(sb, errno, nbuf);
 689                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 690                        sb->s_id, function, line, errstr);
 691         }
 692
 693         save_error_info(sb, -errno, 0, 0, function, line);
 694         ext4_handle_error(sb);
 695 }
 696
 697 /*
 698  * ext4_abort is a much stronger failure handler than ext4_error.  The
 699  * abort function may be used to deal with unrecoverable failures such
 700  * as journal IO errors or ENOMEM at a critical moment in log management.
 701  *
 702  * We unconditionally force the filesystem into an ABORT|READONLY state,
 703  * unless the error response on the fs has been set to panic in which
 704  * case we take the easy way out and panic immediately.
 705  */
 706
 707 void __ext4_abort(struct super_block *sb, const char *function,
 708                   unsigned int line, int error, const char *fmt, ...)
 709 {
 710         struct va_format vaf;
 711         va_list args;
 712
 713         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 714                 return;
 715
 716         save_error_info(sb, error, 0, 0, function, line);
 717         va_start(args, fmt);
 718         vaf.fmt = fmt;
 719         vaf.va = &args;
 720         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: %pV\n",
 721                sb->s_id, function, line, &vaf);
 722         va_end(args);
 723
 724         if (sb_rdonly(sb) == 0) {
 725                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 726                 if (EXT4_SB(sb)->s_journal)
 727                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 728
 729                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 730                 /*
 731                  * Make sure updated value of ->s_mount_flags will be visible
 732                  * before ->s_flags update
 733                  */
 734                 smp_wmb();
 735                 sb->s_flags |= SB_RDONLY;
 736         }
 737         if (test_opt(sb, ERRORS_PANIC) && !system_going_down())
 738                 panic("EXT4-fs panic from previous error\n");
 739 }
 740
 741 void __ext4_msg(struct super_block *sb,
 742                 const char *prefix, const char *fmt, ...)
 743 {
 744         struct va_format vaf;
 745         va_list args;
 746
 747         atomic_inc(&EXT4_SB(sb)->s_msg_count);
 748         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 749                 return;
 750
 751         va_start(args, fmt);
 752         vaf.fmt = fmt;
 753         vaf.va = &args;
 754         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 755         va_end(args);
 756 }
 757
 758 static int ext4_warning_ratelimit(struct super_block *sb)
 759 {
 760         atomic_inc(&EXT4_SB(sb)->s_warning_count);
 761         return ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state),
 762                             "EXT4-fs warning");
 763 }
 764
 765 void __ext4_warning(struct super_block *sb, const char *function,
 766                     unsigned int line, const char *fmt, ...)
 767 {
 768         struct va_format vaf;
 769         va_list args;
 770
 771         if (!ext4_warning_ratelimit(sb))
 772                 return;
 773
 774         va_start(args, fmt);
 775         vaf.fmt = fmt;
 776         vaf.va = &args;
 777         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 778                sb->s_id, function, line, &vaf);
 779         va_end(args);
 780 }
 781
 782 void __ext4_warning_inode(const struct inode *inode, const char *function,
 783                           unsigned int line, const char *fmt, ...)
 784 {
 785         struct va_format vaf;
 786         va_list args;
 787
 788         if (!ext4_warning_ratelimit(inode->i_sb))
 789                 return;
 790
 791         va_start(args, fmt);
 792         vaf.fmt = fmt;
 793         vaf.va = &args;
 794         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 795                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 796                function, line, inode->i_ino, current->comm, &vaf);
 797         va_end(args);
 798 }
 799
 800 void __ext4_grp_locked_error(const char *function, unsigned int line,
 801                              struct super_block *sb, ext4_group_t grp,
 802                              unsigned long ino, ext4_fsblk_t block,
 803                              const char *fmt, ...)
 804 __releases(bitlock)
 805 __acquires(bitlock)
 806 {
 807         struct va_format vaf;
 808         va_list args;
 809
 810         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 811                 return;
 812
 813         trace_ext4_error(sb, function, line);
 814         __save_error_info(sb, EFSCORRUPTED, ino, block, function, line);
 815
 816         if (ext4_error_ratelimit(sb)) {
 817                 va_start(args, fmt);
 818                 vaf.fmt = fmt;
 819                 vaf.va = &args;
 820                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 821                        sb->s_id, function, line, grp);
 822                 if (ino)
 823                         printk(KERN_CONT "inode %lu: ", ino);
 824                 if (block)
 825                         printk(KERN_CONT "block %llu:",
 826                                (unsigned long long) block);
 827                 printk(KERN_CONT "%pV\n", &vaf);
 828                 va_end(args);
 829         }
 830
 831         if (test_opt(sb, WARN_ON_ERROR))
 832                 WARN_ON_ONCE(1);
 833
 834         if (test_opt(sb, ERRORS_CONT)) {
 835                 ext4_commit_super(sb, 0);
 836                 return;
 837         }
 838
 839         ext4_unlock_group(sb, grp);
 840         ext4_commit_super(sb, 1);
 841         ext4_handle_error(sb);
 842         /*
 843          * We only get here in the ERRORS_RO case; relocking the group
 844          * may be dangerous, but nothing bad will happen since the
 845          * filesystem will have already been marked read/only and the
 846          * journal has been aborted.  We return 1 as a hint to callers
 847          * who might what to use the return value from
 848          * ext4_grp_locked_error() to distinguish between the
 849          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 850          * aggressively from the ext4 function in question, with a
 851          * more appropriate error code.
 852          */
 853         ext4_lock_group(sb, grp);
 854         return;
 855 }
 856
 857 void ext4_mark_group_bitmap_corrupted(struct super_block *sb,
 858                                      ext4_group_t group,
 859                                      unsigned int flags)
 860 {
 861         struct ext4_sb_info *sbi = EXT4_SB(sb);
 862         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
 863         struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
 864         int ret;
 865
 866         if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) {
 867                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT,
 868                                             &grp->bb_state);
 869                 if (!ret)
 870                         percpu_counter_sub(&sbi->s_freeclusters_counter,
 871                                            grp->bb_free);
 872         }
 873
 874         if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) {
 875                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT,
 876                                             &grp->bb_state);
 877                 if (!ret && gdp) {
 878                         int count;
 879
 880                         count = ext4_free_inodes_count(sb, gdp);
 881                         percpu_counter_sub(&sbi->s_freeinodes_counter,
 882                                            count);
 883                 }
 884         }
 885 }
 886
 887 void ext4_update_dynamic_rev(struct super_block *sb)
 888 {
 889         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 890
 891         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 892                 return;
 893
 894         ext4_warning(sb,
 895                      "updating to rev %d because of new feature flag, "
 896                      "running e2fsck is recommended",
 897                      EXT4_DYNAMIC_REV);
 898
 899         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 900         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 901         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 902         /* leave es->s_feature_*compat flags alone */
 903         /* es->s_uuid will be set by e2fsck if empty */
 904
 905         /*
 906          * The rest of the superblock fields should be zero, and if not it
 907          * means they are likely already in use, so leave them alone.  We
 908          * can leave it up to e2fsck to clean up any inconsistencies there.
 909          */
 910 }
 911
 912 /*
 913  * Open the external journal device
 914  */
 915 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 916 {
 917         struct block_device *bdev;
 918
 919         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 920         if (IS_ERR(bdev))
 921                 goto fail;
 922         return bdev;
 923
 924 fail:
 925         ext4_msg(sb, KERN_ERR,
 926                  "failed to open journal device unknown-block(%u,%u) %ld",
 927                  MAJOR(dev), MINOR(dev), PTR_ERR(bdev));
 928         return NULL;
 929 }
 930
 931 /*
 932  * Release the journal device
 933  */
 934 static void ext4_blkdev_put(struct block_device *bdev)
 935 {
 936         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 937 }
 938
 939 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 940 {
 941         struct block_device *bdev;
 942         bdev = sbi->journal_bdev;
 943         if (bdev) {
 944                 ext4_blkdev_put(bdev);
 945                 sbi->journal_bdev = NULL;
 946         }
 947 }
 948
 949 static inline struct inode *orphan_list_entry(struct list_head *l)
 950 {
 951         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 952 }
 953
 954 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 955 {
 956         struct list_head *l;
 957
 958         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 959                  le32_to_cpu(sbi->s_es->s_last_orphan));
 960
 961         printk(KERN_ERR "sb_info orphan list:\n");
 962         list_for_each(l, &sbi->s_orphan) {
 963                 struct inode *inode = orphan_list_entry(l);
 964                 printk(KERN_ERR "  "
 965                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 966                        inode->i_sb->s_id, inode->i_ino, inode,
 967                        inode->i_mode, inode->i_nlink,
 968                        NEXT_ORPHAN(inode));
 969         }
 970 }
 971
 972 #ifdef CONFIG_QUOTA
 973 static int ext4_quota_off(struct super_block *sb, int type);
 974
 975 static inline void ext4_quota_off_umount(struct super_block *sb)
 976 {
 977         int type;
 978
 979         /* Use our quota_off function to clear inode flags etc. */
 980         for (type = 0; type < EXT4_MAXQUOTAS; type++)
 981                 ext4_quota_off(sb, type);
 982 }
 983
 984 /*
 985  * This is a helper function which is used in the mount/remount
 986  * codepaths (which holds s_umount) to fetch the quota file name.
 987  */
 988 static inline char *get_qf_name(struct super_block *sb,
 989                                 struct ext4_sb_info *sbi,
 990                                 int type)
 991 {
 992         return rcu_dereference_protected(sbi->s_qf_names[type],
 993                                          lockdep_is_held(&sb->s_umount));
 994 }
 995 #else
 996 static inline void ext4_quota_off_umount(struct super_block *sb)
 997 {
 998 }
 999 #endif
1000
1001 static void ext4_put_super(struct super_block *sb)
1002 {
1003         struct ext4_sb_info *sbi = EXT4_SB(sb);
1004         struct ext4_super_block *es = sbi->s_es;
1005         struct buffer_head **group_desc;
1006         struct flex_groups **flex_groups;
1007         int aborted = 0;
1008         int i, err;
1009
1010         ext4_unregister_li_request(sb);
1011         ext4_quota_off_umount(sb);
1012
1013         destroy_workqueue(sbi->rsv_conversion_wq);
1014
1015         /*
1016          * Unregister sysfs before destroying jbd2 journal.
1017          * Since we could still access attr_journal_task attribute via sysfs
1018          * path which could have sbi->s_journal->j_task as NULL
1019          */
1020         ext4_unregister_sysfs(sb);
1021
1022         if (sbi->s_journal) {
1023                 aborted = is_journal_aborted(sbi->s_journal);
1024                 err = jbd2_journal_destroy(sbi->s_journal);
1025                 sbi->s_journal = NULL;
1026                 if ((err < 0) && !aborted) {
1027                         ext4_abort(sb, -err, "Couldn't clean up the journal");
1028                 }
1029         }
1030
1031         ext4_es_unregister_shrinker(sbi);
1032         del_timer_sync(&sbi->s_err_report);
1033         ext4_release_system_zone(sb);
1034         ext4_mb_release(sb);
1035         ext4_ext_release(sb);
1036
1037         if (!sb_rdonly(sb) && !aborted) {
1038                 ext4_clear_feature_journal_needs_recovery(sb);
1039                 es->s_state = cpu_to_le16(sbi->s_mount_state);
1040         }
1041         if (!sb_rdonly(sb))
1042                 ext4_commit_super(sb, 1);
1043
1044         rcu_read_lock();
1045         group_desc = rcu_dereference(sbi->s_group_desc);
1046         for (i = 0; i < sbi->s_gdb_count; i++)
1047                 brelse(group_desc[i]);
1048         kvfree(group_desc);
1049         flex_groups = rcu_dereference(sbi->s_flex_groups);
1050         if (flex_groups) {
1051                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
1052                         kvfree(flex_groups[i]);
1053                 kvfree(flex_groups);
1054         }
1055         rcu_read_unlock();
1056         percpu_counter_destroy(&sbi->s_freeclusters_counter);
1057         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1058         percpu_counter_destroy(&sbi->s_dirs_counter);
1059         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
1060         percpu_free_rwsem(&sbi->s_writepages_rwsem);
1061 #ifdef CONFIG_QUOTA
1062         for (i = 0; i < EXT4_MAXQUOTAS; i++)
1063                 kfree(get_qf_name(sb, sbi, i));
1064 #endif
1065
1066         /* Debugging code just in case the in-memory inode orphan list
1067          * isn't empty.  The on-disk one can be non-empty if we've
1068          * detected an error and taken the fs readonly, but the
1069          * in-memory list had better be clean by this point. */
1070         if (!list_empty(&sbi->s_orphan))
1071                 dump_orphan_list(sb, sbi);
1072         J_ASSERT(list_empty(&sbi->s_orphan));
1073
1074         sync_blockdev(sb->s_bdev);
1075         invalidate_bdev(sb->s_bdev);
1076         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
1077                 /*
1078                  * Invalidate the journal device's buffers.  We don't want them
1079                  * floating about in memory - the physical journal device may
1080                  * hotswapped, and it breaks the `ro-after' testing code.
1081                  */
1082                 sync_blockdev(sbi->journal_bdev);
1083                 invalidate_bdev(sbi->journal_bdev);
1084                 ext4_blkdev_remove(sbi);
1085         }
1086
1087         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
1088         sbi->s_ea_inode_cache = NULL;
1089
1090         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
1091         sbi->s_ea_block_cache = NULL;
1092
1093         if (sbi->s_mmp_tsk)
1094                 kthread_stop(sbi->s_mmp_tsk);
1095         brelse(sbi->s_sbh);
1096         sb->s_fs_info = NULL;
1097         /*
1098          * Now that we are completely done shutting down the
1099          * superblock, we need to actually destroy the kobject.
1100          */
1101         kobject_put(&sbi->s_kobj);
1102         wait_for_completion(&sbi->s_kobj_unregister);
1103         if (sbi->s_chksum_driver)
1104                 crypto_free_shash(sbi->s_chksum_driver);
1105         kfree(sbi->s_blockgroup_lock);
1106         fs_put_dax(sbi->s_daxdev);
1107         fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy);
1108 #ifdef CONFIG_UNICODE
1109         utf8_unload(sbi->s_encoding);
1110 #endif
1111         kfree(sbi);
1112 }
1113
1114 static struct kmem_cache *ext4_inode_cachep;
1115
1116 /*
1117  * Called inside transaction, so use GFP_NOFS
1118  */
1119 static struct inode *ext4_alloc_inode(struct super_block *sb)
1120 {
1121         struct ext4_inode_info *ei;
1122
1123         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
1124         if (!ei)
1125                 return NULL;
1126
1127         inode_set_iversion(&ei->vfs_inode, 1);
1128         spin_lock_init(&ei->i_raw_lock);
1129         INIT_LIST_HEAD(&ei->i_prealloc_list);
1130         atomic_set(&ei->i_prealloc_active, 0);
1131         spin_lock_init(&ei->i_prealloc_lock);
1132         ext4_es_init_tree(&ei->i_es_tree);
1133         rwlock_init(&ei->i_es_lock);
1134         INIT_LIST_HEAD(&ei->i_es_list);
1135         ei->i_es_all_nr = 0;
1136         ei->i_es_shk_nr = 0;
1137         ei->i_es_shrink_lblk = 0;
1138         ei->i_reserved_data_blocks = 0;
1139         spin_lock_init(&(ei->i_block_reservation_lock));
1140         ext4_init_pending_tree(&ei->i_pending_tree);
1141 #ifdef CONFIG_QUOTA
1142         ei->i_reserved_quota = 0;
1143         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
1144 #endif
1145         ei->jinode = NULL;
1146         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
1147         spin_lock_init(&ei->i_completed_io_lock);
1148         ei->i_sync_tid = 0;
1149         ei->i_datasync_tid = 0;
1150         atomic_set(&ei->i_unwritten, 0);
1151         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
1152         return &ei->vfs_inode;
1153 }
1154
1155 static int ext4_drop_inode(struct inode *inode)
1156 {
1157         int drop = generic_drop_inode(inode);
1158
1159         if (!drop)
1160                 drop = fscrypt_drop_inode(inode);
1161
1162         trace_ext4_drop_inode(inode, drop);
1163         return drop;
1164 }
1165
1166 static void ext4_free_in_core_inode(struct inode *inode)
1167 {
1168         fscrypt_free_inode(inode);
1169         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1170 }
1171
1172 static void ext4_destroy_inode(struct inode *inode)
1173 {
1174         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1175                 ext4_msg(inode->i_sb, KERN_ERR,
1176                          "Inode %lu (%p): orphan list check failed!",
1177                          inode->i_ino, EXT4_I(inode));
1178                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1179                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1180                                 true);
1181                 dump_stack();
1182         }
1183 }
1184
1185 static void init_once(void *foo)
1186 {
1187         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1188
1189         INIT_LIST_HEAD(&ei->i_orphan);
1190         init_rwsem(&ei->xattr_sem);
1191         init_rwsem(&ei->i_data_sem);
1192         init_rwsem(&ei->i_mmap_sem);
1193         inode_init_once(&ei->vfs_inode);
1194 }
1195
1196 static int __init init_inodecache(void)
1197 {
1198         ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache",
1199                                 sizeof(struct ext4_inode_info), 0,
1200                                 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
1201                                         SLAB_ACCOUNT),
1202                                 offsetof(struct ext4_inode_info, i_data),
1203                                 sizeof_field(struct ext4_inode_info, i_data),
1204                                 init_once);
1205         if (ext4_inode_cachep == NULL)
1206                 return -ENOMEM;
1207         return 0;
1208 }
1209
1210 static void destroy_inodecache(void)
1211 {
1212         /*
1213          * Make sure all delayed rcu free inodes are flushed before we
1214          * destroy cache.
1215          */
1216         rcu_barrier();
1217         kmem_cache_destroy(ext4_inode_cachep);
1218 }
1219
1220 void ext4_clear_inode(struct inode *inode)
1221 {
1222         invalidate_inode_buffers(inode);
1223         clear_inode(inode);
1224         ext4_discard_preallocations(inode, 0);
1225         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1226         dquot_drop(inode);
1227         if (EXT4_I(inode)->jinode) {
1228                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1229                                                EXT4_I(inode)->jinode);
1230                 jbd2_free_inode(EXT4_I(inode)->jinode);
1231                 EXT4_I(inode)->jinode = NULL;
1232         }
1233         fscrypt_put_encryption_info(inode);
1234         fsverity_cleanup_inode(inode);
1235 }
1236
1237 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1238                                         u64 ino, u32 generation)
1239 {
1240         struct inode *inode;
1241
1242         /*
1243          * Currently we don't know the generation for parent directory, so
1244          * a generation of 0 means "accept any"
1245          */
1246         inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE);
1247         if (IS_ERR(inode))
1248                 return ERR_CAST(inode);
1249         if (generation && inode->i_generation != generation) {
1250                 iput(inode);
1251                 return ERR_PTR(-ESTALE);
1252         }
1253
1254         return inode;
1255 }
1256
1257 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1258                                         int fh_len, int fh_type)
1259 {
1260         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1261                                     ext4_nfs_get_inode);
1262 }
1263
1264 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1265                                         int fh_len, int fh_type)
1266 {
1267         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1268                                     ext4_nfs_get_inode);
1269 }
1270
1271 static int ext4_nfs_commit_metadata(struct inode *inode)
1272 {
1273         struct writeback_control wbc = {
1274                 .sync_mode = WB_SYNC_ALL
1275         };
1276
1277         trace_ext4_nfs_commit_metadata(inode);
1278         return ext4_write_inode(inode, &wbc);
1279 }
1280
1281 /*
1282  * Try to release metadata pages (indirect blocks, directories) which are
1283  * mapped via the block device.  Since these pages could have journal heads
1284  * which would prevent try_to_free_buffers() from freeing them, we must use
1285  * jbd2 layer's try_to_free_buffers() function to release them.
1286  */
1287 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1288                                  gfp_t wait)
1289 {
1290         journal_t *journal = EXT4_SB(sb)->s_journal;
1291
1292         WARN_ON(PageChecked(page));
1293         if (!page_has_buffers(page))
1294                 return 0;
1295         if (journal)
1296                 return jbd2_journal_try_to_free_buffers(journal, page);
1297
1298         return try_to_free_buffers(page);
1299 }
1300
1301 #ifdef CONFIG_FS_ENCRYPTION
1302 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1303 {
1304         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1305                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1306 }
1307
1308 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1309                                                         void *fs_data)
1310 {
1311         handle_t *handle = fs_data;
1312         int res, res2, credits, retries = 0;
1313
1314         /*
1315          * Encrypting the root directory is not allowed because e2fsck expects
1316          * lost+found to exist and be unencrypted, and encrypting the root
1317          * directory would imply encrypting the lost+found directory as well as
1318          * the filename "lost+found" itself.
1319          */
1320         if (inode->i_ino == EXT4_ROOT_INO)
1321                 return -EPERM;
1322
1323         if (WARN_ON_ONCE(IS_DAX(inode) && i_size_read(inode)))
1324                 return -EINVAL;
1325
1326         if (ext4_test_inode_flag(inode, EXT4_INODE_DAX))
1327                 return -EOPNOTSUPP;
1328
1329         res = ext4_convert_inline_data(inode);
1330         if (res)
1331                 return res;
1332
1333         /*
1334          * If a journal handle was specified, then the encryption context is
1335          * being set on a new inode via inheritance and is part of a larger
1336          * transaction to create the inode.  Otherwise the encryption context is
1337          * being set on an existing inode in its own transaction.  Only in the
1338          * latter case should the "retry on ENOSPC" logic be used.
1339          */
1340
1341         if (handle) {
1342                 res = ext4_xattr_set_handle(handle, inode,
1343                                             EXT4_XATTR_INDEX_ENCRYPTION,
1344                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1345                                             ctx, len, 0);
1346                 if (!res) {
1347                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1348                         ext4_clear_inode_state(inode,
1349                                         EXT4_STATE_MAY_INLINE_DATA);
1350                         /*
1351                          * Update inode->i_flags - S_ENCRYPTED will be enabled,
1352                          * S_DAX may be disabled
1353                          */
1354                         ext4_set_inode_flags(inode, false);
1355                 }
1356                 return res;
1357         }
1358
1359         res = dquot_initialize(inode);
1360         if (res)
1361                 return res;
1362 retry:
1363         res = ext4_xattr_set_credits(inode, len, false /* is_create */,
1364                                      &credits);
1365         if (res)
1366                 return res;
1367
1368         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
1369         if (IS_ERR(handle))
1370                 return PTR_ERR(handle);
1371
1372         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1373                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1374                                     ctx, len, 0);
1375         if (!res) {
1376                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1377                 /*
1378                  * Update inode->i_flags - S_ENCRYPTED will be enabled,
1379                  * S_DAX may be disabled
1380                  */
1381                 ext4_set_inode_flags(inode, false);
1382                 res = ext4_mark_inode_dirty(handle, inode);
1383                 if (res)
1384                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1385         }
1386         res2 = ext4_journal_stop(handle);
1387
1388         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1389                 goto retry;
1390         if (!res)
1391                 res = res2;
1392         return res;
1393 }
1394
1395 static const union fscrypt_policy *ext4_get_dummy_policy(struct super_block *sb)
1396 {
1397         return EXT4_SB(sb)->s_dummy_enc_policy.policy;
1398 }
1399
1400 static bool ext4_has_stable_inodes(struct super_block *sb)
1401 {
1402         return ext4_has_feature_stable_inodes(sb);
1403 }
1404
1405 static void ext4_get_ino_and_lblk_bits(struct super_block *sb,
1406                                        int *ino_bits_ret, int *lblk_bits_ret)
1407 {
1408         *ino_bits_ret = 8 * sizeof(EXT4_SB(sb)->s_es->s_inodes_count);
1409         *lblk_bits_ret = 8 * sizeof(ext4_lblk_t);
1410 }
1411
1412 static const struct fscrypt_operations ext4_cryptops = {
1413         .key_prefix             = "ext4:",
1414         .get_context            = ext4_get_context,
1415         .set_context            = ext4_set_context,
1416         .get_dummy_policy       = ext4_get_dummy_policy,
1417         .empty_dir              = ext4_empty_dir,
1418         .max_namelen            = EXT4_NAME_LEN,
1419         .has_stable_inodes      = ext4_has_stable_inodes,
1420         .get_ino_and_lblk_bits  = ext4_get_ino_and_lblk_bits,
1421 };
1422 #endif
1423
1424 #ifdef CONFIG_QUOTA
1425 static const char * const quotatypes[] = INITQFNAMES;
1426 #define QTYPE2NAME(t) (quotatypes[t])
1427
1428 static int ext4_write_dquot(struct dquot *dquot);
1429 static int ext4_acquire_dquot(struct dquot *dquot);
1430 static int ext4_release_dquot(struct dquot *dquot);
1431 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1432 static int ext4_write_info(struct super_block *sb, int type);
1433 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1434                          const struct path *path);
1435 static int ext4_quota_on_mount(struct super_block *sb, int type);
1436 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1437                                size_t len, loff_t off);
1438 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1439                                 const char *data, size_t len, loff_t off);
1440 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1441                              unsigned int flags);
1442 static int ext4_enable_quotas(struct super_block *sb);
1443
1444 static struct dquot **ext4_get_dquots(struct inode *inode)
1445 {
1446         return EXT4_I(inode)->i_dquot;
1447 }
1448
1449 static const struct dquot_operations ext4_quota_operations = {
1450         .get_reserved_space     = ext4_get_reserved_space,
1451         .write_dquot            = ext4_write_dquot,
1452         .acquire_dquot          = ext4_acquire_dquot,
1453         .release_dquot          = ext4_release_dquot,
1454         .mark_dirty             = ext4_mark_dquot_dirty,
1455         .write_info             = ext4_write_info,
1456         .alloc_dquot            = dquot_alloc,
1457         .destroy_dquot          = dquot_destroy,
1458         .get_projid             = ext4_get_projid,
1459         .get_inode_usage        = ext4_get_inode_usage,
1460         .get_next_id            = dquot_get_next_id,
1461 };
1462
1463 static const struct quotactl_ops ext4_qctl_operations = {
1464         .quota_on       = ext4_quota_on,
1465         .quota_off      = ext4_quota_off,
1466         .quota_sync     = dquot_quota_sync,
1467         .get_state      = dquot_get_state,
1468         .set_info       = dquot_set_dqinfo,
1469         .get_dqblk      = dquot_get_dqblk,
1470         .set_dqblk      = dquot_set_dqblk,
1471         .get_nextdqblk  = dquot_get_next_dqblk,
1472 };
1473 #endif
1474
1475 static const struct super_operations ext4_sops = {
1476         .alloc_inode    = ext4_alloc_inode,
1477         .free_inode     = ext4_free_in_core_inode,
1478         .destroy_inode  = ext4_destroy_inode,
1479         .write_inode    = ext4_write_inode,
1480         .dirty_inode    = ext4_dirty_inode,
1481         .drop_inode     = ext4_drop_inode,
1482         .evict_inode    = ext4_evict_inode,
1483         .put_super      = ext4_put_super,
1484         .sync_fs        = ext4_sync_fs,
1485         .freeze_fs      = ext4_freeze,
1486         .unfreeze_fs    = ext4_unfreeze,
1487         .statfs         = ext4_statfs,
1488         .remount_fs     = ext4_remount,
1489         .show_options   = ext4_show_options,
1490 #ifdef CONFIG_QUOTA
1491         .quota_read     = ext4_quota_read,
1492         .quota_write    = ext4_quota_write,
1493         .get_dquots     = ext4_get_dquots,
1494 #endif
1495         .bdev_try_to_free_page = bdev_try_to_free_page,
1496 };
1497
1498 static const struct export_operations ext4_export_ops = {
1499         .fh_to_dentry = ext4_fh_to_dentry,
1500         .fh_to_parent = ext4_fh_to_parent,
1501         .get_parent = ext4_get_parent,
1502         .commit_metadata = ext4_nfs_commit_metadata,
1503 };
1504
1505 enum {
1506         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1507         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1508         Opt_nouid32, Opt_debug, Opt_removed,
1509         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1510         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1511         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1512         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1513         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1514         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1515         Opt_inlinecrypt,
1516         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1517         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1518         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1519         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version,
1520         Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never,
1521         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
1522         Opt_nowarn_on_error, Opt_mblk_io_submit,
1523         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1524         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1525         Opt_inode_readahead_blks, Opt_journal_ioprio,
1526         Opt_dioread_nolock, Opt_dioread_lock,
1527         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1528         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1529         Opt_prefetch_block_bitmaps,
1530 };
1531
1532 static const match_table_t tokens = {
1533         {Opt_bsd_df, "bsddf"},
1534         {Opt_minix_df, "minixdf"},
1535         {Opt_grpid, "grpid"},
1536         {Opt_grpid, "bsdgroups"},
1537         {Opt_nogrpid, "nogrpid"},
1538         {Opt_nogrpid, "sysvgroups"},
1539         {Opt_resgid, "resgid=%u"},
1540         {Opt_resuid, "resuid=%u"},
1541         {Opt_sb, "sb=%u"},
1542         {Opt_err_cont, "errors=continue"},
1543         {Opt_err_panic, "errors=panic"},
1544         {Opt_err_ro, "errors=remount-ro"},
1545         {Opt_nouid32, "nouid32"},
1546         {Opt_debug, "debug"},
1547         {Opt_removed, "oldalloc"},
1548         {Opt_removed, "orlov"},
1549         {Opt_user_xattr, "user_xattr"},
1550         {Opt_nouser_xattr, "nouser_xattr"},
1551         {Opt_acl, "acl"},
1552         {Opt_noacl, "noacl"},
1553         {Opt_noload, "norecovery"},
1554         {Opt_noload, "noload"},
1555         {Opt_removed, "nobh"},
1556         {Opt_removed, "bh"},
1557         {Opt_commit, "commit=%u"},
1558         {Opt_min_batch_time, "min_batch_time=%u"},
1559         {Opt_max_batch_time, "max_batch_time=%u"},
1560         {Opt_journal_dev, "journal_dev=%u"},
1561         {Opt_journal_path, "journal_path=%s"},
1562         {Opt_journal_checksum, "journal_checksum"},
1563         {Opt_nojournal_checksum, "nojournal_checksum"},
1564         {Opt_journal_async_commit, "journal_async_commit"},
1565         {Opt_abort, "abort"},
1566         {Opt_data_journal, "data=journal"},
1567         {Opt_data_ordered, "data=ordered"},
1568         {Opt_data_writeback, "data=writeback"},
1569         {Opt_data_err_abort, "data_err=abort"},
1570         {Opt_data_err_ignore, "data_err=ignore"},
1571         {Opt_offusrjquota, "usrjquota="},
1572         {Opt_usrjquota, "usrjquota=%s"},
1573         {Opt_offgrpjquota, "grpjquota="},
1574         {Opt_grpjquota, "grpjquota=%s"},
1575         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1576         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1577         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1578         {Opt_grpquota, "grpquota"},
1579         {Opt_noquota, "noquota"},
1580         {Opt_quota, "quota"},
1581         {Opt_usrquota, "usrquota"},
1582         {Opt_prjquota, "prjquota"},
1583         {Opt_barrier, "barrier=%u"},
1584         {Opt_barrier, "barrier"},
1585         {Opt_nobarrier, "nobarrier"},
1586         {Opt_i_version, "i_version"},
1587         {Opt_dax, "dax"},
1588         {Opt_dax_always, "dax=always"},
1589         {Opt_dax_inode, "dax=inode"},
1590         {Opt_dax_never, "dax=never"},
1591         {Opt_stripe, "stripe=%u"},
1592         {Opt_delalloc, "delalloc"},
1593         {Opt_warn_on_error, "warn_on_error"},
1594         {Opt_nowarn_on_error, "nowarn_on_error"},
1595         {Opt_lazytime, "lazytime"},
1596         {Opt_nolazytime, "nolazytime"},
1597         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1598         {Opt_nodelalloc, "nodelalloc"},
1599         {Opt_removed, "mblk_io_submit"},
1600         {Opt_removed, "nomblk_io_submit"},
1601         {Opt_block_validity, "block_validity"},
1602         {Opt_noblock_validity, "noblock_validity"},
1603         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1604         {Opt_journal_ioprio, "journal_ioprio=%u"},
1605         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1606         {Opt_auto_da_alloc, "auto_da_alloc"},
1607         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1608         {Opt_dioread_nolock, "dioread_nolock"},
1609         {Opt_dioread_lock, "nodioread_nolock"},
1610         {Opt_dioread_lock, "dioread_lock"},
1611         {Opt_discard, "discard"},
1612         {Opt_nodiscard, "nodiscard"},
1613         {Opt_init_itable, "init_itable=%u"},
1614         {Opt_init_itable, "init_itable"},
1615         {Opt_noinit_itable, "noinit_itable"},
1616         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1617         {Opt_test_dummy_encryption, "test_dummy_encryption=%s"},
1618         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1619         {Opt_inlinecrypt, "inlinecrypt"},
1620         {Opt_nombcache, "nombcache"},
1621         {Opt_nombcache, "no_mbcache"},  /* for backward compatibility */
1622         {Opt_prefetch_block_bitmaps, "prefetch_block_bitmaps"},
1623         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1624         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1625         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1626         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1627         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1628         {Opt_err, NULL},
1629 };
1630
1631 static ext4_fsblk_t get_sb_block(void **data)
1632 {
1633         ext4_fsblk_t    sb_block;
1634         char            *options = (char *) *data;
1635
1636         if (!options || strncmp(options, "sb=", 3) != 0)
1637                 return 1;       /* Default location */
1638
1639         options += 3;
1640         /* TODO: use simple_strtoll with >32bit ext4 */
1641         sb_block = simple_strtoul(options, &options, 0);
1642         if (*options && *options != ',') {
1643                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1644                        (char *) *data);
1645                 return 1;
1646         }
1647         if (*options == ',')
1648                 options++;
1649         *data = (void *) options;
1650
1651         return sb_block;
1652 }
1653
1654 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1655 static const char deprecated_msg[] =
1656         "Mount option \"%s\" will be removed by %s\n"
1657         "Contact [email protected] if you think we should keep it.\n";
1658
1659 #ifdef CONFIG_QUOTA
1660 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1661 {
1662         struct ext4_sb_info *sbi = EXT4_SB(sb);
1663         char *qname, *old_qname = get_qf_name(sb, sbi, qtype);
1664         int ret = -1;
1665
1666         if (sb_any_quota_loaded(sb) && !old_qname) {
1667                 ext4_msg(sb, KERN_ERR,
1668                         "Cannot change journaled "
1669                         "quota options when quota turned on");
1670                 return -1;
1671         }
1672         if (ext4_has_feature_quota(sb)) {
1673                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1674                          "ignored when QUOTA feature is enabled");
1675                 return 1;
1676         }
1677         qname = match_strdup(args);
1678         if (!qname) {
1679                 ext4_msg(sb, KERN_ERR,
1680                         "Not enough memory for storing quotafile name");
1681                 return -1;
1682         }
1683         if (old_qname) {
1684                 if (strcmp(old_qname, qname) == 0)
1685                         ret = 1;
1686                 else
1687                         ext4_msg(sb, KERN_ERR,
1688                                  "%s quota file already specified",
1689                                  QTYPE2NAME(qtype));
1690                 goto errout;
1691         }
1692         if (strchr(qname, '/')) {
1693                 ext4_msg(sb, KERN_ERR,
1694                         "quotafile must be on filesystem root");
1695                 goto errout;
1696         }
1697         rcu_assign_pointer(sbi->s_qf_names[qtype], qname);
1698         set_opt(sb, QUOTA);
1699         return 1;
1700 errout:
1701         kfree(qname);
1702         return ret;
1703 }
1704
1705 static int clear_qf_name(struct super_block *sb, int qtype)
1706 {
1707
1708         struct ext4_sb_info *sbi = EXT4_SB(sb);
1709         char *old_qname = get_qf_name(sb, sbi, qtype);
1710
1711         if (sb_any_quota_loaded(sb) && old_qname) {
1712                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1713                         " when quota turned on");
1714                 return -1;
1715         }
1716         rcu_assign_pointer(sbi->s_qf_names[qtype], NULL);
1717         synchronize_rcu();
1718         kfree(old_qname);
1719         return 1;
1720 }
1721 #endif
1722
1723 #define MOPT_SET        0x0001
1724 #define MOPT_CLEAR      0x0002
1725 #define MOPT_NOSUPPORT  0x0004
1726 #define MOPT_EXPLICIT   0x0008
1727 #define MOPT_CLEAR_ERR  0x0010
1728 #define MOPT_GTE0       0x0020
1729 #ifdef CONFIG_QUOTA
1730 #define MOPT_Q          0
1731 #define MOPT_QFMT       0x0040
1732 #else
1733 #define MOPT_Q          MOPT_NOSUPPORT
1734 #define MOPT_QFMT       MOPT_NOSUPPORT
1735 #endif
1736 #define MOPT_DATAJ      0x0080
1737 #define MOPT_NO_EXT2    0x0100
1738 #define MOPT_NO_EXT3    0x0200
1739 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1740 #define MOPT_STRING     0x0400
1741 #define MOPT_SKIP       0x0800
1742
1743 static const struct mount_opts {
1744         int     token;
1745         int     mount_opt;
1746         int     flags;
1747 } ext4_mount_opts[] = {
1748         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1749         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1750         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1751         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1752         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1753         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1754         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1755          MOPT_EXT4_ONLY | MOPT_SET},
1756         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1757          MOPT_EXT4_ONLY | MOPT_CLEAR},
1758         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1759         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1760         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1761          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1762         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1763          MOPT_EXT4_ONLY | MOPT_CLEAR},
1764         {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET},
1765         {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR},
1766         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1767          MOPT_EXT4_ONLY | MOPT_CLEAR},
1768         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1769          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1770         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1771                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1772          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1773         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1774         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1775         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1776         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1777         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1778          MOPT_NO_EXT2},
1779         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1780          MOPT_NO_EXT2},
1781         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1782         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1783         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1784         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1785         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1786         {Opt_commit, 0, MOPT_GTE0},
1787         {Opt_max_batch_time, 0, MOPT_GTE0},
1788         {Opt_min_batch_time, 0, MOPT_GTE0},
1789         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1790         {Opt_init_itable, 0, MOPT_GTE0},
1791         {Opt_dax, EXT4_MOUNT_DAX_ALWAYS, MOPT_SET | MOPT_SKIP},
1792         {Opt_dax_always, EXT4_MOUNT_DAX_ALWAYS,
1793                 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP},
1794         {Opt_dax_inode, EXT4_MOUNT2_DAX_INODE,
1795                 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP},
1796         {Opt_dax_never, EXT4_MOUNT2_DAX_NEVER,
1797                 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP},
1798         {Opt_stripe, 0, MOPT_GTE0},
1799         {Opt_resuid, 0, MOPT_GTE0},
1800         {Opt_resgid, 0, MOPT_GTE0},
1801         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1802         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1803         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1804         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1805         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1806         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1807          MOPT_NO_EXT2 | MOPT_DATAJ},
1808         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1809         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1810 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1811         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1812         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1813 #else
1814         {Opt_acl, 0, MOPT_NOSUPPORT},
1815         {Opt_noacl, 0, MOPT_NOSUPPORT},
1816 #endif
1817         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1818         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1819         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1820         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1821         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1822                                                         MOPT_SET | MOPT_Q},
1823         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1824                                                         MOPT_SET | MOPT_Q},
1825         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1826                                                         MOPT_SET | MOPT_Q},
1827         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1828                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1829                                                         MOPT_CLEAR | MOPT_Q},
1830         {Opt_usrjquota, 0, MOPT_Q},
1831         {Opt_grpjquota, 0, MOPT_Q},
1832         {Opt_offusrjquota, 0, MOPT_Q},
1833         {Opt_offgrpjquota, 0, MOPT_Q},
1834         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1835         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1836         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1837         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1838         {Opt_test_dummy_encryption, 0, MOPT_STRING},
1839         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1840         {Opt_prefetch_block_bitmaps, EXT4_MOUNT_PREFETCH_BLOCK_BITMAPS,
1841          MOPT_SET},
1842         {Opt_err, 0, 0}
1843 };
1844
1845 #ifdef CONFIG_UNICODE
1846 static const struct ext4_sb_encodings {
1847         __u16 magic;
1848         char *name;
1849         char *version;
1850 } ext4_sb_encoding_map[] = {
1851         {EXT4_ENC_UTF8_12_1, "utf8", "12.1.0"},
1852 };
1853
1854 static int ext4_sb_read_encoding(const struct ext4_super_block *es,
1855                                  const struct ext4_sb_encodings **encoding,
1856                                  __u16 *flags)
1857 {
1858         __u16 magic = le16_to_cpu(es->s_encoding);
1859         int i;
1860
1861         for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++)
1862                 if (magic == ext4_sb_encoding_map[i].magic)
1863                         break;
1864
1865         if (i >= ARRAY_SIZE(ext4_sb_encoding_map))
1866                 return -EINVAL;
1867
1868         *encoding = &ext4_sb_encoding_map[i];
1869         *flags = le16_to_cpu(es->s_encoding_flags);
1870
1871         return 0;
1872 }
1873 #endif
1874
1875 static int ext4_set_test_dummy_encryption(struct super_block *sb,
1876                                           const char *opt,
1877                                           const substring_t *arg,
1878                                           bool is_remount)
1879 {
1880 #ifdef CONFIG_FS_ENCRYPTION
1881         struct ext4_sb_info *sbi = EXT4_SB(sb);
1882         int err;
1883
1884         /*
1885          * This mount option is just for testing, and it's not worthwhile to
1886          * implement the extra complexity (e.g. RCU protection) that would be
1887          * needed to allow it to be set or changed during remount.  We do allow
1888          * it to be specified during remount, but only if there is no change.
1889          */
1890         if (is_remount && !sbi->s_dummy_enc_policy.policy) {
1891                 ext4_msg(sb, KERN_WARNING,
1892                          "Can't set test_dummy_encryption on remount");
1893                 return -1;
1894         }
1895         err = fscrypt_set_test_dummy_encryption(sb, arg->from,
1896                                                 &sbi->s_dummy_enc_policy);
1897         if (err) {
1898                 if (err == -EEXIST)
1899                         ext4_msg(sb, KERN_WARNING,
1900                                  "Can't change test_dummy_encryption on remount");
1901                 else if (err == -EINVAL)
1902                         ext4_msg(sb, KERN_WARNING,
1903                                  "Value of option \"%s\" is unrecognized", opt);
1904                 else
1905                         ext4_msg(sb, KERN_WARNING,
1906                                  "Error processing option \"%s\" [%d]",
1907                                  opt, err);
1908                 return -1;
1909         }
1910         ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled");
1911 #else
1912         ext4_msg(sb, KERN_WARNING,
1913                  "Test dummy encryption mount option ignored");
1914 #endif
1915         return 1;
1916 }
1917
1918 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1919                             substring_t *args, unsigned long *journal_devnum,
1920                             unsigned int *journal_ioprio, int is_remount)
1921 {
1922         struct ext4_sb_info *sbi = EXT4_SB(sb);
1923         const struct mount_opts *m;
1924         kuid_t uid;
1925         kgid_t gid;
1926         int arg = 0;
1927
1928 #ifdef CONFIG_QUOTA
1929         if (token == Opt_usrjquota)
1930                 return set_qf_name(sb, USRQUOTA, &args[0]);
1931         else if (token == Opt_grpjquota)
1932                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1933         else if (token == Opt_offusrjquota)
1934                 return clear_qf_name(sb, USRQUOTA);
1935         else if (token == Opt_offgrpjquota)
1936                 return clear_qf_name(sb, GRPQUOTA);
1937 #endif
1938         switch (token) {
1939         case Opt_noacl:
1940         case Opt_nouser_xattr:
1941                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1942                 break;
1943         case Opt_sb:
1944                 return 1;       /* handled by get_sb_block() */
1945         case Opt_removed:
1946                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1947                 return 1;
1948         case Opt_abort:
1949                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1950                 return 1;
1951         case Opt_i_version:
1952                 sb->s_flags |= SB_I_VERSION;
1953                 return 1;
1954         case Opt_lazytime:
1955                 sb->s_flags |= SB_LAZYTIME;
1956                 return 1;
1957         case Opt_nolazytime:
1958                 sb->s_flags &= ~SB_LAZYTIME;
1959                 return 1;
1960         case Opt_inlinecrypt:
1961 #ifdef CONFIG_FS_ENCRYPTION_INLINE_CRYPT
1962                 sb->s_flags |= SB_INLINECRYPT;
1963 #else
1964                 ext4_msg(sb, KERN_ERR, "inline encryption not supported");
1965 #endif
1966                 return 1;
1967         }
1968
1969         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1970                 if (token == m->token)
1971                         break;
1972
1973         if (m->token == Opt_err) {
1974                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1975                          "or missing value", opt);
1976                 return -1;
1977         }
1978
1979         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1980                 ext4_msg(sb, KERN_ERR,
1981                          "Mount option \"%s\" incompatible with ext2", opt);
1982                 return -1;
1983         }
1984         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1985                 ext4_msg(sb, KERN_ERR,
1986                          "Mount option \"%s\" incompatible with ext3", opt);
1987                 return -1;
1988         }
1989
1990         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1991                 return -1;
1992         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1993                 return -1;
1994         if (m->flags & MOPT_EXPLICIT) {
1995                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1996                         set_opt2(sb, EXPLICIT_DELALLOC);
1997                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1998                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1999                 } else
2000                         return -1;
2001         }
2002         if (m->flags & MOPT_CLEAR_ERR)
2003                 clear_opt(sb, ERRORS_MASK);
2004         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
2005                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
2006                          "options when quota turned on");
2007                 return -1;
2008         }
2009
2010         if (m->flags & MOPT_NOSUPPORT) {
2011                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
2012         } else if (token == Opt_commit) {
2013                 if (arg == 0)
2014                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
2015                 else if (arg > INT_MAX / HZ) {
2016                         ext4_msg(sb, KERN_ERR,
2017                                  "Invalid commit interval %d, "
2018                                  "must be smaller than %d",
2019                                  arg, INT_MAX / HZ);
2020                         return -1;
2021                 }
2022                 sbi->s_commit_interval = HZ * arg;
2023         } else if (token == Opt_debug_want_extra_isize) {
2024                 if ((arg & 1) ||
2025                     (arg < 4) ||
2026                     (arg > (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE))) {
2027                         ext4_msg(sb, KERN_ERR,
2028                                  "Invalid want_extra_isize %d", arg);
2029                         return -1;
2030                 }
2031                 sbi->s_want_extra_isize = arg;
2032         } else if (token == Opt_max_batch_time) {
2033                 sbi->s_max_batch_time = arg;
2034         } else if (token == Opt_min_batch_time) {
2035                 sbi->s_min_batch_time = arg;
2036         } else if (token == Opt_inode_readahead_blks) {
2037                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
2038                         ext4_msg(sb, KERN_ERR,
2039                                  "EXT4-fs: inode_readahead_blks must be "
2040                                  "0 or a power of 2 smaller than 2^31");
2041                         return -1;
2042                 }
2043                 sbi->s_inode_readahead_blks = arg;
2044         } else if (token == Opt_init_itable) {
2045                 set_opt(sb, INIT_INODE_TABLE);
2046                 if (!args->from)
2047                         arg = EXT4_DEF_LI_WAIT_MULT;
2048                 sbi->s_li_wait_mult = arg;
2049         } else if (token == Opt_max_dir_size_kb) {
2050                 sbi->s_max_dir_size_kb = arg;
2051         } else if (token == Opt_stripe) {
2052                 sbi->s_stripe = arg;
2053         } else if (token == Opt_resuid) {
2054                 uid = make_kuid(current_user_ns(), arg);
2055                 if (!uid_valid(uid)) {
2056                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
2057                         return -1;
2058                 }
2059                 sbi->s_resuid = uid;
2060         } else if (token == Opt_resgid) {
2061                 gid = make_kgid(current_user_ns(), arg);
2062                 if (!gid_valid(gid)) {
2063                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
2064                         return -1;
2065                 }
2066                 sbi->s_resgid = gid;
2067         } else if (token == Opt_journal_dev) {
2068                 if (is_remount) {
2069                         ext4_msg(sb, KERN_ERR,
2070                                  "Cannot specify journal on remount");
2071                         return -1;
2072                 }
2073                 *journal_devnum = arg;
2074         } else if (token == Opt_journal_path) {
2075                 char *journal_path;
2076                 struct inode *journal_inode;
2077                 struct path path;
2078                 int error;
2079
2080                 if (is_remount) {
2081                         ext4_msg(sb, KERN_ERR,
2082                                  "Cannot specify journal on remount");
2083                         return -1;
2084                 }
2085                 journal_path = match_strdup(&args[0]);
2086                 if (!journal_path) {
2087                         ext4_msg(sb, KERN_ERR, "error: could not dup "
2088                                 "journal device string");
2089                         return -1;
2090                 }
2091
2092                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
2093                 if (error) {
2094                         ext4_msg(sb, KERN_ERR, "error: could not find "
2095                                 "journal device path: error %d", error);
2096                         kfree(journal_path);
2097                         return -1;
2098                 }
2099
2100                 journal_inode = d_inode(path.dentry);
2101                 if (!S_ISBLK(journal_inode->i_mode)) {
2102                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
2103                                 "is not a block device", journal_path);
2104                         path_put(&path);
2105                         kfree(journal_path);
2106                         return -1;
2107                 }
2108
2109                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
2110                 path_put(&path);
2111                 kfree(journal_path);
2112         } else if (token == Opt_journal_ioprio) {
2113                 if (arg > 7) {
2114                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
2115                                  " (must be 0-7)");
2116                         return -1;
2117                 }
2118                 *journal_ioprio =
2119                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
2120         } else if (token == Opt_test_dummy_encryption) {
2121                 return ext4_set_test_dummy_encryption(sb, opt, &args[0],
2122                                                       is_remount);
2123         } else if (m->flags & MOPT_DATAJ) {
2124                 if (is_remount) {
2125                         if (!sbi->s_journal)
2126                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
2127                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
2128                                 ext4_msg(sb, KERN_ERR,
2129                                          "Cannot change data mode on remount");
2130                                 return -1;
2131                         }
2132                 } else {
2133                         clear_opt(sb, DATA_FLAGS);
2134                         sbi->s_mount_opt |= m->mount_opt;
2135                 }
2136 #ifdef CONFIG_QUOTA
2137         } else if (m->flags & MOPT_QFMT) {
2138                 if (sb_any_quota_loaded(sb) &&
2139                     sbi->s_jquota_fmt != m->mount_opt) {
2140                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
2141                                  "quota options when quota turned on");
2142                         return -1;
2143                 }
2144                 if (ext4_has_feature_quota(sb)) {
2145                         ext4_msg(sb, KERN_INFO,
2146                                  "Quota format mount options ignored "
2147                                  "when QUOTA feature is enabled");
2148                         return 1;
2149                 }
2150                 sbi->s_jquota_fmt = m->mount_opt;
2151 #endif
2152         } else if (token == Opt_dax || token == Opt_dax_always ||
2153                    token == Opt_dax_inode || token == Opt_dax_never) {
2154 #ifdef CONFIG_FS_DAX
2155                 switch (token) {
2156                 case Opt_dax:
2157                 case Opt_dax_always:
2158                         if (is_remount &&
2159                             (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2160                              (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) {
2161                         fail_dax_change_remount:
2162                                 ext4_msg(sb, KERN_ERR, "can't change "
2163                                          "dax mount option while remounting");
2164                                 return -1;
2165                         }
2166                         if (is_remount &&
2167                             (test_opt(sb, DATA_FLAGS) ==
2168                              EXT4_MOUNT_JOURNAL_DATA)) {
2169                                     ext4_msg(sb, KERN_ERR, "can't mount with "
2170                                              "both data=journal and dax");
2171                                     return -1;
2172                         }
2173                         ext4_msg(sb, KERN_WARNING,
2174                                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
2175                         sbi->s_mount_opt |= EXT4_MOUNT_DAX_ALWAYS;
2176                         sbi->s_mount_opt2 &= ~EXT4_MOUNT2_DAX_NEVER;
2177                         break;
2178                 case Opt_dax_never:
2179                         if (is_remount &&
2180                             (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2181                              (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS)))
2182                                 goto fail_dax_change_remount;
2183                         sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_NEVER;
2184                         sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS;
2185                         break;
2186                 case Opt_dax_inode:
2187                         if (is_remount &&
2188                             ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2189                              (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2190                              !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE)))
2191                                 goto fail_dax_change_remount;
2192                         sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS;
2193                         sbi->s_mount_opt2 &= ~EXT4_MOUNT2_DAX_NEVER;
2194                         /* Strictly for printing options */
2195                         sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_INODE;
2196                         break;
2197                 }
2198 #else
2199                 ext4_msg(sb, KERN_INFO, "dax option not supported");
2200                 sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_NEVER;
2201                 sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS;
2202                 return -1;
2203 #endif
2204         } else if (token == Opt_data_err_abort) {
2205                 sbi->s_mount_opt |= m->mount_opt;
2206         } else if (token == Opt_data_err_ignore) {
2207                 sbi->s_mount_opt &= ~m->mount_opt;
2208         } else {
2209                 if (!args->from)
2210                         arg = 1;
2211                 if (m->flags & MOPT_CLEAR)
2212                         arg = !arg;
2213                 else if (unlikely(!(m->flags & MOPT_SET))) {
2214                         ext4_msg(sb, KERN_WARNING,
2215                                  "buggy handling of option %s", opt);
2216                         WARN_ON(1);
2217                         return -1;
2218                 }
2219                 if (arg != 0)
2220                         sbi->s_mount_opt |= m->mount_opt;
2221                 else
2222                         sbi->s_mount_opt &= ~m->mount_opt;
2223         }
2224         return 1;
2225 }
2226
2227 static int parse_options(char *options, struct super_block *sb,
2228                          unsigned long *journal_devnum,
2229                          unsigned int *journal_ioprio,
2230                          int is_remount)
2231 {
2232         struct ext4_sb_info __maybe_unused *sbi = EXT4_SB(sb);
2233         char *p, __maybe_unused *usr_qf_name, __maybe_unused *grp_qf_name;
2234         substring_t args[MAX_OPT_ARGS];
2235         int token;
2236
2237         if (!options)
2238                 return 1;
2239
2240         while ((p = strsep(&options, ",")) != NULL) {
2241                 if (!*p)
2242                         continue;
2243                 /*
2244                  * Initialize args struct so we know whether arg was
2245                  * found; some options take optional arguments.
2246                  */
2247                 args[0].to = args[0].from = NULL;
2248                 token = match_token(p, tokens, args);
2249                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
2250                                      journal_ioprio, is_remount) < 0)
2251                         return 0;
2252         }
2253 #ifdef CONFIG_QUOTA
2254         /*
2255          * We do the test below only for project quotas. 'usrquota' and
2256          * 'grpquota' mount options are allowed even without quota feature
2257          * to support legacy quotas in quota files.
2258          */
2259         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
2260                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
2261                          "Cannot enable project quota enforcement.");
2262                 return 0;
2263         }
2264         usr_qf_name = get_qf_name(sb, sbi, USRQUOTA);
2265         grp_qf_name = get_qf_name(sb, sbi, GRPQUOTA);
2266         if (usr_qf_name || grp_qf_name) {
2267                 if (test_opt(sb, USRQUOTA) && usr_qf_name)
2268                         clear_opt(sb, USRQUOTA);
2269
2270                 if (test_opt(sb, GRPQUOTA) && grp_qf_name)
2271                         clear_opt(sb, GRPQUOTA);
2272
2273                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
2274                         ext4_msg(sb, KERN_ERR, "old and new quota "
2275                                         "format mixing");
2276                         return 0;
2277                 }
2278
2279                 if (!sbi->s_jquota_fmt) {
2280                         ext4_msg(sb, KERN_ERR, "journaled quota format "
2281                                         "not specified");
2282                         return 0;
2283                 }
2284         }
2285 #endif
2286         if (test_opt(sb, DIOREAD_NOLOCK)) {
2287                 int blocksize =
2288                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
2289                 if (blocksize < PAGE_SIZE)
2290                         ext4_msg(sb, KERN_WARNING, "Warning: mounting with an "
2291                                  "experimental mount option 'dioread_nolock' "
2292                                  "for blocksize < PAGE_SIZE");
2293         }
2294         return 1;
2295 }
2296
2297 static inline void ext4_show_quota_options(struct seq_file *seq,
2298                                            struct super_block *sb)
2299 {
2300 #if defined(CONFIG_QUOTA)
2301         struct ext4_sb_info *sbi = EXT4_SB(sb);
2302         char *usr_qf_name, *grp_qf_name;
2303
2304         if (sbi->s_jquota_fmt) {
2305                 char *fmtname = "";
2306
2307                 switch (sbi->s_jquota_fmt) {
2308                 case QFMT_VFS_OLD:
2309                         fmtname = "vfsold";
2310                         break;
2311                 case QFMT_VFS_V0:
2312                         fmtname = "vfsv0";
2313                         break;
2314                 case QFMT_VFS_V1:
2315                         fmtname = "vfsv1";
2316                         break;
2317                 }
2318                 seq_printf(seq, ",jqfmt=%s", fmtname);
2319         }
2320
2321         rcu_read_lock();
2322         usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]);
2323         grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]);
2324         if (usr_qf_name)
2325                 seq_show_option(seq, "usrjquota", usr_qf_name);
2326         if (grp_qf_name)
2327                 seq_show_option(seq, "grpjquota", grp_qf_name);
2328         rcu_read_unlock();
2329 #endif
2330 }
2331
2332 static const char *token2str(int token)
2333 {
2334         const struct match_token *t;
2335
2336         for (t = tokens; t->token != Opt_err; t++)
2337                 if (t->token == token && !strchr(t->pattern, '='))
2338                         break;
2339         return t->pattern;
2340 }
2341
2342 /*
2343  * Show an option if
2344  *  - it's set to a non-default value OR
2345  *  - if the per-sb default is different from the global default
2346  */
2347 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2348                               int nodefs)
2349 {
2350         struct ext4_sb_info *sbi = EXT4_SB(sb);
2351         struct ext4_super_block *es = sbi->s_es;
2352         int def_errors, def_mount_opt = sbi->s_def_mount_opt;
2353         const struct mount_opts *m;
2354         char sep = nodefs ? '\n' : ',';
2355
2356 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2357 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2358
2359         if (sbi->s_sb_block != 1)
2360                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2361
2362         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2363                 int want_set = m->flags & MOPT_SET;
2364                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2365                     (m->flags & MOPT_CLEAR_ERR) || m->flags & MOPT_SKIP)
2366                         continue;
2367                 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2368                         continue; /* skip if same as the default */
2369                 if ((want_set &&
2370                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2371                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2372                         continue; /* select Opt_noFoo vs Opt_Foo */
2373                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2374         }
2375
2376         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2377             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2378                 SEQ_OPTS_PRINT("resuid=%u",
2379                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2380         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2381             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2382                 SEQ_OPTS_PRINT("resgid=%u",
2383                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2384         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2385         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2386                 SEQ_OPTS_PUTS("errors=remount-ro");
2387         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2388                 SEQ_OPTS_PUTS("errors=continue");
2389         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2390                 SEQ_OPTS_PUTS("errors=panic");
2391         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2392                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2393         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2394                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2395         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2396                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2397         if (sb->s_flags & SB_I_VERSION)
2398                 SEQ_OPTS_PUTS("i_version");
2399         if (nodefs || sbi->s_stripe)
2400                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2401         if (nodefs || EXT4_MOUNT_DATA_FLAGS &
2402                         (sbi->s_mount_opt ^ def_mount_opt)) {
2403                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2404                         SEQ_OPTS_PUTS("data=journal");
2405                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2406                         SEQ_OPTS_PUTS("data=ordered");
2407                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2408                         SEQ_OPTS_PUTS("data=writeback");
2409         }
2410         if (nodefs ||
2411             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2412                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2413                                sbi->s_inode_readahead_blks);
2414
2415         if (test_opt(sb, INIT_INODE_TABLE) && (nodefs ||
2416                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2417                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2418         if (nodefs || sbi->s_max_dir_size_kb)
2419                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2420         if (test_opt(sb, DATA_ERR_ABORT))
2421                 SEQ_OPTS_PUTS("data_err=abort");
2422
2423         fscrypt_show_test_dummy_encryption(seq, sep, sb);
2424
2425         if (sb->s_flags & SB_INLINECRYPT)
2426                 SEQ_OPTS_PUTS("inlinecrypt");
2427
2428         if (test_opt(sb, DAX_ALWAYS)) {
2429                 if (IS_EXT2_SB(sb))
2430                         SEQ_OPTS_PUTS("dax");
2431                 else
2432                         SEQ_OPTS_PUTS("dax=always");
2433         } else if (test_opt2(sb, DAX_NEVER)) {
2434                 SEQ_OPTS_PUTS("dax=never");
2435         } else if (test_opt2(sb, DAX_INODE)) {
2436                 SEQ_OPTS_PUTS("dax=inode");
2437         }
2438
2439         ext4_show_quota_options(seq, sb);
2440         return 0;
2441 }
2442
2443 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2444 {
2445         return _ext4_show_options(seq, root->d_sb, 0);
2446 }
2447
2448 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2449 {
2450         struct super_block *sb = seq->private;
2451         int rc;
2452
2453         seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
2454         rc = _ext4_show_options(seq, sb, 1);
2455         seq_puts(seq, "\n");
2456         return rc;
2457 }
2458
2459 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2460                             int read_only)
2461 {
2462         struct ext4_sb_info *sbi = EXT4_SB(sb);
2463         int err = 0;
2464
2465         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2466                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2467                          "forcing read-only mode");
2468                 err = -EROFS;
2469                 goto done;
2470         }
2471         if (read_only)
2472                 goto done;
2473         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2474                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2475                          "running e2fsck is recommended");
2476         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2477                 ext4_msg(sb, KERN_WARNING,
2478                          "warning: mounting fs with errors, "
2479                          "running e2fsck is recommended");
2480         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2481                  le16_to_cpu(es->s_mnt_count) >=
2482                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2483                 ext4_msg(sb, KERN_WARNING,
2484                          "warning: maximal mount count reached, "
2485                          "running e2fsck is recommended");
2486         else if (le32_to_cpu(es->s_checkinterval) &&
2487                  (ext4_get_tstamp(es, s_lastcheck) +
2488                   le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds()))
2489                 ext4_msg(sb, KERN_WARNING,
2490                          "warning: checktime reached, "
2491                          "running e2fsck is recommended");
2492         if (!sbi->s_journal)
2493                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2494         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2495                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2496         le16_add_cpu(&es->s_mnt_count, 1);
2497         ext4_update_tstamp(es, s_mtime);
2498         if (sbi->s_journal)
2499                 ext4_set_feature_journal_needs_recovery(sb);
2500
2501         err = ext4_commit_super(sb, 1);
2502 done:
2503         if (test_opt(sb, DEBUG))
2504                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2505                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2506                         sb->s_blocksize,
2507                         sbi->s_groups_count,
2508                         EXT4_BLOCKS_PER_GROUP(sb),
2509                         EXT4_INODES_PER_GROUP(sb),
2510                         sbi->s_mount_opt, sbi->s_mount_opt2);
2511
2512         cleancache_init_fs(sb);
2513         return err;
2514 }
2515
2516 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2517 {
2518         struct ext4_sb_info *sbi = EXT4_SB(sb);
2519         struct flex_groups **old_groups, **new_groups;
2520         int size, i, j;
2521
2522         if (!sbi->s_log_groups_per_flex)
2523                 return 0;
2524
2525         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2526         if (size <= sbi->s_flex_groups_allocated)
2527                 return 0;
2528
2529         new_groups = kvzalloc(roundup_pow_of_two(size *
2530                               sizeof(*sbi->s_flex_groups)), GFP_KERNEL);
2531         if (!new_groups) {
2532                 ext4_msg(sb, KERN_ERR,
2533                          "not enough memory for %d flex group pointers", size);
2534                 return -ENOMEM;
2535         }
2536         for (i = sbi->s_flex_groups_allocated; i < size; i++) {
2537                 new_groups[i] = kvzalloc(roundup_pow_of_two(
2538                                          sizeof(struct flex_groups)),
2539                                          GFP_KERNEL);
2540                 if (!new_groups[i]) {
2541                         for (j = sbi->s_flex_groups_allocated; j < i; j++)
2542                                 kvfree(new_groups[j]);
2543                         kvfree(new_groups);
2544                         ext4_msg(sb, KERN_ERR,
2545                                  "not enough memory for %d flex groups", size);
2546                         return -ENOMEM;
2547                 }
2548         }
2549         rcu_read_lock();
2550         old_groups = rcu_dereference(sbi->s_flex_groups);
2551         if (old_groups)
2552                 memcpy(new_groups, old_groups,
2553                        (sbi->s_flex_groups_allocated *
2554                         sizeof(struct flex_groups *)));
2555         rcu_read_unlock();
2556         rcu_assign_pointer(sbi->s_flex_groups, new_groups);
2557         sbi->s_flex_groups_allocated = size;
2558         if (old_groups)
2559                 ext4_kvfree_array_rcu(old_groups);
2560         return 0;
2561 }
2562
2563 static int ext4_fill_flex_info(struct super_block *sb)
2564 {
2565         struct ext4_sb_info *sbi = EXT4_SB(sb);
2566         struct ext4_group_desc *gdp = NULL;
2567         struct flex_groups *fg;
2568         ext4_group_t flex_group;
2569         int i, err;
2570
2571         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2572         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2573                 sbi->s_log_groups_per_flex = 0;
2574                 return 1;
2575         }
2576
2577         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2578         if (err)
2579                 goto failed;
2580
2581         for (i = 0; i < sbi->s_groups_count; i++) {
2582                 gdp = ext4_get_group_desc(sb, i, NULL);
2583
2584                 flex_group = ext4_flex_group(sbi, i);
2585                 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
2586                 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes);
2587                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2588                              &fg->free_clusters);
2589                 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs);
2590         }
2591
2592         return 1;
2593 failed:
2594         return 0;
2595 }
2596
2597 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2598                                    struct ext4_group_desc *gdp)
2599 {
2600         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2601         __u16 crc = 0;
2602         __le32 le_group = cpu_to_le32(block_group);
2603         struct ext4_sb_info *sbi = EXT4_SB(sb);
2604
2605         if (ext4_has_metadata_csum(sbi->s_sb)) {
2606                 /* Use new metadata_csum algorithm */
2607                 __u32 csum32;
2608                 __u16 dummy_csum = 0;
2609
2610                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2611                                      sizeof(le_group));
2612                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2613                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2614                                      sizeof(dummy_csum));
2615                 offset += sizeof(dummy_csum);
2616                 if (offset < sbi->s_desc_size)
2617                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2618                                              sbi->s_desc_size - offset);
2619
2620                 crc = csum32 & 0xFFFF;
2621                 goto out;
2622         }
2623
2624         /* old crc16 code */
2625         if (!ext4_has_feature_gdt_csum(sb))
2626                 return 0;
2627
2628         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2629         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2630         crc = crc16(crc, (__u8 *)gdp, offset);
2631         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2632         /* for checksum of struct ext4_group_desc do the rest...*/
2633         if (ext4_has_feature_64bit(sb) &&
2634             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2635                 crc = crc16(crc, (__u8 *)gdp + offset,
2636                             le16_to_cpu(sbi->s_es->s_desc_size) -
2637                                 offset);
2638
2639 out:
2640         return cpu_to_le16(crc);
2641 }
2642
2643 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2644                                 struct ext4_group_desc *gdp)
2645 {
2646         if (ext4_has_group_desc_csum(sb) &&
2647             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2648                 return 0;
2649
2650         return 1;
2651 }
2652
2653 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2654                               struct ext4_group_desc *gdp)
2655 {
2656         if (!ext4_has_group_desc_csum(sb))
2657                 return;
2658         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2659 }
2660
2661 /* Called at mount-time, super-block is locked */
2662 static int ext4_check_descriptors(struct super_block *sb,
2663                                   ext4_fsblk_t sb_block,
2664                                   ext4_group_t *first_not_zeroed)
2665 {
2666         struct ext4_sb_info *sbi = EXT4_SB(sb);
2667         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2668         ext4_fsblk_t last_block;
2669         ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0);
2670         ext4_fsblk_t block_bitmap;
2671         ext4_fsblk_t inode_bitmap;
2672         ext4_fsblk_t inode_table;
2673         int flexbg_flag = 0;
2674         ext4_group_t i, grp = sbi->s_groups_count;
2675
2676         if (ext4_has_feature_flex_bg(sb))
2677                 flexbg_flag = 1;
2678
2679         ext4_debug("Checking group descriptors");
2680
2681         for (i = 0; i < sbi->s_groups_count; i++) {
2682                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2683
2684                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2685                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2686                 else
2687                         last_block = first_block +
2688                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2689
2690                 if ((grp == sbi->s_groups_count) &&
2691                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2692                         grp = i;
2693
2694                 block_bitmap = ext4_block_bitmap(sb, gdp);
2695                 if (block_bitmap == sb_block) {
2696                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2697                                  "Block bitmap for group %u overlaps "
2698                                  "superblock", i);
2699                         if (!sb_rdonly(sb))
2700                                 return 0;
2701                 }
2702                 if (block_bitmap >= sb_block + 1 &&
2703                     block_bitmap <= last_bg_block) {
2704                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2705                                  "Block bitmap for group %u overlaps "
2706                                  "block group descriptors", i);
2707                         if (!sb_rdonly(sb))
2708                                 return 0;
2709                 }
2710                 if (block_bitmap < first_block || block_bitmap > last_block) {
2711                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2712                                "Block bitmap for group %u not in group "
2713                                "(block %llu)!", i, block_bitmap);
2714                         return 0;
2715                 }
2716                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2717                 if (inode_bitmap == sb_block) {
2718                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2719                                  "Inode bitmap for group %u overlaps "
2720                                  "superblock", i);
2721                         if (!sb_rdonly(sb))
2722                                 return 0;
2723                 }
2724                 if (inode_bitmap >= sb_block + 1 &&
2725                     inode_bitmap <= last_bg_block) {
2726                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2727                                  "Inode bitmap for group %u overlaps "
2728                                  "block group descriptors", i);
2729                         if (!sb_rdonly(sb))
2730                                 return 0;
2731                 }
2732                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2733                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2734                                "Inode bitmap for group %u not in group "
2735                                "(block %llu)!", i, inode_bitmap);
2736                         return 0;
2737                 }
2738                 inode_table = ext4_inode_table(sb, gdp);
2739                 if (inode_table == sb_block) {
2740                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2741                                  "Inode table for group %u overlaps "
2742                                  "superblock", i);
2743                         if (!sb_rdonly(sb))
2744                                 return 0;
2745                 }
2746                 if (inode_table >= sb_block + 1 &&
2747                     inode_table <= last_bg_block) {
2748                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2749                                  "Inode table for group %u overlaps "
2750                                  "block group descriptors", i);
2751                         if (!sb_rdonly(sb))
2752                                 return 0;
2753                 }
2754                 if (inode_table < first_block ||
2755                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2756                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2757                                "Inode table for group %u not in group "
2758                                "(block %llu)!", i, inode_table);
2759                         return 0;
2760                 }
2761                 ext4_lock_group(sb, i);
2762                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2763                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2764                                  "Checksum for group %u failed (%u!=%u)",
2765                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2766                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2767                         if (!sb_rdonly(sb)) {
2768                                 ext4_unlock_group(sb, i);
2769                                 return 0;
2770                         }
2771                 }
2772                 ext4_unlock_group(sb, i);
2773                 if (!flexbg_flag)
2774                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2775         }
2776         if (NULL != first_not_zeroed)
2777                 *first_not_zeroed = grp;
2778         return 1;
2779 }
2780
2781 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2782  * the superblock) which were deleted from all directories, but held open by
2783  * a process at the time of a crash.  We walk the list and try to delete these
2784  * inodes at recovery time (only with a read-write filesystem).
2785  *
2786  * In order to keep the orphan inode chain consistent during traversal (in
2787  * case of crash during recovery), we link each inode into the superblock
2788  * orphan list_head and handle it the same way as an inode deletion during
2789  * normal operation (which journals the operations for us).
2790  *
2791  * We only do an iget() and an iput() on each inode, which is very safe if we
2792  * accidentally point at an in-use or already deleted inode.  The worst that
2793  * can happen in this case is that we get a "bit already cleared" message from
2794  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2795  * e2fsck was run on this filesystem, and it must have already done the orphan
2796  * inode cleanup for us, so we can safely abort without any further action.
2797  */
2798 static void ext4_orphan_cleanup(struct super_block *sb,
2799                                 struct ext4_super_block *es)
2800 {
2801         unsigned int s_flags = sb->s_flags;
2802         int ret, nr_orphans = 0, nr_truncates = 0;
2803 #ifdef CONFIG_QUOTA
2804         int quota_update = 0;
2805         int i;
2806 #endif
2807         if (!es->s_last_orphan) {
2808                 jbd_debug(4, "no orphan inodes to clean up\n");
2809                 return;
2810         }
2811
2812         if (bdev_read_only(sb->s_bdev)) {
2813                 ext4_msg(sb, KERN_ERR, "write access "
2814                         "unavailable, skipping orphan cleanup");
2815                 return;
2816         }
2817
2818         /* Check if feature set would not allow a r/w mount */
2819         if (!ext4_feature_set_ok(sb, 0)) {
2820                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2821                          "unknown ROCOMPAT features");
2822                 return;
2823         }
2824
2825         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2826                 /* don't clear list on RO mount w/ errors */
2827                 if (es->s_last_orphan && !(s_flags & SB_RDONLY)) {
2828                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2829                                   "clearing orphan list.\n");
2830                         es->s_last_orphan = 0;
2831                 }
2832                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2833                 return;
2834         }
2835
2836         if (s_flags & SB_RDONLY) {
2837                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2838                 sb->s_flags &= ~SB_RDONLY;
2839         }
2840 #ifdef CONFIG_QUOTA
2841         /* Needed for iput() to work correctly and not trash data */
2842         sb->s_flags |= SB_ACTIVE;
2843
2844         /*
2845          * Turn on quotas which were not enabled for read-only mounts if
2846          * filesystem has quota feature, so that they are updated correctly.
2847          */
2848         if (ext4_has_feature_quota(sb) && (s_flags & SB_RDONLY)) {
2849                 int ret = ext4_enable_quotas(sb);
2850
2851                 if (!ret)
2852                         quota_update = 1;
2853                 else
2854                         ext4_msg(sb, KERN_ERR,
2855                                 "Cannot turn on quotas: error %d", ret);
2856         }
2857
2858         /* Turn on journaled quotas used for old sytle */
2859         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2860                 if (EXT4_SB(sb)->s_qf_names[i]) {
2861                         int ret = ext4_quota_on_mount(sb, i);
2862
2863                         if (!ret)
2864                                 quota_update = 1;
2865                         else
2866                                 ext4_msg(sb, KERN_ERR,
2867                                         "Cannot turn on journaled "
2868                                         "quota: type %d: error %d", i, ret);
2869                 }
2870         }
2871 #endif
2872
2873         while (es->s_last_orphan) {
2874                 struct inode *inode;
2875
2876                 /*
2877                  * We may have encountered an error during cleanup; if
2878                  * so, skip the rest.
2879                  */
2880                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2881                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2882                         es->s_last_orphan = 0;
2883                         break;
2884                 }
2885
2886                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2887                 if (IS_ERR(inode)) {
2888                         es->s_last_orphan = 0;
2889                         break;
2890                 }
2891
2892                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2893                 dquot_initialize(inode);
2894                 if (inode->i_nlink) {
2895                         if (test_opt(sb, DEBUG))
2896                                 ext4_msg(sb, KERN_DEBUG,
2897                                         "%s: truncating inode %lu to %lld bytes",
2898                                         __func__, inode->i_ino, inode->i_size);
2899                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2900                                   inode->i_ino, inode->i_size);
2901                         inode_lock(inode);
2902                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2903                         ret = ext4_truncate(inode);
2904                         if (ret)
2905                                 ext4_std_error(inode->i_sb, ret);
2906                         inode_unlock(inode);
2907                         nr_truncates++;
2908                 } else {
2909                         if (test_opt(sb, DEBUG))
2910                                 ext4_msg(sb, KERN_DEBUG,
2911                                         "%s: deleting unreferenced inode %lu",
2912                                         __func__, inode->i_ino);
2913                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2914                                   inode->i_ino);
2915                         nr_orphans++;
2916                 }
2917                 iput(inode);  /* The delete magic happens here! */
2918         }
2919
2920 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2921
2922         if (nr_orphans)
2923                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2924                        PLURAL(nr_orphans));
2925         if (nr_truncates)
2926                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2927                        PLURAL(nr_truncates));
2928 #ifdef CONFIG_QUOTA
2929         /* Turn off quotas if they were enabled for orphan cleanup */
2930         if (quota_update) {
2931                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2932                         if (sb_dqopt(sb)->files[i])
2933                                 dquot_quota_off(sb, i);
2934                 }
2935         }
2936 #endif
2937         sb->s_flags = s_flags; /* Restore SB_RDONLY status */
2938 }
2939
2940 /*
2941  * Maximal extent format file size.
2942  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2943  * extent format containers, within a sector_t, and within i_blocks
2944  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2945  * so that won't be a limiting factor.
2946  *
2947  * However there is other limiting factor. We do store extents in the form
2948  * of starting block and length, hence the resulting length of the extent
2949  * covering maximum file size must fit into on-disk format containers as
2950  * well. Given that length is always by 1 unit bigger than max unit (because
2951  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2952  *
2953  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2954  */
2955 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2956 {
2957         loff_t res;
2958         loff_t upper_limit = MAX_LFS_FILESIZE;
2959
2960         BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64));
2961
2962         if (!has_huge_files) {
2963                 upper_limit = (1LL << 32) - 1;
2964
2965                 /* total blocks in file system block size */
2966                 upper_limit >>= (blkbits - 9);
2967                 upper_limit <<= blkbits;
2968         }
2969
2970         /*
2971          * 32-bit extent-start container, ee_block. We lower the maxbytes
2972          * by one fs block, so ee_len can cover the extent of maximum file
2973          * size
2974          */
2975         res = (1LL << 32) - 1;
2976         res <<= blkbits;
2977
2978         /* Sanity check against vm- & vfs- imposed limits */
2979         if (res > upper_limit)
2980                 res = upper_limit;
2981
2982         return res;
2983 }
2984
2985 /*
2986  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2987  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2988  * We need to be 1 filesystem block less than the 2^48 sector limit.
2989  */
2990 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2991 {
2992         loff_t res = EXT4_NDIR_BLOCKS;
2993         int meta_blocks;
2994         loff_t upper_limit;
2995         /* This is calculated to be the largest file size for a dense, block
2996          * mapped file such that the file's total number of 512-byte sectors,
2997          * including data and all indirect blocks, does not exceed (2^48 - 1).
2998          *
2999          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
3000          * number of 512-byte sectors of the file.
3001          */
3002
3003         if (!has_huge_files) {
3004                 /*
3005                  * !has_huge_files or implies that the inode i_block field
3006                  * represents total file blocks in 2^32 512-byte sectors ==
3007                  * size of vfs inode i_blocks * 8
3008                  */
3009                 upper_limit = (1LL << 32) - 1;
3010
3011                 /* total blocks in file system block size */
3012                 upper_limit >>= (bits - 9);
3013
3014         } else {
3015                 /*
3016                  * We use 48 bit ext4_inode i_blocks
3017                  * With EXT4_HUGE_FILE_FL set the i_blocks
3018                  * represent total number of blocks in
3019                  * file system block size
3020                  */
3021                 upper_limit = (1LL << 48) - 1;
3022
3023         }
3024
3025         /* indirect blocks */
3026         meta_blocks = 1;
3027         /* double indirect blocks */
3028         meta_blocks += 1 + (1LL << (bits-2));
3029         /* tripple indirect blocks */
3030         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
3031
3032         upper_limit -= meta_blocks;
3033         upper_limit <<= bits;
3034
3035         res += 1LL << (bits-2);
3036         res += 1LL << (2*(bits-2));
3037         res += 1LL << (3*(bits-2));
3038         res <<= bits;
3039         if (res > upper_limit)
3040                 res = upper_limit;
3041
3042         if (res > MAX_LFS_FILESIZE)
3043                 res = MAX_LFS_FILESIZE;
3044
3045         return res;
3046 }
3047
3048 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
3049                                    ext4_fsblk_t logical_sb_block, int nr)
3050 {
3051         struct ext4_sb_info *sbi = EXT4_SB(sb);
3052         ext4_group_t bg, first_meta_bg;
3053         int has_super = 0;
3054
3055         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
3056
3057         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
3058                 return logical_sb_block + nr + 1;
3059         bg = sbi->s_desc_per_block * nr;
3060         if (ext4_bg_has_super(sb, bg))
3061                 has_super = 1;
3062
3063         /*
3064          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
3065          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
3066          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
3067          * compensate.
3068          */
3069         if (sb->s_blocksize == 1024 && nr == 0 &&
3070             le32_to_cpu(sbi->s_es->s_first_data_block) == 0)
3071                 has_super++;
3072
3073         return (has_super + ext4_group_first_block_no(sb, bg));
3074 }
3075
3076 /**
3077  * ext4_get_stripe_size: Get the stripe size.
3078  * @sbi: In memory super block info
3079  *
3080  * If we have specified it via mount option, then
3081  * use the mount option value. If the value specified at mount time is
3082  * greater than the blocks per group use the super block value.
3083  * If the super block value is greater than blocks per group return 0.
3084  * Allocator needs it be less than blocks per group.
3085  *
3086  */
3087 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
3088 {
3089         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
3090         unsigned long stripe_width =
3091                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
3092         int ret;
3093
3094         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
3095                 ret = sbi->s_stripe;
3096         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
3097                 ret = stripe_width;
3098         else if (stride && stride <= sbi->s_blocks_per_group)
3099                 ret = stride;
3100         else
3101                 ret = 0;
3102
3103         /*
3104          * If the stripe width is 1, this makes no sense and
3105          * we set it to 0 to turn off stripe handling code.
3106          */
3107         if (ret <= 1)
3108                 ret = 0;
3109
3110         return ret;
3111 }
3112
3113 /*
3114  * Check whether this filesystem can be mounted based on
3115  * the features present and the RDONLY/RDWR mount requested.
3116  * Returns 1 if this filesystem can be mounted as requested,
3117  * 0 if it cannot be.
3118  */
3119 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
3120 {
3121         if (ext4_has_unknown_ext4_incompat_features(sb)) {
3122                 ext4_msg(sb, KERN_ERR,
3123                         "Couldn't mount because of "
3124                         "unsupported optional features (%x)",
3125                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
3126                         ~EXT4_FEATURE_INCOMPAT_SUPP));
3127                 return 0;
3128         }
3129
3130 #ifndef CONFIG_UNICODE
3131         if (ext4_has_feature_casefold(sb)) {
3132                 ext4_msg(sb, KERN_ERR,
3133                          "Filesystem with casefold feature cannot be "
3134                          "mounted without CONFIG_UNICODE");
3135                 return 0;
3136         }
3137 #endif
3138
3139         if (readonly)
3140                 return 1;
3141
3142         if (ext4_has_feature_readonly(sb)) {
3143                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
3144                 sb->s_flags |= SB_RDONLY;
3145                 return 1;
3146         }
3147
3148         /* Check that feature set is OK for a read-write mount */
3149         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
3150                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
3151                          "unsupported optional features (%x)",
3152                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
3153                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
3154                 return 0;
3155         }
3156         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
3157                 ext4_msg(sb, KERN_ERR,
3158                          "Can't support bigalloc feature without "
3159                          "extents feature\n");
3160                 return 0;
3161         }
3162
3163 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2)
3164         if (!readonly && (ext4_has_feature_quota(sb) ||
3165                           ext4_has_feature_project(sb))) {
3166                 ext4_msg(sb, KERN_ERR,
3167                          "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2");
3168                 return 0;
3169         }
3170 #endif  /* CONFIG_QUOTA */
3171         return 1;
3172 }
3173
3174 /*
3175  * This function is called once a day if we have errors logged
3176  * on the file system
3177  */
3178 static void print_daily_error_info(struct timer_list *t)
3179 {
3180         struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report);
3181         struct super_block *sb = sbi->s_sb;
3182         struct ext4_super_block *es = sbi->s_es;
3183
3184         if (es->s_error_count)
3185                 /* fsck newer than v1.41.13 is needed to clean this condition. */
3186                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
3187                          le32_to_cpu(es->s_error_count));
3188         if (es->s_first_error_time) {
3189                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d",
3190                        sb->s_id,
3191                        ext4_get_tstamp(es, s_first_error_time),
3192                        (int) sizeof(es->s_first_error_func),
3193                        es->s_first_error_func,
3194                        le32_to_cpu(es->s_first_error_line));
3195                 if (es->s_first_error_ino)
3196                         printk(KERN_CONT ": inode %u",
3197                                le32_to_cpu(es->s_first_error_ino));
3198                 if (es->s_first_error_block)
3199                         printk(KERN_CONT ": block %llu", (unsigned long long)
3200                                le64_to_cpu(es->s_first_error_block));
3201                 printk(KERN_CONT "\n");
3202         }
3203         if (es->s_last_error_time) {
3204                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d",
3205                        sb->s_id,
3206                        ext4_get_tstamp(es, s_last_error_time),
3207                        (int) sizeof(es->s_last_error_func),
3208                        es->s_last_error_func,
3209                        le32_to_cpu(es->s_last_error_line));
3210                 if (es->s_last_error_ino)
3211                         printk(KERN_CONT ": inode %u",
3212                                le32_to_cpu(es->s_last_error_ino));
3213                 if (es->s_last_error_block)
3214                         printk(KERN_CONT ": block %llu", (unsigned long long)
3215                                le64_to_cpu(es->s_last_error_block));
3216                 printk(KERN_CONT "\n");
3217         }
3218         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
3219 }
3220
3221 /* Find next suitable group and run ext4_init_inode_table */
3222 static int ext4_run_li_request(struct ext4_li_request *elr)
3223 {
3224         struct ext4_group_desc *gdp = NULL;
3225         struct super_block *sb = elr->lr_super;
3226         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3227         ext4_group_t group = elr->lr_next_group;
3228         unsigned long timeout = 0;
3229         unsigned int prefetch_ios = 0;
3230         int ret = 0;
3231
3232         if (elr->lr_mode == EXT4_LI_MODE_PREFETCH_BBITMAP) {
3233                 elr->lr_next_group = ext4_mb_prefetch(sb, group,
3234                                 EXT4_SB(sb)->s_mb_prefetch, &prefetch_ios);
3235                 if (prefetch_ios)
3236                         ext4_mb_prefetch_fini(sb, elr->lr_next_group,
3237                                               prefetch_ios);
3238                 trace_ext4_prefetch_bitmaps(sb, group, elr->lr_next_group,
3239                                             prefetch_ios);
3240                 if (group >= elr->lr_next_group) {
3241                         ret = 1;
3242                         if (elr->lr_first_not_zeroed != ngroups &&
3243                             !sb_rdonly(sb) && test_opt(sb, INIT_INODE_TABLE)) {
3244                                 elr->lr_next_group = elr->lr_first_not_zeroed;
3245                                 elr->lr_mode = EXT4_LI_MODE_ITABLE;
3246                                 ret = 0;
3247                         }
3248                 }
3249                 return ret;
3250         }
3251
3252         for (; group < ngroups; group++) {
3253                 gdp = ext4_get_group_desc(sb, group, NULL);
3254                 if (!gdp) {
3255                         ret = 1;
3256                         break;
3257                 }
3258
3259                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3260                         break;
3261         }
3262
3263         if (group >= ngroups)
3264                 ret = 1;
3265
3266         if (!ret) {
3267                 timeout = jiffies;
3268                 ret = ext4_init_inode_table(sb, group,
3269                                             elr->lr_timeout ? 0 : 1);
3270                 trace_ext4_lazy_itable_init(sb, group);
3271                 if (elr->lr_timeout == 0) {
3272                         timeout = (jiffies - timeout) *
3273                                 EXT4_SB(elr->lr_super)->s_li_wait_mult;
3274                         elr->lr_timeout = timeout;
3275                 }
3276                 elr->lr_next_sched = jiffies + elr->lr_timeout;
3277                 elr->lr_next_group = group + 1;
3278         }
3279         return ret;
3280 }
3281
3282 /*
3283  * Remove lr_request from the list_request and free the
3284  * request structure. Should be called with li_list_mtx held
3285  */
3286 static void ext4_remove_li_request(struct ext4_li_request *elr)
3287 {
3288         if (!elr)
3289                 return;
3290
3291         list_del(&elr->lr_request);
3292         EXT4_SB(elr->lr_super)->s_li_request = NULL;
3293         kfree(elr);
3294 }
3295
3296 static void ext4_unregister_li_request(struct super_block *sb)
3297 {
3298         mutex_lock(&ext4_li_mtx);
3299         if (!ext4_li_info) {
3300                 mutex_unlock(&ext4_li_mtx);
3301                 return;
3302         }
3303
3304         mutex_lock(&ext4_li_info->li_list_mtx);
3305         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
3306         mutex_unlock(&ext4_li_info->li_list_mtx);
3307         mutex_unlock(&ext4_li_mtx);
3308 }
3309
3310 static struct task_struct *ext4_lazyinit_task;
3311
3312 /*
3313  * This is the function where ext4lazyinit thread lives. It walks
3314  * through the request list searching for next scheduled filesystem.
3315  * When such a fs is found, run the lazy initialization request
3316  * (ext4_rn_li_request) and keep track of the time spend in this
3317  * function. Based on that time we compute next schedule time of
3318  * the request. When walking through the list is complete, compute
3319  * next waking time and put itself into sleep.
3320  */
3321 static int ext4_lazyinit_thread(void *arg)
3322 {
3323         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
3324         struct list_head *pos, *n;
3325         struct ext4_li_request *elr;
3326         unsigned long next_wakeup, cur;
3327
3328         BUG_ON(NULL == eli);
3329
3330 cont_thread:
3331         while (true) {
3332                 next_wakeup = MAX_JIFFY_OFFSET;
3333
3334                 mutex_lock(&eli->li_list_mtx);
3335                 if (list_empty(&eli->li_request_list)) {
3336                         mutex_unlock(&eli->li_list_mtx);
3337                         goto exit_thread;
3338                 }
3339                 list_for_each_safe(pos, n, &eli->li_request_list) {
3340                         int err = 0;
3341                         int progress = 0;
3342                         elr = list_entry(pos, struct ext4_li_request,
3343                                          lr_request);
3344
3345                         if (time_before(jiffies, elr->lr_next_sched)) {
3346                                 if (time_before(elr->lr_next_sched, next_wakeup))
3347                                         next_wakeup = elr->lr_next_sched;
3348                                 continue;
3349                         }
3350                         if (down_read_trylock(&elr->lr_super->s_umount)) {
3351                                 if (sb_start_write_trylock(elr->lr_super)) {
3352                                         progress = 1;
3353                                         /*
3354                                          * We hold sb->s_umount, sb can not
3355                                          * be removed from the list, it is
3356                                          * now safe to drop li_list_mtx
3357                                          */
3358                                         mutex_unlock(&eli->li_list_mtx);
3359                                         err = ext4_run_li_request(elr);
3360                                         sb_end_write(elr->lr_super);
3361                                         mutex_lock(&eli->li_list_mtx);
3362                                         n = pos->next;
3363                                 }
3364                                 up_read((&elr->lr_super->s_umount));
3365                         }
3366                         /* error, remove the lazy_init job */
3367                         if (err) {
3368                                 ext4_remove_li_request(elr);
3369                                 continue;
3370                         }
3371                         if (!progress) {
3372                                 elr->lr_next_sched = jiffies +
3373                                         (prandom_u32()
3374                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3375                         }
3376                         if (time_before(elr->lr_next_sched, next_wakeup))
3377                                 next_wakeup = elr->lr_next_sched;
3378                 }
3379                 mutex_unlock(&eli->li_list_mtx);
3380
3381                 try_to_freeze();
3382
3383                 cur = jiffies;
3384                 if ((time_after_eq(cur, next_wakeup)) ||
3385                     (MAX_JIFFY_OFFSET == next_wakeup)) {
3386                         cond_resched();
3387                         continue;
3388                 }
3389
3390                 schedule_timeout_interruptible(next_wakeup - cur);
3391
3392                 if (kthread_should_stop()) {
3393                         ext4_clear_request_list();
3394                         goto exit_thread;
3395                 }
3396         }
3397
3398 exit_thread:
3399         /*
3400          * It looks like the request list is empty, but we need
3401          * to check it under the li_list_mtx lock, to prevent any
3402          * additions into it, and of course we should lock ext4_li_mtx
3403          * to atomically free the list and ext4_li_info, because at
3404          * this point another ext4 filesystem could be registering
3405          * new one.
3406          */
3407         mutex_lock(&ext4_li_mtx);
3408         mutex_lock(&eli->li_list_mtx);
3409         if (!list_empty(&eli->li_request_list)) {
3410                 mutex_unlock(&eli->li_list_mtx);
3411                 mutex_unlock(&ext4_li_mtx);
3412                 goto cont_thread;
3413         }
3414         mutex_unlock(&eli->li_list_mtx);
3415         kfree(ext4_li_info);
3416         ext4_li_info = NULL;
3417         mutex_unlock(&ext4_li_mtx);
3418
3419         return 0;
3420 }
3421
3422 static void ext4_clear_request_list(void)
3423 {
3424         struct list_head *pos, *n;
3425         struct ext4_li_request *elr;
3426
3427         mutex_lock(&ext4_li_info->li_list_mtx);
3428         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3429                 elr = list_entry(pos, struct ext4_li_request,
3430                                  lr_request);
3431                 ext4_remove_li_request(elr);
3432         }
3433         mutex_unlock(&ext4_li_info->li_list_mtx);
3434 }
3435
3436 static int ext4_run_lazyinit_thread(void)
3437 {
3438         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3439                                          ext4_li_info, "ext4lazyinit");
3440         if (IS_ERR(ext4_lazyinit_task)) {
3441                 int err = PTR_ERR(ext4_lazyinit_task);
3442                 ext4_clear_request_list();
3443                 kfree(ext4_li_info);
3444                 ext4_li_info = NULL;
3445                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3446                                  "initialization thread\n",
3447                                  err);
3448                 return err;
3449         }
3450         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3451         return 0;
3452 }
3453
3454 /*
3455  * Check whether it make sense to run itable init. thread or not.
3456  * If there is at least one uninitialized inode table, return
3457  * corresponding group number, else the loop goes through all
3458  * groups and return total number of groups.
3459  */
3460 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3461 {
3462         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3463         struct ext4_group_desc *gdp = NULL;
3464
3465         if (!ext4_has_group_desc_csum(sb))
3466                 return ngroups;
3467
3468         for (group = 0; group < ngroups; group++) {
3469                 gdp = ext4_get_group_desc(sb, group, NULL);
3470                 if (!gdp)
3471                         continue;
3472
3473                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3474                         break;
3475         }
3476
3477         return group;
3478 }
3479
3480 static int ext4_li_info_new(void)
3481 {
3482         struct ext4_lazy_init *eli = NULL;
3483
3484         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3485         if (!eli)
3486                 return -ENOMEM;
3487
3488         INIT_LIST_HEAD(&eli->li_request_list);
3489         mutex_init(&eli->li_list_mtx);
3490
3491         eli->li_state |= EXT4_LAZYINIT_QUIT;
3492
3493         ext4_li_info = eli;
3494
3495         return 0;
3496 }
3497
3498 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3499                                             ext4_group_t start)
3500 {
3501         struct ext4_li_request *elr;
3502
3503         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3504         if (!elr)
3505                 return NULL;
3506
3507         elr->lr_super = sb;
3508         elr->lr_first_not_zeroed = start;
3509         if (test_opt(sb, PREFETCH_BLOCK_BITMAPS))
3510                 elr->lr_mode = EXT4_LI_MODE_PREFETCH_BBITMAP;
3511         else {
3512                 elr->lr_mode = EXT4_LI_MODE_ITABLE;
3513                 elr->lr_next_group = start;
3514         }
3515
3516         /*
3517          * Randomize first schedule time of the request to
3518          * spread the inode table initialization requests
3519          * better.
3520          */
3521         elr->lr_next_sched = jiffies + (prandom_u32() %
3522                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3523         return elr;
3524 }
3525
3526 int ext4_register_li_request(struct super_block *sb,
3527                              ext4_group_t first_not_zeroed)
3528 {
3529         struct ext4_sb_info *sbi = EXT4_SB(sb);
3530         struct ext4_li_request *elr = NULL;
3531         ext4_group_t ngroups = sbi->s_groups_count;
3532         int ret = 0;
3533
3534         mutex_lock(&ext4_li_mtx);
3535         if (sbi->s_li_request != NULL) {
3536                 /*
3537                  * Reset timeout so it can be computed again, because
3538                  * s_li_wait_mult might have changed.
3539                  */
3540                 sbi->s_li_request->lr_timeout = 0;
3541                 goto out;
3542         }
3543
3544         if (!test_opt(sb, PREFETCH_BLOCK_BITMAPS) &&
3545             (first_not_zeroed == ngroups || sb_rdonly(sb) ||
3546              !test_opt(sb, INIT_INODE_TABLE)))
3547                 goto out;
3548
3549         elr = ext4_li_request_new(sb, first_not_zeroed);
3550         if (!elr) {
3551                 ret = -ENOMEM;
3552                 goto out;
3553         }
3554
3555         if (NULL == ext4_li_info) {
3556                 ret = ext4_li_info_new();
3557                 if (ret)
3558                         goto out;
3559         }
3560
3561         mutex_lock(&ext4_li_info->li_list_mtx);
3562         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3563         mutex_unlock(&ext4_li_info->li_list_mtx);
3564
3565         sbi->s_li_request = elr;
3566         /*
3567          * set elr to NULL here since it has been inserted to
3568          * the request_list and the removal and free of it is
3569          * handled by ext4_clear_request_list from now on.
3570          */
3571         elr = NULL;
3572
3573         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3574                 ret = ext4_run_lazyinit_thread();
3575                 if (ret)
3576                         goto out;
3577         }
3578 out:
3579         mutex_unlock(&ext4_li_mtx);
3580         if (ret)
3581                 kfree(elr);
3582         return ret;
3583 }
3584
3585 /*
3586  * We do not need to lock anything since this is called on
3587  * module unload.
3588  */
3589 static void ext4_destroy_lazyinit_thread(void)
3590 {
3591         /*
3592          * If thread exited earlier
3593          * there's nothing to be done.
3594          */
3595         if (!ext4_li_info || !ext4_lazyinit_task)
3596                 return;
3597
3598         kthread_stop(ext4_lazyinit_task);
3599 }
3600
3601 static int set_journal_csum_feature_set(struct super_block *sb)
3602 {
3603         int ret = 1;
3604         int compat, incompat;
3605         struct ext4_sb_info *sbi = EXT4_SB(sb);
3606
3607         if (ext4_has_metadata_csum(sb)) {
3608                 /* journal checksum v3 */
3609                 compat = 0;
3610                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3611         } else {
3612                 /* journal checksum v1 */
3613                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3614                 incompat = 0;
3615         }
3616
3617         jbd2_journal_clear_features(sbi->s_journal,
3618                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3619                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3620                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3621         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3622                 ret = jbd2_journal_set_features(sbi->s_journal,
3623                                 compat, 0,
3624                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3625                                 incompat);
3626         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3627                 ret = jbd2_journal_set_features(sbi->s_journal,
3628                                 compat, 0,
3629                                 incompat);
3630                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3631                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3632         } else {
3633                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3634                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3635         }
3636
3637         return ret;
3638 }
3639
3640 /*
3641  * Note: calculating the overhead so we can be compatible with
3642  * historical BSD practice is quite difficult in the face of
3643  * clusters/bigalloc.  This is because multiple metadata blocks from
3644  * different block group can end up in the same allocation cluster.
3645  * Calculating the exact overhead in the face of clustered allocation
3646  * requires either O(all block bitmaps) in memory or O(number of block
3647  * groups**2) in time.  We will still calculate the superblock for
3648  * older file systems --- and if we come across with a bigalloc file
3649  * system with zero in s_overhead_clusters the estimate will be close to
3650  * correct especially for very large cluster sizes --- but for newer
3651  * file systems, it's better to calculate this figure once at mkfs
3652  * time, and store it in the superblock.  If the superblock value is
3653  * present (even for non-bigalloc file systems), we will use it.
3654  */
3655 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3656                           char *buf)
3657 {
3658         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3659         struct ext4_group_desc  *gdp;
3660         ext4_fsblk_t            first_block, last_block, b;
3661         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3662         int                     s, j, count = 0;
3663
3664         if (!ext4_has_feature_bigalloc(sb))
3665                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3666                         sbi->s_itb_per_group + 2);
3667
3668         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3669                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3670         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3671         for (i = 0; i < ngroups; i++) {
3672                 gdp = ext4_get_group_desc(sb, i, NULL);
3673                 b = ext4_block_bitmap(sb, gdp);
3674                 if (b >= first_block && b <= last_block) {
3675                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3676                         count++;
3677                 }
3678                 b = ext4_inode_bitmap(sb, gdp);
3679                 if (b >= first_block && b <= last_block) {
3680                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3681                         count++;
3682                 }
3683                 b = ext4_inode_table(sb, gdp);
3684                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3685                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3686                                 int c = EXT4_B2C(sbi, b - first_block);
3687                                 ext4_set_bit(c, buf);
3688                                 count++;
3689                         }
3690                 if (i != grp)
3691                         continue;
3692                 s = 0;
3693                 if (ext4_bg_has_super(sb, grp)) {
3694                         ext4_set_bit(s++, buf);
3695                         count++;
3696                 }
3697                 j = ext4_bg_num_gdb(sb, grp);
3698                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3699                         ext4_error(sb, "Invalid number of block group "
3700                                    "descriptor blocks: %d", j);
3701                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3702                 }
3703                 count += j;
3704                 for (; j > 0; j--)
3705                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3706         }
3707         if (!count)
3708                 return 0;
3709         return EXT4_CLUSTERS_PER_GROUP(sb) -
3710                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3711 }
3712
3713 /*
3714  * Compute the overhead and stash it in sbi->s_overhead
3715  */
3716 int ext4_calculate_overhead(struct super_block *sb)
3717 {
3718         struct ext4_sb_info *sbi = EXT4_SB(sb);
3719         struct ext4_super_block *es = sbi->s_es;
3720         struct inode *j_inode;
3721         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3722         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3723         ext4_fsblk_t overhead = 0;
3724         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3725
3726         if (!buf)
3727                 return -ENOMEM;
3728
3729         /*
3730          * Compute the overhead (FS structures).  This is constant
3731          * for a given filesystem unless the number of block groups
3732          * changes so we cache the previous value until it does.
3733          */
3734
3735         /*
3736          * All of the blocks before first_data_block are overhead
3737          */
3738         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3739
3740         /*
3741          * Add the overhead found in each block group
3742          */
3743         for (i = 0; i < ngroups; i++) {
3744                 int blks;
3745
3746                 blks = count_overhead(sb, i, buf);
3747                 overhead += blks;
3748                 if (blks)
3749                         memset(buf, 0, PAGE_SIZE);
3750                 cond_resched();
3751         }
3752
3753         /*
3754          * Add the internal journal blocks whether the journal has been
3755          * loaded or not
3756          */
3757         if (sbi->s_journal && !sbi->journal_bdev)
3758                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3759         else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) {
3760                 /* j_inum for internal journal is non-zero */
3761                 j_inode = ext4_get_journal_inode(sb, j_inum);
3762                 if (j_inode) {
3763                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3764                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
3765                         iput(j_inode);
3766                 } else {
3767                         ext4_msg(sb, KERN_ERR, "can't get journal size");
3768                 }
3769         }
3770         sbi->s_overhead = overhead;
3771         smp_wmb();
3772         free_page((unsigned long) buf);
3773         return 0;
3774 }
3775
3776 static void ext4_set_resv_clusters(struct super_block *sb)
3777 {
3778         ext4_fsblk_t resv_clusters;
3779         struct ext4_sb_info *sbi = EXT4_SB(sb);
3780
3781         /*
3782          * There's no need to reserve anything when we aren't using extents.
3783          * The space estimates are exact, there are no unwritten extents,
3784          * hole punching doesn't need new metadata... This is needed especially
3785          * to keep ext2/3 backward compatibility.
3786          */
3787         if (!ext4_has_feature_extents(sb))
3788                 return;
3789         /*
3790          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3791          * This should cover the situations where we can not afford to run
3792          * out of space like for example punch hole, or converting
3793          * unwritten extents in delalloc path. In most cases such
3794          * allocation would require 1, or 2 blocks, higher numbers are
3795          * very rare.
3796          */
3797         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3798                          sbi->s_cluster_bits);
3799
3800         do_div(resv_clusters, 50);
3801         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3802
3803         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3804 }
3805
3806 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3807 {
3808         struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev);
3809         char *orig_data = kstrdup(data, GFP_KERNEL);
3810         struct buffer_head *bh, **group_desc;
3811         struct ext4_super_block *es = NULL;
3812         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3813         struct flex_groups **flex_groups;
3814         ext4_fsblk_t block;
3815         ext4_fsblk_t sb_block = get_sb_block(&data);
3816         ext4_fsblk_t logical_sb_block;
3817         unsigned long offset = 0;
3818         unsigned long journal_devnum = 0;
3819         unsigned long def_mount_opts;
3820         struct inode *root;
3821         const char *descr;
3822         int ret = -ENOMEM;
3823         int blocksize, clustersize;
3824         unsigned int db_count;
3825         unsigned int i;
3826         int needs_recovery, has_huge_files;
3827         __u64 blocks_count;
3828         int err = 0;
3829         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3830         ext4_group_t first_not_zeroed;
3831
3832         if ((data && !orig_data) || !sbi)
3833                 goto out_free_base;
3834
3835         sbi->s_daxdev = dax_dev;
3836         sbi->s_blockgroup_lock =
3837                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3838         if (!sbi->s_blockgroup_lock)
3839                 goto out_free_base;
3840
3841         sb->s_fs_info = sbi;
3842         sbi->s_sb = sb;
3843         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3844         sbi->s_sb_block = sb_block;
3845         if (sb->s_bdev->bd_part)
3846                 sbi->s_sectors_written_start =
3847                         part_stat_read(sb->s_bdev->bd_part, sectors[STAT_WRITE]);
3848
3849         /* Cleanup superblock name */
3850         strreplace(sb->s_id, '/', '!');
3851
3852         /* -EINVAL is default */
3853         ret = -EINVAL;
3854         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3855         if (!blocksize) {
3856                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3857                 goto out_fail;
3858         }
3859
3860         /*
3861          * The ext4 superblock will not be buffer aligned for other than 1kB
3862          * block sizes.  We need to calculate the offset from buffer start.
3863          */
3864         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3865                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3866                 offset = do_div(logical_sb_block, blocksize);
3867         } else {
3868                 logical_sb_block = sb_block;
3869         }
3870
3871         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3872                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3873                 goto out_fail;
3874         }
3875         /*
3876          * Note: s_es must be initialized as soon as possible because
3877          *       some ext4 macro-instructions depend on its value
3878          */
3879         es = (struct ext4_super_block *) (bh->b_data + offset);
3880         sbi->s_es = es;
3881         sb->s_magic = le16_to_cpu(es->s_magic);
3882         if (sb->s_magic != EXT4_SUPER_MAGIC)
3883                 goto cantfind_ext4;
3884         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3885
3886         /* Warn if metadata_csum and gdt_csum are both set. */
3887         if (ext4_has_feature_metadata_csum(sb) &&
3888             ext4_has_feature_gdt_csum(sb))
3889                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3890                              "redundant flags; please run fsck.");
3891
3892         /* Check for a known checksum algorithm */
3893         if (!ext4_verify_csum_type(sb, es)) {
3894                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3895                          "unknown checksum algorithm.");
3896                 silent = 1;
3897                 goto cantfind_ext4;
3898         }
3899
3900         /* Load the checksum driver */
3901         sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3902         if (IS_ERR(sbi->s_chksum_driver)) {
3903                 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3904                 ret = PTR_ERR(sbi->s_chksum_driver);
3905                 sbi->s_chksum_driver = NULL;
3906                 goto failed_mount;
3907         }
3908
3909         /* Check superblock checksum */
3910         if (!ext4_superblock_csum_verify(sb, es)) {
3911                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3912                          "invalid superblock checksum.  Run e2fsck?");
3913                 silent = 1;
3914                 ret = -EFSBADCRC;
3915                 goto cantfind_ext4;
3916         }
3917
3918         /* Precompute checksum seed for all metadata */
3919         if (ext4_has_feature_csum_seed(sb))
3920                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3921         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
3922                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3923                                                sizeof(es->s_uuid));
3924
3925         /* Set defaults before we parse the mount options */
3926         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3927         set_opt(sb, INIT_INODE_TABLE);
3928         if (def_mount_opts & EXT4_DEFM_DEBUG)
3929                 set_opt(sb, DEBUG);
3930         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3931                 set_opt(sb, GRPID);
3932         if (def_mount_opts & EXT4_DEFM_UID16)
3933                 set_opt(sb, NO_UID32);
3934         /* xattr user namespace & acls are now defaulted on */
3935         set_opt(sb, XATTR_USER);
3936 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3937         set_opt(sb, POSIX_ACL);
3938 #endif
3939         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3940         if (ext4_has_metadata_csum(sb))
3941                 set_opt(sb, JOURNAL_CHECKSUM);
3942
3943         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3944                 set_opt(sb, JOURNAL_DATA);
3945         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3946                 set_opt(sb, ORDERED_DATA);
3947         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3948                 set_opt(sb, WRITEBACK_DATA);
3949
3950         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3951                 set_opt(sb, ERRORS_PANIC);
3952         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3953                 set_opt(sb, ERRORS_CONT);
3954         else
3955                 set_opt(sb, ERRORS_RO);
3956         /* block_validity enabled by default; disable with noblock_validity */
3957         set_opt(sb, BLOCK_VALIDITY);
3958         if (def_mount_opts & EXT4_DEFM_DISCARD)
3959                 set_opt(sb, DISCARD);
3960
3961         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3962         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3963         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3964         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3965         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3966
3967         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3968                 set_opt(sb, BARRIER);
3969
3970         /*
3971          * enable delayed allocation by default
3972          * Use -o nodelalloc to turn it off
3973          */
3974         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3975             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3976                 set_opt(sb, DELALLOC);
3977
3978         /*
3979          * set default s_li_wait_mult for lazyinit, for the case there is
3980          * no mount option specified.
3981          */
3982         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3983
3984         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3985
3986         if (blocksize == PAGE_SIZE)
3987                 set_opt(sb, DIOREAD_NOLOCK);
3988
3989         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3990             blocksize > EXT4_MAX_BLOCK_SIZE) {
3991                 ext4_msg(sb, KERN_ERR,
3992                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3993                          blocksize, le32_to_cpu(es->s_log_block_size));
3994                 goto failed_mount;
3995         }
3996
3997         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3998                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3999                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
4000         } else {
4001                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
4002                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
4003                 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) {
4004                         ext4_msg(sb, KERN_ERR, "invalid first ino: %u",
4005                                  sbi->s_first_ino);
4006                         goto failed_mount;
4007                 }
4008                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
4009                     (!is_power_of_2(sbi->s_inode_size)) ||
4010                     (sbi->s_inode_size > blocksize)) {
4011                         ext4_msg(sb, KERN_ERR,
4012                                "unsupported inode size: %d",
4013                                sbi->s_inode_size);
4014                         ext4_msg(sb, KERN_ERR, "blocksize: %d", blocksize);
4015                         goto failed_mount;
4016                 }
4017                 /*
4018                  * i_atime_extra is the last extra field available for
4019                  * [acm]times in struct ext4_inode. Checking for that
4020                  * field should suffice to ensure we have extra space
4021                  * for all three.
4022                  */
4023                 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) +
4024                         sizeof(((struct ext4_inode *)0)->i_atime_extra)) {
4025                         sb->s_time_gran = 1;
4026                         sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX;
4027                 } else {
4028                         sb->s_time_gran = NSEC_PER_SEC;
4029                         sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX;
4030                 }
4031                 sb->s_time_min = EXT4_TIMESTAMP_MIN;
4032         }
4033         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
4034                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4035                         EXT4_GOOD_OLD_INODE_SIZE;
4036                 if (ext4_has_feature_extra_isize(sb)) {
4037                         unsigned v, max = (sbi->s_inode_size -
4038                                            EXT4_GOOD_OLD_INODE_SIZE);
4039
4040                         v = le16_to_cpu(es->s_want_extra_isize);
4041                         if (v > max) {
4042                                 ext4_msg(sb, KERN_ERR,
4043                                          "bad s_want_extra_isize: %d", v);
4044                                 goto failed_mount;
4045                         }
4046                         if (sbi->s_want_extra_isize < v)
4047                                 sbi->s_want_extra_isize = v;
4048
4049                         v = le16_to_cpu(es->s_min_extra_isize);
4050                         if (v > max) {
4051                                 ext4_msg(sb, KERN_ERR,
4052                                          "bad s_min_extra_isize: %d", v);
4053                                 goto failed_mount;
4054                         }
4055                         if (sbi->s_want_extra_isize < v)
4056                                 sbi->s_want_extra_isize = v;
4057                 }
4058         }
4059
4060         if (sbi->s_es->s_mount_opts[0]) {
4061                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
4062                                               sizeof(sbi->s_es->s_mount_opts),
4063                                               GFP_KERNEL);
4064                 if (!s_mount_opts)
4065                         goto failed_mount;
4066                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
4067                                    &journal_ioprio, 0)) {
4068                         ext4_msg(sb, KERN_WARNING,
4069                                  "failed to parse options in superblock: %s",
4070                                  s_mount_opts);
4071                 }
4072                 kfree(s_mount_opts);
4073         }
4074         sbi->s_def_mount_opt = sbi->s_mount_opt;
4075         if (!parse_options((char *) data, sb, &journal_devnum,
4076                            &journal_ioprio, 0))
4077                 goto failed_mount;
4078
4079 #ifdef CONFIG_UNICODE
4080         if (ext4_has_feature_casefold(sb) && !sbi->s_encoding) {
4081                 const struct ext4_sb_encodings *encoding_info;
4082                 struct unicode_map *encoding;
4083                 __u16 encoding_flags;
4084
4085                 if (ext4_has_feature_encrypt(sb)) {
4086                         ext4_msg(sb, KERN_ERR,
4087                                  "Can't mount with encoding and encryption");
4088                         goto failed_mount;
4089                 }
4090
4091                 if (ext4_sb_read_encoding(es, &encoding_info,
4092                                           &encoding_flags)) {
4093                         ext4_msg(sb, KERN_ERR,
4094                                  "Encoding requested by superblock is unknown");
4095                         goto failed_mount;
4096                 }
4097
4098                 encoding = utf8_load(encoding_info->version);
4099                 if (IS_ERR(encoding)) {
4100                         ext4_msg(sb, KERN_ERR,
4101                                  "can't mount with superblock charset: %s-%s "
4102                                  "not supported by the kernel. flags: 0x%x.",
4103                                  encoding_info->name, encoding_info->version,
4104                                  encoding_flags);
4105                         goto failed_mount;
4106                 }
4107                 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: "
4108                          "%s-%s with flags 0x%hx", encoding_info->name,
4109                          encoding_info->version?:"\b", encoding_flags);
4110
4111                 sbi->s_encoding = encoding;
4112                 sbi->s_encoding_flags = encoding_flags;
4113         }
4114 #endif
4115
4116         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4117                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with data=journal disables delayed allocation, dioread_nolock, and O_DIRECT support!\n");
4118                 /* can't mount with both data=journal and dioread_nolock. */
4119                 clear_opt(sb, DIOREAD_NOLOCK);
4120                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4121                         ext4_msg(sb, KERN_ERR, "can't mount with "
4122                                  "both data=journal and delalloc");
4123                         goto failed_mount;
4124                 }
4125                 if (test_opt(sb, DAX_ALWAYS)) {
4126                         ext4_msg(sb, KERN_ERR, "can't mount with "
4127                                  "both data=journal and dax");
4128                         goto failed_mount;
4129                 }
4130                 if (ext4_has_feature_encrypt(sb)) {
4131                         ext4_msg(sb, KERN_WARNING,
4132                                  "encrypted files will use data=ordered "
4133                                  "instead of data journaling mode");
4134                 }
4135                 if (test_opt(sb, DELALLOC))
4136                         clear_opt(sb, DELALLOC);
4137         } else {
4138                 sb->s_iflags |= SB_I_CGROUPWB;
4139         }
4140
4141         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
4142                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
4143
4144         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
4145             (ext4_has_compat_features(sb) ||
4146              ext4_has_ro_compat_features(sb) ||
4147              ext4_has_incompat_features(sb)))
4148                 ext4_msg(sb, KERN_WARNING,
4149                        "feature flags set on rev 0 fs, "
4150                        "running e2fsck is recommended");
4151
4152         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
4153                 set_opt2(sb, HURD_COMPAT);
4154                 if (ext4_has_feature_64bit(sb)) {
4155                         ext4_msg(sb, KERN_ERR,
4156                                  "The Hurd can't support 64-bit file systems");
4157                         goto failed_mount;
4158                 }
4159
4160                 /*
4161                  * ea_inode feature uses l_i_version field which is not
4162                  * available in HURD_COMPAT mode.
4163                  */
4164                 if (ext4_has_feature_ea_inode(sb)) {
4165                         ext4_msg(sb, KERN_ERR,
4166                                  "ea_inode feature is not supported for Hurd");
4167                         goto failed_mount;
4168                 }
4169         }
4170
4171         if (IS_EXT2_SB(sb)) {
4172                 if (ext2_feature_set_ok(sb))
4173                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
4174                                  "using the ext4 subsystem");
4175                 else {
4176                         /*
4177                          * If we're probing be silent, if this looks like
4178                          * it's actually an ext[34] filesystem.
4179                          */
4180                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4181                                 goto failed_mount;
4182                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
4183                                  "to feature incompatibilities");
4184                         goto failed_mount;
4185                 }
4186         }
4187
4188         if (IS_EXT3_SB(sb)) {
4189                 if (ext3_feature_set_ok(sb))
4190                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
4191                                  "using the ext4 subsystem");
4192                 else {
4193                         /*
4194                          * If we're probing be silent, if this looks like
4195                          * it's actually an ext4 filesystem.
4196                          */
4197                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4198                                 goto failed_mount;
4199                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
4200                                  "to feature incompatibilities");
4201                         goto failed_mount;
4202                 }
4203         }
4204
4205         /*
4206          * Check feature flags regardless of the revision level, since we
4207          * previously didn't change the revision level when setting the flags,
4208          * so there is a chance incompat flags are set on a rev 0 filesystem.
4209          */
4210         if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
4211                 goto failed_mount;
4212
4213         if (le32_to_cpu(es->s_log_block_size) >
4214             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4215                 ext4_msg(sb, KERN_ERR,
4216                          "Invalid log block size: %u",
4217                          le32_to_cpu(es->s_log_block_size));
4218                 goto failed_mount;
4219         }
4220         if (le32_to_cpu(es->s_log_cluster_size) >
4221             (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4222                 ext4_msg(sb, KERN_ERR,
4223                          "Invalid log cluster size: %u",
4224                          le32_to_cpu(es->s_log_cluster_size));
4225                 goto failed_mount;
4226         }
4227
4228         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
4229                 ext4_msg(sb, KERN_ERR,
4230                          "Number of reserved GDT blocks insanely large: %d",
4231                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
4232                 goto failed_mount;
4233         }
4234
4235         if (bdev_dax_supported(sb->s_bdev, blocksize))
4236                 set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags);
4237
4238         if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) {
4239                 if (ext4_has_feature_inline_data(sb)) {
4240                         ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
4241                                         " that may contain inline data");
4242                         goto failed_mount;
4243                 }
4244                 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) {
4245                         ext4_msg(sb, KERN_ERR,
4246                                 "DAX unsupported by block device.");
4247                         goto failed_mount;
4248                 }
4249         }
4250
4251         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
4252                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
4253                          es->s_encryption_level);
4254                 goto failed_mount;
4255         }
4256
4257         if (sb->s_blocksize != blocksize) {
4258                 /* Validate the filesystem blocksize */
4259                 if (!sb_set_blocksize(sb, blocksize)) {
4260                         ext4_msg(sb, KERN_ERR, "bad block size %d",
4261                                         blocksize);
4262                         goto failed_mount;
4263                 }
4264
4265                 brelse(bh);
4266                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
4267                 offset = do_div(logical_sb_block, blocksize);
4268                 bh = sb_bread_unmovable(sb, logical_sb_block);
4269                 if (!bh) {
4270                         ext4_msg(sb, KERN_ERR,
4271                                "Can't read superblock on 2nd try");
4272                         goto failed_mount;
4273                 }
4274                 es = (struct ext4_super_block *)(bh->b_data + offset);
4275                 sbi->s_es = es;
4276                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
4277                         ext4_msg(sb, KERN_ERR,
4278                                "Magic mismatch, very weird!");
4279                         goto failed_mount;
4280                 }
4281         }
4282
4283         has_huge_files = ext4_has_feature_huge_file(sb);
4284         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
4285                                                       has_huge_files);
4286         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
4287
4288         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
4289         if (ext4_has_feature_64bit(sb)) {
4290                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
4291                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
4292                     !is_power_of_2(sbi->s_desc_size)) {
4293                         ext4_msg(sb, KERN_ERR,
4294                                "unsupported descriptor size %lu",
4295                                sbi->s_desc_size);
4296                         goto failed_mount;
4297                 }
4298         } else
4299                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
4300
4301         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
4302         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
4303
4304         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
4305         if (sbi->s_inodes_per_block == 0)
4306                 goto cantfind_ext4;
4307         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
4308             sbi->s_inodes_per_group > blocksize * 8) {
4309                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
4310                          sbi->s_inodes_per_group);
4311                 goto failed_mount;
4312         }
4313         sbi->s_itb_per_group = sbi->s_inodes_per_group /
4314                                         sbi->s_inodes_per_block;
4315         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
4316         sbi->s_sbh = bh;
4317         sbi->s_mount_state = le16_to_cpu(es->s_state);
4318         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
4319         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
4320
4321         for (i = 0; i < 4; i++)
4322                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
4323         sbi->s_def_hash_version = es->s_def_hash_version;
4324         if (ext4_has_feature_dir_index(sb)) {
4325                 i = le32_to_cpu(es->s_flags);
4326                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
4327                         sbi->s_hash_unsigned = 3;
4328                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
4329 #ifdef __CHAR_UNSIGNED__
4330                         if (!sb_rdonly(sb))
4331                                 es->s_flags |=
4332                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
4333                         sbi->s_hash_unsigned = 3;
4334 #else
4335                         if (!sb_rdonly(sb))
4336                                 es->s_flags |=
4337                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
4338 #endif
4339                 }
4340         }
4341
4342         /* Handle clustersize */
4343         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
4344         if (ext4_has_feature_bigalloc(sb)) {
4345                 if (clustersize < blocksize) {
4346                         ext4_msg(sb, KERN_ERR,
4347                                  "cluster size (%d) smaller than "
4348                                  "block size (%d)", clustersize, blocksize);
4349                         goto failed_mount;
4350                 }
4351                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
4352                         le32_to_cpu(es->s_log_block_size);
4353                 sbi->s_clusters_per_group =
4354                         le32_to_cpu(es->s_clusters_per_group);
4355                 if (sbi->s_clusters_per_group > blocksize * 8) {
4356                         ext4_msg(sb, KERN_ERR,
4357                                  "#clusters per group too big: %lu",
4358                                  sbi->s_clusters_per_group);
4359                         goto failed_mount;
4360                 }
4361                 if (sbi->s_blocks_per_group !=
4362                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
4363                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
4364                                  "clusters per group (%lu) inconsistent",
4365                                  sbi->s_blocks_per_group,
4366                                  sbi->s_clusters_per_group);
4367                         goto failed_mount;
4368                 }
4369         } else {
4370                 if (clustersize != blocksize) {
4371                         ext4_msg(sb, KERN_ERR,
4372                                  "fragment/cluster size (%d) != "
4373                                  "block size (%d)", clustersize, blocksize);
4374                         goto failed_mount;
4375                 }
4376                 if (sbi->s_blocks_per_group > blocksize * 8) {
4377                         ext4_msg(sb, KERN_ERR,
4378                                  "#blocks per group too big: %lu",
4379                                  sbi->s_blocks_per_group);
4380                         goto failed_mount;
4381                 }
4382                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
4383                 sbi->s_cluster_bits = 0;
4384         }
4385         sbi->s_cluster_ratio = clustersize / blocksize;
4386
4387         /* Do we have standard group size of clustersize * 8 blocks ? */
4388         if (sbi->s_blocks_per_group == clustersize << 3)
4389                 set_opt2(sb, STD_GROUP_SIZE);
4390
4391         /*
4392          * Test whether we have more sectors than will fit in sector_t,
4393          * and whether the max offset is addressable by the page cache.
4394          */
4395         err = generic_check_addressable(sb->s_blocksize_bits,
4396                                         ext4_blocks_count(es));
4397         if (err) {
4398                 ext4_msg(sb, KERN_ERR, "filesystem"
4399                          " too large to mount safely on this system");
4400                 goto failed_mount;
4401         }
4402
4403         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
4404                 goto cantfind_ext4;
4405
4406         /* check blocks count against device size */
4407         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
4408         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
4409                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
4410                        "exceeds size of device (%llu blocks)",
4411                        ext4_blocks_count(es), blocks_count);
4412                 goto failed_mount;
4413         }
4414
4415         /*
4416          * It makes no sense for the first data block to be beyond the end
4417          * of the filesystem.
4418          */
4419         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
4420                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4421                          "block %u is beyond end of filesystem (%llu)",
4422                          le32_to_cpu(es->s_first_data_block),
4423                          ext4_blocks_count(es));
4424                 goto failed_mount;
4425         }
4426         if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) &&
4427             (sbi->s_cluster_ratio == 1)) {
4428                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4429                          "block is 0 with a 1k block and cluster size");
4430                 goto failed_mount;
4431         }
4432
4433         blocks_count = (ext4_blocks_count(es) -
4434                         le32_to_cpu(es->s_first_data_block) +
4435                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
4436         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
4437         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
4438                 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu "
4439                        "(block count %llu, first data block %u, "
4440                        "blocks per group %lu)", blocks_count,
4441                        ext4_blocks_count(es),
4442                        le32_to_cpu(es->s_first_data_block),
4443                        EXT4_BLOCKS_PER_GROUP(sb));
4444                 goto failed_mount;
4445         }
4446         sbi->s_groups_count = blocks_count;
4447         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
4448                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
4449         if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) !=
4450             le32_to_cpu(es->s_inodes_count)) {
4451                 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu",
4452                          le32_to_cpu(es->s_inodes_count),
4453                          ((u64)sbi->s_groups_count * sbi->s_inodes_per_group));
4454                 ret = -EINVAL;
4455                 goto failed_mount;
4456         }
4457         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
4458                    EXT4_DESC_PER_BLOCK(sb);
4459         if (ext4_has_feature_meta_bg(sb)) {
4460                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
4461                         ext4_msg(sb, KERN_WARNING,
4462                                  "first meta block group too large: %u "
4463                                  "(group descriptor block count %u)",
4464                                  le32_to_cpu(es->s_first_meta_bg), db_count);
4465                         goto failed_mount;
4466                 }
4467         }
4468         rcu_assign_pointer(sbi->s_group_desc,
4469                            kvmalloc_array(db_count,
4470                                           sizeof(struct buffer_head *),
4471                                           GFP_KERNEL));
4472         if (sbi->s_group_desc == NULL) {
4473                 ext4_msg(sb, KERN_ERR, "not enough memory");
4474                 ret = -ENOMEM;
4475                 goto failed_mount;
4476         }
4477
4478         bgl_lock_init(sbi->s_blockgroup_lock);
4479
4480         /* Pre-read the descriptors into the buffer cache */
4481         for (i = 0; i < db_count; i++) {
4482                 block = descriptor_loc(sb, logical_sb_block, i);
4483                 sb_breadahead_unmovable(sb, block);
4484         }
4485
4486         for (i = 0; i < db_count; i++) {
4487                 struct buffer_head *bh;
4488
4489                 block = descriptor_loc(sb, logical_sb_block, i);
4490                 bh = sb_bread_unmovable(sb, block);
4491                 if (!bh) {
4492                         ext4_msg(sb, KERN_ERR,
4493                                "can't read group descriptor %d", i);
4494                         db_count = i;
4495                         goto failed_mount2;
4496                 }
4497                 rcu_read_lock();
4498                 rcu_dereference(sbi->s_group_desc)[i] = bh;
4499                 rcu_read_unlock();
4500         }
4501         sbi->s_gdb_count = db_count;
4502         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
4503                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
4504                 ret = -EFSCORRUPTED;
4505                 goto failed_mount2;
4506         }
4507
4508         timer_setup(&sbi->s_err_report, print_daily_error_info, 0);
4509
4510         /* Register extent status tree shrinker */
4511         if (ext4_es_register_shrinker(sbi))
4512                 goto failed_mount3;
4513
4514         sbi->s_stripe = ext4_get_stripe_size(sbi);
4515         sbi->s_extent_max_zeroout_kb = 32;
4516
4517         /*
4518          * set up enough so that it can read an inode
4519          */
4520         sb->s_op = &ext4_sops;
4521         sb->s_export_op = &ext4_export_ops;
4522         sb->s_xattr = ext4_xattr_handlers;
4523 #ifdef CONFIG_FS_ENCRYPTION
4524         sb->s_cop = &ext4_cryptops;
4525 #endif
4526 #ifdef CONFIG_FS_VERITY
4527         sb->s_vop = &ext4_verityops;
4528 #endif
4529 #ifdef CONFIG_QUOTA
4530         sb->dq_op = &ext4_quota_operations;
4531         if (ext4_has_feature_quota(sb))
4532                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4533         else
4534                 sb->s_qcop = &ext4_qctl_operations;
4535         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
4536 #endif
4537         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
4538
4539         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
4540         mutex_init(&sbi->s_orphan_lock);
4541
4542         sb->s_root = NULL;
4543
4544         needs_recovery = (es->s_last_orphan != 0 ||
4545                           ext4_has_feature_journal_needs_recovery(sb));
4546
4547         if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb))
4548                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
4549                         goto failed_mount3a;
4550
4551         /*
4552          * The first inode we look at is the journal inode.  Don't try
4553          * root first: it may be modified in the journal!
4554          */
4555         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
4556                 err = ext4_load_journal(sb, es, journal_devnum);
4557                 if (err)
4558                         goto failed_mount3a;
4559         } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
4560                    ext4_has_feature_journal_needs_recovery(sb)) {
4561                 ext4_msg(sb, KERN_ERR, "required journal recovery "
4562                        "suppressed and not mounted read-only");
4563                 goto failed_mount_wq;
4564         } else {
4565                 /* Nojournal mode, all journal mount options are illegal */
4566                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
4567                         ext4_msg(sb, KERN_ERR, "can't mount with "
4568                                  "journal_checksum, fs mounted w/o journal");
4569                         goto failed_mount_wq;
4570                 }
4571                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4572                         ext4_msg(sb, KERN_ERR, "can't mount with "
4573                                  "journal_async_commit, fs mounted w/o journal");
4574                         goto failed_mount_wq;
4575                 }
4576                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
4577                         ext4_msg(sb, KERN_ERR, "can't mount with "
4578                                  "commit=%lu, fs mounted w/o journal",
4579                                  sbi->s_commit_interval / HZ);
4580                         goto failed_mount_wq;
4581                 }
4582                 if (EXT4_MOUNT_DATA_FLAGS &
4583                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
4584                         ext4_msg(sb, KERN_ERR, "can't mount with "
4585                                  "data=, fs mounted w/o journal");
4586                         goto failed_mount_wq;
4587                 }
4588                 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM;
4589                 clear_opt(sb, JOURNAL_CHECKSUM);
4590                 clear_opt(sb, DATA_FLAGS);
4591                 sbi->s_journal = NULL;
4592                 needs_recovery = 0;
4593                 goto no_journal;
4594         }
4595
4596         if (ext4_has_feature_64bit(sb) &&
4597             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4598                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4599                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4600                 goto failed_mount_wq;
4601         }
4602
4603         if (!set_journal_csum_feature_set(sb)) {
4604                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4605                          "feature set");
4606                 goto failed_mount_wq;
4607         }
4608
4609         /* We have now updated the journal if required, so we can
4610          * validate the data journaling mode. */
4611         switch (test_opt(sb, DATA_FLAGS)) {
4612         case 0:
4613                 /* No mode set, assume a default based on the journal
4614                  * capabilities: ORDERED_DATA if the journal can
4615                  * cope, else JOURNAL_DATA
4616                  */
4617                 if (jbd2_journal_check_available_features
4618                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4619                         set_opt(sb, ORDERED_DATA);
4620                         sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
4621                 } else {
4622                         set_opt(sb, JOURNAL_DATA);
4623                         sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
4624                 }
4625                 break;
4626
4627         case EXT4_MOUNT_ORDERED_DATA:
4628         case EXT4_MOUNT_WRITEBACK_DATA:
4629                 if (!jbd2_journal_check_available_features
4630                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4631                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4632                                "requested data journaling mode");
4633                         goto failed_mount_wq;
4634                 }
4635         default:
4636                 break;
4637         }
4638
4639         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4640             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4641                 ext4_msg(sb, KERN_ERR, "can't mount with "
4642                         "journal_async_commit in data=ordered mode");
4643                 goto failed_mount_wq;
4644         }
4645
4646         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4647
4648         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4649
4650 no_journal:
4651         if (!test_opt(sb, NO_MBCACHE)) {
4652                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
4653                 if (!sbi->s_ea_block_cache) {
4654                         ext4_msg(sb, KERN_ERR,
4655                                  "Failed to create ea_block_cache");
4656                         goto failed_mount_wq;
4657                 }
4658
4659                 if (ext4_has_feature_ea_inode(sb)) {
4660                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
4661                         if (!sbi->s_ea_inode_cache) {
4662                                 ext4_msg(sb, KERN_ERR,
4663                                          "Failed to create ea_inode_cache");
4664                                 goto failed_mount_wq;
4665                         }
4666                 }
4667         }
4668
4669         if (ext4_has_feature_verity(sb) && blocksize != PAGE_SIZE) {
4670                 ext4_msg(sb, KERN_ERR, "Unsupported blocksize for fs-verity");
4671                 goto failed_mount_wq;
4672         }
4673
4674         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) &&
4675             !ext4_has_feature_encrypt(sb)) {
4676                 ext4_set_feature_encrypt(sb);
4677                 ext4_commit_super(sb, 1);
4678         }
4679
4680         /*
4681          * Get the # of file system overhead blocks from the
4682          * superblock if present.
4683          */
4684         if (es->s_overhead_clusters)
4685                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4686         else {
4687                 err = ext4_calculate_overhead(sb);
4688                 if (err)
4689                         goto failed_mount_wq;
4690         }
4691
4692         /*
4693          * The maximum number of concurrent works can be high and
4694          * concurrency isn't really necessary.  Limit it to 1.
4695          */
4696         EXT4_SB(sb)->rsv_conversion_wq =
4697                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4698         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4699                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4700                 ret = -ENOMEM;
4701                 goto failed_mount4;
4702         }
4703
4704         /*
4705          * The jbd2_journal_load will have done any necessary log recovery,
4706          * so we can safely mount the rest of the filesystem now.
4707          */
4708
4709         root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL);
4710         if (IS_ERR(root)) {
4711                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4712                 ret = PTR_ERR(root);
4713                 root = NULL;
4714                 goto failed_mount4;
4715         }
4716         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4717                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4718                 iput(root);
4719                 goto failed_mount4;
4720         }
4721
4722 #ifdef CONFIG_UNICODE
4723         if (sbi->s_encoding)
4724                 sb->s_d_op = &ext4_dentry_ops;
4725 #endif
4726
4727         sb->s_root = d_make_root(root);
4728         if (!sb->s_root) {
4729                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4730                 ret = -ENOMEM;
4731                 goto failed_mount4;
4732         }
4733
4734         ret = ext4_setup_super(sb, es, sb_rdonly(sb));
4735         if (ret == -EROFS) {
4736                 sb->s_flags |= SB_RDONLY;
4737                 ret = 0;
4738         } else if (ret)
4739                 goto failed_mount4a;
4740
4741         ext4_set_resv_clusters(sb);
4742
4743         if (test_opt(sb, BLOCK_VALIDITY)) {
4744                 err = ext4_setup_system_zone(sb);
4745                 if (err) {
4746                         ext4_msg(sb, KERN_ERR, "failed to initialize system "
4747                                  "zone (%d)", err);
4748                         goto failed_mount4a;
4749                 }
4750         }
4751
4752         ext4_ext_init(sb);
4753         err = ext4_mb_init(sb);
4754         if (err) {
4755                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4756                          err);
4757                 goto failed_mount5;
4758         }
4759
4760         block = ext4_count_free_clusters(sb);
4761         ext4_free_blocks_count_set(sbi->s_es,
4762                                    EXT4_C2B(sbi, block));
4763         ext4_superblock_csum_set(sb);
4764         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4765                                   GFP_KERNEL);
4766         if (!err) {
4767                 unsigned long freei = ext4_count_free_inodes(sb);
4768                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4769                 ext4_superblock_csum_set(sb);
4770                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4771                                           GFP_KERNEL);
4772         }
4773         if (!err)
4774                 err = percpu_counter_init(&sbi->s_dirs_counter,
4775                                           ext4_count_dirs(sb), GFP_KERNEL);
4776         if (!err)
4777                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4778                                           GFP_KERNEL);
4779         if (!err)
4780                 err = percpu_init_rwsem(&sbi->s_writepages_rwsem);
4781
4782         if (err) {
4783                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4784                 goto failed_mount6;
4785         }
4786
4787         if (ext4_has_feature_flex_bg(sb))
4788                 if (!ext4_fill_flex_info(sb)) {
4789                         ext4_msg(sb, KERN_ERR,
4790                                "unable to initialize "
4791                                "flex_bg meta info!");
4792                         goto failed_mount6;
4793                 }
4794
4795         err = ext4_register_li_request(sb, first_not_zeroed);
4796         if (err)
4797                 goto failed_mount6;
4798
4799         err = ext4_register_sysfs(sb);
4800         if (err)
4801                 goto failed_mount7;
4802
4803 #ifdef CONFIG_QUOTA
4804         /* Enable quota usage during mount. */
4805         if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
4806                 err = ext4_enable_quotas(sb);
4807                 if (err)
4808                         goto failed_mount8;
4809         }
4810 #endif  /* CONFIG_QUOTA */
4811
4812         /*
4813          * Save the original bdev mapping's wb_err value which could be
4814          * used to detect the metadata async write error.
4815          */
4816         spin_lock_init(&sbi->s_bdev_wb_lock);
4817         if (!sb_rdonly(sb))
4818                 errseq_check_and_advance(&sb->s_bdev->bd_inode->i_mapping->wb_err,
4819                                          &sbi->s_bdev_wb_err);
4820         sb->s_bdev->bd_super = sb;
4821         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4822         ext4_orphan_cleanup(sb, es);
4823         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4824         if (needs_recovery) {
4825                 ext4_msg(sb, KERN_INFO, "recovery complete");
4826                 err = ext4_mark_recovery_complete(sb, es);
4827                 if (err)
4828                         goto failed_mount8;
4829         }
4830         if (EXT4_SB(sb)->s_journal) {
4831                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4832                         descr = " journalled data mode";
4833                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4834                         descr = " ordered data mode";
4835                 else
4836                         descr = " writeback data mode";
4837         } else
4838                 descr = "out journal";
4839
4840         if (test_opt(sb, DISCARD)) {
4841                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4842                 if (!blk_queue_discard(q))
4843                         ext4_msg(sb, KERN_WARNING,
4844                                  "mounting with \"discard\" option, but "
4845                                  "the device does not support discard");
4846         }
4847
4848         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4849                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4850                          "Opts: %.*s%s%s", descr,
4851                          (int) sizeof(sbi->s_es->s_mount_opts),
4852                          sbi->s_es->s_mount_opts,
4853                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4854
4855         if (es->s_error_count)
4856                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4857
4858         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4859         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4860         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4861         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4862         atomic_set(&sbi->s_warning_count, 0);
4863         atomic_set(&sbi->s_msg_count, 0);
4864
4865         kfree(orig_data);
4866         return 0;
4867
4868 cantfind_ext4:
4869         if (!silent)
4870                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4871         goto failed_mount;
4872
4873 failed_mount8:
4874         ext4_unregister_sysfs(sb);
4875 failed_mount7:
4876         ext4_unregister_li_request(sb);
4877 failed_mount6:
4878         ext4_mb_release(sb);
4879         rcu_read_lock();
4880         flex_groups = rcu_dereference(sbi->s_flex_groups);
4881         if (flex_groups) {
4882                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
4883                         kvfree(flex_groups[i]);
4884                 kvfree(flex_groups);
4885         }
4886         rcu_read_unlock();
4887         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4888         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4889         percpu_counter_destroy(&sbi->s_dirs_counter);
4890         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4891         percpu_free_rwsem(&sbi->s_writepages_rwsem);
4892 failed_mount5:
4893         ext4_ext_release(sb);
4894         ext4_release_system_zone(sb);
4895 failed_mount4a:
4896         dput(sb->s_root);
4897         sb->s_root = NULL;
4898 failed_mount4:
4899         ext4_msg(sb, KERN_ERR, "mount failed");
4900         if (EXT4_SB(sb)->rsv_conversion_wq)
4901                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4902 failed_mount_wq:
4903         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
4904         sbi->s_ea_inode_cache = NULL;
4905
4906         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
4907         sbi->s_ea_block_cache = NULL;
4908
4909         if (sbi->s_journal) {
4910                 jbd2_journal_destroy(sbi->s_journal);
4911                 sbi->s_journal = NULL;
4912         }
4913 failed_mount3a:
4914         ext4_es_unregister_shrinker(sbi);
4915 failed_mount3:
4916         del_timer_sync(&sbi->s_err_report);
4917         if (sbi->s_mmp_tsk)
4918                 kthread_stop(sbi->s_mmp_tsk);
4919 failed_mount2:
4920         rcu_read_lock();
4921         group_desc = rcu_dereference(sbi->s_group_desc);
4922         for (i = 0; i < db_count; i++)
4923                 brelse(group_desc[i]);
4924         kvfree(group_desc);
4925         rcu_read_unlock();
4926 failed_mount:
4927         if (sbi->s_chksum_driver)
4928                 crypto_free_shash(sbi->s_chksum_driver);
4929
4930 #ifdef CONFIG_UNICODE
4931         utf8_unload(sbi->s_encoding);
4932 #endif
4933
4934 #ifdef CONFIG_QUOTA
4935         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4936                 kfree(get_qf_name(sb, sbi, i));
4937 #endif
4938         fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy);
4939         ext4_blkdev_remove(sbi);
4940         brelse(bh);
4941 out_fail:
4942         sb->s_fs_info = NULL;
4943         kfree(sbi->s_blockgroup_lock);
4944 out_free_base:
4945         kfree(sbi);
4946         kfree(orig_data);
4947         fs_put_dax(dax_dev);
4948         return err ? err : ret;
4949 }
4950
4951 /*
4952  * Setup any per-fs journal parameters now.  We'll do this both on
4953  * initial mount, once the journal has been initialised but before we've
4954  * done any recovery; and again on any subsequent remount.
4955  */
4956 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4957 {
4958         struct ext4_sb_info *sbi = EXT4_SB(sb);
4959
4960         journal->j_commit_interval = sbi->s_commit_interval;
4961         journal->j_min_batch_time = sbi->s_min_batch_time;
4962         journal->j_max_batch_time = sbi->s_max_batch_time;
4963
4964         write_lock(&journal->j_state_lock);
4965         if (test_opt(sb, BARRIER))
4966                 journal->j_flags |= JBD2_BARRIER;
4967         else
4968                 journal->j_flags &= ~JBD2_BARRIER;
4969         if (test_opt(sb, DATA_ERR_ABORT))
4970                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4971         else
4972                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4973         write_unlock(&journal->j_state_lock);
4974 }
4975
4976 static struct inode *ext4_get_journal_inode(struct super_block *sb,
4977                                              unsigned int journal_inum)
4978 {
4979         struct inode *journal_inode;
4980
4981         /*
4982          * Test for the existence of a valid inode on disk.  Bad things
4983          * happen if we iget() an unused inode, as the subsequent iput()
4984          * will try to delete it.
4985          */
4986         journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL);
4987         if (IS_ERR(journal_inode)) {
4988                 ext4_msg(sb, KERN_ERR, "no journal found");
4989                 return NULL;
4990         }
4991         if (!journal_inode->i_nlink) {
4992                 make_bad_inode(journal_inode);
4993                 iput(journal_inode);
4994                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4995                 return NULL;
4996         }
4997
4998         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4999                   journal_inode, journal_inode->i_size);
5000         if (!S_ISREG(journal_inode->i_mode)) {
5001                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
5002                 iput(journal_inode);
5003                 return NULL;
5004         }
5005         return journal_inode;
5006 }
5007
5008 static journal_t *ext4_get_journal(struct super_block *sb,
5009                                    unsigned int journal_inum)
5010 {
5011         struct inode *journal_inode;
5012         journal_t *journal;
5013
5014         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5015                 return NULL;
5016
5017         journal_inode = ext4_get_journal_inode(sb, journal_inum);
5018         if (!journal_inode)
5019                 return NULL;
5020
5021         journal = jbd2_journal_init_inode(journal_inode);
5022         if (!journal) {
5023                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
5024                 iput(journal_inode);
5025                 return NULL;
5026         }
5027         journal->j_private = sb;
5028         ext4_init_journal_params(sb, journal);
5029         return journal;
5030 }
5031
5032 static journal_t *ext4_get_dev_journal(struct super_block *sb,
5033                                        dev_t j_dev)
5034 {
5035         struct buffer_head *bh;
5036         journal_t *journal;
5037         ext4_fsblk_t start;
5038         ext4_fsblk_t len;
5039         int hblock, blocksize;
5040         ext4_fsblk_t sb_block;
5041         unsigned long offset;
5042         struct ext4_super_block *es;
5043         struct block_device *bdev;
5044
5045         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5046                 return NULL;
5047
5048         bdev = ext4_blkdev_get(j_dev, sb);
5049         if (bdev == NULL)
5050                 return NULL;
5051
5052         blocksize = sb->s_blocksize;
5053         hblock = bdev_logical_block_size(bdev);
5054         if (blocksize < hblock) {
5055                 ext4_msg(sb, KERN_ERR,
5056                         "blocksize too small for journal device");
5057                 goto out_bdev;
5058         }
5059
5060         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
5061         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
5062         set_blocksize(bdev, blocksize);
5063         if (!(bh = __bread(bdev, sb_block, blocksize))) {
5064                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
5065                        "external journal");
5066                 goto out_bdev;
5067         }
5068
5069         es = (struct ext4_super_block *) (bh->b_data + offset);
5070         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
5071             !(le32_to_cpu(es->s_feature_incompat) &
5072               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
5073                 ext4_msg(sb, KERN_ERR, "external journal has "
5074                                         "bad superblock");
5075                 brelse(bh);
5076                 goto out_bdev;
5077         }
5078
5079         if ((le32_to_cpu(es->s_feature_ro_compat) &
5080              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
5081             es->s_checksum != ext4_superblock_csum(sb, es)) {
5082                 ext4_msg(sb, KERN_ERR, "external journal has "
5083                                        "corrupt superblock");
5084                 brelse(bh);
5085                 goto out_bdev;
5086         }
5087
5088         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
5089                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
5090                 brelse(bh);
5091                 goto out_bdev;
5092         }
5093
5094         len = ext4_blocks_count(es);
5095         start = sb_block + 1;
5096         brelse(bh);     /* we're done with the superblock */
5097
5098         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
5099                                         start, len, blocksize);
5100         if (!journal) {
5101                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
5102                 goto out_bdev;
5103         }
5104         journal->j_private = sb;
5105         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
5106         wait_on_buffer(journal->j_sb_buffer);
5107         if (!buffer_uptodate(journal->j_sb_buffer)) {
5108                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
5109                 goto out_journal;
5110         }
5111         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
5112                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
5113                                         "user (unsupported) - %d",
5114                         be32_to_cpu(journal->j_superblock->s_nr_users));
5115                 goto out_journal;
5116         }
5117         EXT4_SB(sb)->journal_bdev = bdev;
5118         ext4_init_journal_params(sb, journal);
5119         return journal;
5120
5121 out_journal:
5122         jbd2_journal_destroy(journal);
5123 out_bdev:
5124         ext4_blkdev_put(bdev);
5125         return NULL;
5126 }
5127
5128 static int ext4_load_journal(struct super_block *sb,
5129                              struct ext4_super_block *es,
5130                              unsigned long journal_devnum)
5131 {
5132         journal_t *journal;
5133         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
5134         dev_t journal_dev;
5135         int err = 0;
5136         int really_read_only;
5137         int journal_dev_ro;
5138
5139         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5140                 return -EFSCORRUPTED;
5141
5142         if (journal_devnum &&
5143             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5144                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
5145                         "numbers have changed");
5146                 journal_dev = new_decode_dev(journal_devnum);
5147         } else
5148                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
5149
5150         if (journal_inum && journal_dev) {
5151                 ext4_msg(sb, KERN_ERR,
5152                          "filesystem has both journal inode and journal device!");
5153                 return -EINVAL;
5154         }
5155
5156         if (journal_inum) {
5157                 journal = ext4_get_journal(sb, journal_inum);
5158                 if (!journal)
5159                         return -EINVAL;
5160         } else {
5161                 journal = ext4_get_dev_journal(sb, journal_dev);
5162                 if (!journal)
5163                         return -EINVAL;
5164         }
5165
5166         journal_dev_ro = bdev_read_only(journal->j_dev);
5167         really_read_only = bdev_read_only(sb->s_bdev) | journal_dev_ro;
5168
5169         if (journal_dev_ro && !sb_rdonly(sb)) {
5170                 ext4_msg(sb, KERN_ERR,
5171                          "journal device read-only, try mounting with '-o ro'");
5172                 err = -EROFS;
5173                 goto err_out;
5174         }
5175
5176         /*
5177          * Are we loading a blank journal or performing recovery after a
5178          * crash?  For recovery, we need to check in advance whether we
5179          * can get read-write access to the device.
5180          */
5181         if (ext4_has_feature_journal_needs_recovery(sb)) {
5182                 if (sb_rdonly(sb)) {
5183                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
5184                                         "required on readonly filesystem");
5185                         if (really_read_only) {
5186                                 ext4_msg(sb, KERN_ERR, "write access "
5187                                         "unavailable, cannot proceed "
5188                                         "(try mounting with noload)");
5189                                 err = -EROFS;
5190                                 goto err_out;
5191                         }
5192                         ext4_msg(sb, KERN_INFO, "write access will "
5193                                "be enabled during recovery");
5194                 }
5195         }
5196
5197         if (!(journal->j_flags & JBD2_BARRIER))
5198                 ext4_msg(sb, KERN_INFO, "barriers disabled");
5199
5200         if (!ext4_has_feature_journal_needs_recovery(sb))
5201                 err = jbd2_journal_wipe(journal, !really_read_only);
5202         if (!err) {
5203                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
5204                 if (save)
5205                         memcpy(save, ((char *) es) +
5206                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
5207                 err = jbd2_journal_load(journal);
5208                 if (save)
5209                         memcpy(((char *) es) + EXT4_S_ERR_START,
5210                                save, EXT4_S_ERR_LEN);
5211                 kfree(save);
5212         }
5213
5214         if (err) {
5215                 ext4_msg(sb, KERN_ERR, "error loading journal");
5216                 goto err_out;
5217         }
5218
5219         EXT4_SB(sb)->s_journal = journal;
5220         err = ext4_clear_journal_err(sb, es);
5221         if (err) {
5222                 EXT4_SB(sb)->s_journal = NULL;
5223                 jbd2_journal_destroy(journal);
5224                 return err;
5225         }
5226
5227         if (!really_read_only && journal_devnum &&
5228             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5229                 es->s_journal_dev = cpu_to_le32(journal_devnum);
5230
5231                 /* Make sure we flush the recovery flag to disk. */
5232                 ext4_commit_super(sb, 1);
5233         }
5234
5235         return 0;
5236
5237 err_out:
5238         jbd2_journal_destroy(journal);
5239         return err;
5240 }
5241
5242 static int ext4_commit_super(struct super_block *sb, int sync)
5243 {
5244         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
5245         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
5246         int error = 0;
5247
5248         if (!sbh || block_device_ejected(sb))
5249                 return error;
5250
5251         /*
5252          * If the file system is mounted read-only, don't update the
5253          * superblock write time.  This avoids updating the superblock
5254          * write time when we are mounting the root file system
5255          * read/only but we need to replay the journal; at that point,
5256          * for people who are east of GMT and who make their clock
5257          * tick in localtime for Windows bug-for-bug compatibility,
5258          * the clock is set in the future, and this will cause e2fsck
5259          * to complain and force a full file system check.
5260          */
5261         if (!(sb->s_flags & SB_RDONLY))
5262                 ext4_update_tstamp(es, s_wtime);
5263         if (sb->s_bdev->bd_part)
5264                 es->s_kbytes_written =
5265                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
5266                             ((part_stat_read(sb->s_bdev->bd_part,
5267                                              sectors[STAT_WRITE]) -
5268                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
5269         else
5270                 es->s_kbytes_written =
5271                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
5272         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
5273                 ext4_free_blocks_count_set(es,
5274                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
5275                                 &EXT4_SB(sb)->s_freeclusters_counter)));
5276         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
5277                 es->s_free_inodes_count =
5278                         cpu_to_le32(percpu_counter_sum_positive(
5279                                 &EXT4_SB(sb)->s_freeinodes_counter));
5280         BUFFER_TRACE(sbh, "marking dirty");
5281         ext4_superblock_csum_set(sb);
5282         if (sync)
5283                 lock_buffer(sbh);
5284         if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
5285                 /*
5286                  * Oh, dear.  A previous attempt to write the
5287                  * superblock failed.  This could happen because the
5288                  * USB device was yanked out.  Or it could happen to
5289                  * be a transient write error and maybe the block will
5290                  * be remapped.  Nothing we can do but to retry the
5291                  * write and hope for the best.
5292                  */
5293                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
5294                        "superblock detected");
5295                 clear_buffer_write_io_error(sbh);
5296                 set_buffer_uptodate(sbh);
5297         }
5298         mark_buffer_dirty(sbh);
5299         if (sync) {
5300                 unlock_buffer(sbh);
5301                 error = __sync_dirty_buffer(sbh,
5302                         REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
5303                 if (buffer_write_io_error(sbh)) {
5304                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
5305                                "superblock");
5306                         clear_buffer_write_io_error(sbh);
5307                         set_buffer_uptodate(sbh);
5308                 }
5309         }
5310         return error;
5311 }
5312
5313 /*
5314  * Have we just finished recovery?  If so, and if we are mounting (or
5315  * remounting) the filesystem readonly, then we will end up with a
5316  * consistent fs on disk.  Record that fact.
5317  */
5318 static int ext4_mark_recovery_complete(struct super_block *sb,
5319                                        struct ext4_super_block *es)
5320 {
5321         int err;
5322         journal_t *journal = EXT4_SB(sb)->s_journal;
5323
5324         if (!ext4_has_feature_journal(sb)) {
5325                 if (journal != NULL) {
5326                         ext4_error(sb, "Journal got removed while the fs was "
5327                                    "mounted!");
5328                         return -EFSCORRUPTED;
5329                 }
5330                 return 0;
5331         }
5332         jbd2_journal_lock_updates(journal);
5333         err = jbd2_journal_flush(journal);
5334         if (err < 0)
5335                 goto out;
5336
5337         if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) {
5338                 ext4_clear_feature_journal_needs_recovery(sb);
5339                 ext4_commit_super(sb, 1);
5340         }
5341 out:
5342         jbd2_journal_unlock_updates(journal);
5343         return err;
5344 }
5345
5346 /*
5347  * If we are mounting (or read-write remounting) a filesystem whose journal
5348  * has recorded an error from a previous lifetime, move that error to the
5349  * main filesystem now.
5350  */
5351 static int ext4_clear_journal_err(struct super_block *sb,
5352                                    struct ext4_super_block *es)
5353 {
5354         journal_t *journal;
5355         int j_errno;
5356         const char *errstr;
5357
5358         if (!ext4_has_feature_journal(sb)) {
5359                 ext4_error(sb, "Journal got removed while the fs was mounted!");
5360                 return -EFSCORRUPTED;
5361         }
5362
5363         journal = EXT4_SB(sb)->s_journal;
5364
5365         /*
5366          * Now check for any error status which may have been recorded in the
5367          * journal by a prior ext4_error() or ext4_abort()
5368          */
5369
5370         j_errno = jbd2_journal_errno(journal);
5371         if (j_errno) {
5372                 char nbuf[16];
5373
5374                 errstr = ext4_decode_error(sb, j_errno, nbuf);
5375                 ext4_warning(sb, "Filesystem error recorded "
5376                              "from previous mount: %s", errstr);
5377                 ext4_warning(sb, "Marking fs in need of filesystem check.");
5378
5379                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
5380                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
5381                 ext4_commit_super(sb, 1);
5382
5383                 jbd2_journal_clear_err(journal);
5384                 jbd2_journal_update_sb_errno(journal);
5385         }
5386         return 0;
5387 }
5388
5389 /*
5390  * Force the running and committing transactions to commit,
5391  * and wait on the commit.
5392  */
5393 int ext4_force_commit(struct super_block *sb)
5394 {
5395         journal_t *journal;
5396
5397         if (sb_rdonly(sb))
5398                 return 0;
5399
5400         journal = EXT4_SB(sb)->s_journal;
5401         return ext4_journal_force_commit(journal);
5402 }
5403
5404 static int ext4_sync_fs(struct super_block *sb, int wait)
5405 {
5406         int ret = 0;
5407         tid_t target;
5408         bool needs_barrier = false;
5409         struct ext4_sb_info *sbi = EXT4_SB(sb);
5410
5411         if (unlikely(ext4_forced_shutdown(sbi)))
5412                 return 0;
5413
5414         trace_ext4_sync_fs(sb, wait);
5415         flush_workqueue(sbi->rsv_conversion_wq);
5416         /*
5417          * Writeback quota in non-journalled quota case - journalled quota has
5418          * no dirty dquots
5419          */
5420         dquot_writeback_dquots(sb, -1);
5421         /*
5422          * Data writeback is possible w/o journal transaction, so barrier must
5423          * being sent at the end of the function. But we can skip it if
5424          * transaction_commit will do it for us.
5425          */
5426         if (sbi->s_journal) {
5427                 target = jbd2_get_latest_transaction(sbi->s_journal);
5428                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
5429                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
5430                         needs_barrier = true;
5431
5432                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
5433                         if (wait)
5434                                 ret = jbd2_log_wait_commit(sbi->s_journal,
5435                                                            target);
5436                 }
5437         } else if (wait && test_opt(sb, BARRIER))
5438                 needs_barrier = true;
5439         if (needs_barrier) {
5440                 int err;
5441                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL);
5442                 if (!ret)
5443                         ret = err;
5444         }
5445
5446         return ret;
5447 }
5448
5449 /*
5450  * LVM calls this function before a (read-only) snapshot is created.  This
5451  * gives us a chance to flush the journal completely and mark the fs clean.
5452  *
5453  * Note that only this function cannot bring a filesystem to be in a clean
5454  * state independently. It relies on upper layer to stop all data & metadata
5455  * modifications.
5456  */
5457 static int ext4_freeze(struct super_block *sb)
5458 {
5459         int error = 0;
5460         journal_t *journal;
5461
5462         if (sb_rdonly(sb))
5463                 return 0;
5464
5465         journal = EXT4_SB(sb)->s_journal;
5466
5467         if (journal) {
5468                 /* Now we set up the journal barrier. */
5469                 jbd2_journal_lock_updates(journal);
5470
5471                 /*
5472                  * Don't clear the needs_recovery flag if we failed to
5473                  * flush the journal.
5474                  */
5475                 error = jbd2_journal_flush(journal);
5476                 if (error < 0)
5477                         goto out;
5478
5479                 /* Journal blocked and flushed, clear needs_recovery flag. */
5480                 ext4_clear_feature_journal_needs_recovery(sb);
5481         }
5482
5483         error = ext4_commit_super(sb, 1);
5484 out:
5485         if (journal)
5486                 /* we rely on upper layer to stop further updates */
5487                 jbd2_journal_unlock_updates(journal);
5488         return error;
5489 }
5490
5491 /*
5492  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
5493  * flag here, even though the filesystem is not technically dirty yet.
5494  */
5495 static int ext4_unfreeze(struct super_block *sb)
5496 {
5497         if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
5498                 return 0;
5499
5500         if (EXT4_SB(sb)->s_journal) {
5501                 /* Reset the needs_recovery flag before the fs is unlocked. */
5502                 ext4_set_feature_journal_needs_recovery(sb);
5503         }
5504
5505         ext4_commit_super(sb, 1);
5506         return 0;
5507 }
5508
5509 /*
5510  * Structure to save mount options for ext4_remount's benefit
5511  */
5512 struct ext4_mount_options {
5513         unsigned long s_mount_opt;
5514         unsigned long s_mount_opt2;
5515         kuid_t s_resuid;
5516         kgid_t s_resgid;
5517         unsigned long s_commit_interval;
5518         u32 s_min_batch_time, s_max_batch_time;
5519 #ifdef CONFIG_QUOTA
5520         int s_jquota_fmt;
5521         char *s_qf_names[EXT4_MAXQUOTAS];
5522 #endif
5523 };
5524
5525 static int ext4_remount(struct super_block *sb, int *flags, char *data)
5526 {
5527         struct ext4_super_block *es;
5528         struct ext4_sb_info *sbi = EXT4_SB(sb);
5529         unsigned long old_sb_flags, vfs_flags;
5530         struct ext4_mount_options old_opts;
5531         int enable_quota = 0;
5532         ext4_group_t g;
5533         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
5534         int err = 0;
5535 #ifdef CONFIG_QUOTA
5536         int i, j;
5537         char *to_free[EXT4_MAXQUOTAS];
5538 #endif
5539         char *orig_data = kstrdup(data, GFP_KERNEL);
5540
5541         if (data && !orig_data)
5542                 return -ENOMEM;
5543
5544         /* Store the original options */
5545         old_sb_flags = sb->s_flags;
5546         old_opts.s_mount_opt = sbi->s_mount_opt;
5547         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
5548         old_opts.s_resuid = sbi->s_resuid;
5549         old_opts.s_resgid = sbi->s_resgid;
5550         old_opts.s_commit_interval = sbi->s_commit_interval;
5551         old_opts.s_min_batch_time = sbi->s_min_batch_time;
5552         old_opts.s_max_batch_time = sbi->s_max_batch_time;
5553 #ifdef CONFIG_QUOTA
5554         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
5555         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5556                 if (sbi->s_qf_names[i]) {
5557                         char *qf_name = get_qf_name(sb, sbi, i);
5558
5559                         old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL);
5560                         if (!old_opts.s_qf_names[i]) {
5561                                 for (j = 0; j < i; j++)
5562                                         kfree(old_opts.s_qf_names[j]);
5563                                 kfree(orig_data);
5564                                 return -ENOMEM;
5565                         }
5566                 } else
5567                         old_opts.s_qf_names[i] = NULL;
5568 #endif
5569         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
5570                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
5571
5572         /*
5573          * Some options can be enabled by ext4 and/or by VFS mount flag
5574          * either way we need to make sure it matches in both *flags and
5575          * s_flags. Copy those selected flags from *flags to s_flags
5576          */
5577         vfs_flags = SB_LAZYTIME | SB_I_VERSION;
5578         sb->s_flags = (sb->s_flags & ~vfs_flags) | (*flags & vfs_flags);
5579
5580         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
5581                 err = -EINVAL;
5582                 goto restore_opts;
5583         }
5584
5585         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
5586             test_opt(sb, JOURNAL_CHECKSUM)) {
5587                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
5588                          "during remount not supported; ignoring");
5589                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
5590         }
5591
5592         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
5593                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
5594                         ext4_msg(sb, KERN_ERR, "can't mount with "
5595                                  "both data=journal and delalloc");
5596                         err = -EINVAL;
5597                         goto restore_opts;
5598                 }
5599                 if (test_opt(sb, DIOREAD_NOLOCK)) {
5600                         ext4_msg(sb, KERN_ERR, "can't mount with "
5601                                  "both data=journal and dioread_nolock");
5602                         err = -EINVAL;
5603                         goto restore_opts;
5604                 }
5605         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
5606                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5607                         ext4_msg(sb, KERN_ERR, "can't mount with "
5608                                 "journal_async_commit in data=ordered mode");
5609                         err = -EINVAL;
5610                         goto restore_opts;
5611                 }
5612         }
5613
5614         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
5615                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
5616                 err = -EINVAL;
5617                 goto restore_opts;
5618         }
5619
5620         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
5621                 ext4_abort(sb, EXT4_ERR_ESHUTDOWN, "Abort forced by user");
5622
5623         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
5624                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
5625
5626         es = sbi->s_es;
5627
5628         if (sbi->s_journal) {
5629                 ext4_init_journal_params(sb, sbi->s_journal);
5630                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
5631         }
5632
5633         if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) {
5634                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
5635                         err = -EROFS;
5636                         goto restore_opts;
5637                 }
5638
5639                 if (*flags & SB_RDONLY) {
5640                         err = sync_filesystem(sb);
5641                         if (err < 0)
5642                                 goto restore_opts;
5643                         err = dquot_suspend(sb, -1);
5644                         if (err < 0)
5645                                 goto restore_opts;
5646
5647                         /*
5648                          * First of all, the unconditional stuff we have to do
5649                          * to disable replay of the journal when we next remount
5650                          */
5651                         sb->s_flags |= SB_RDONLY;
5652
5653                         /*
5654                          * OK, test if we are remounting a valid rw partition
5655                          * readonly, and if so set the rdonly flag and then
5656                          * mark the partition as valid again.
5657                          */
5658                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5659                             (sbi->s_mount_state & EXT4_VALID_FS))
5660                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5661
5662                         if (sbi->s_journal) {
5663                                 /*
5664                                  * We let remount-ro finish even if marking fs
5665                                  * as clean failed...
5666                                  */
5667                                 ext4_mark_recovery_complete(sb, es);
5668                         }
5669                         if (sbi->s_mmp_tsk)
5670                                 kthread_stop(sbi->s_mmp_tsk);
5671                 } else {
5672                         /* Make sure we can mount this feature set readwrite */
5673                         if (ext4_has_feature_readonly(sb) ||
5674                             !ext4_feature_set_ok(sb, 0)) {
5675                                 err = -EROFS;
5676                                 goto restore_opts;
5677                         }
5678                         /*
5679                          * Make sure the group descriptor checksums
5680                          * are sane.  If they aren't, refuse to remount r/w.
5681                          */
5682                         for (g = 0; g < sbi->s_groups_count; g++) {
5683                                 struct ext4_group_desc *gdp =
5684                                         ext4_get_group_desc(sb, g, NULL);
5685
5686                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5687                                         ext4_msg(sb, KERN_ERR,
5688                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5689                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
5690                                                le16_to_cpu(gdp->bg_checksum));
5691                                         err = -EFSBADCRC;
5692                                         goto restore_opts;
5693                                 }
5694                         }
5695
5696                         /*
5697                          * If we have an unprocessed orphan list hanging
5698                          * around from a previously readonly bdev mount,
5699                          * require a full umount/remount for now.
5700                          */
5701                         if (es->s_last_orphan) {
5702                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5703                                        "remount RDWR because of unprocessed "
5704                                        "orphan inode list.  Please "
5705                                        "umount/remount instead");
5706                                 err = -EINVAL;
5707                                 goto restore_opts;
5708                         }
5709
5710                         /*
5711                          * Update the original bdev mapping's wb_err value
5712                          * which could be used to detect the metadata async
5713                          * write error.
5714                          */
5715                         errseq_check_and_advance(&sb->s_bdev->bd_inode->i_mapping->wb_err,
5716                                                  &sbi->s_bdev_wb_err);
5717
5718                         /*
5719                          * Mounting a RDONLY partition read-write, so reread
5720                          * and store the current valid flag.  (It may have
5721                          * been changed by e2fsck since we originally mounted
5722                          * the partition.)
5723                          */
5724                         if (sbi->s_journal) {
5725                                 err = ext4_clear_journal_err(sb, es);
5726                                 if (err)
5727                                         goto restore_opts;
5728                         }
5729                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5730
5731                         err = ext4_setup_super(sb, es, 0);
5732                         if (err)
5733                                 goto restore_opts;
5734
5735                         sb->s_flags &= ~SB_RDONLY;
5736                         if (ext4_has_feature_mmp(sb))
5737                                 if (ext4_multi_mount_protect(sb,
5738                                                 le64_to_cpu(es->s_mmp_block))) {
5739                                         err = -EROFS;
5740                                         goto restore_opts;
5741                                 }
5742                         enable_quota = 1;
5743                 }
5744         }
5745
5746         /*
5747          * Reinitialize lazy itable initialization thread based on
5748          * current settings
5749          */
5750         if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
5751                 ext4_unregister_li_request(sb);
5752         else {
5753                 ext4_group_t first_not_zeroed;
5754                 first_not_zeroed = ext4_has_uninit_itable(sb);
5755                 ext4_register_li_request(sb, first_not_zeroed);
5756         }
5757
5758         /*
5759          * Handle creation of system zone data early because it can fail.
5760          * Releasing of existing data is done when we are sure remount will
5761          * succeed.
5762          */
5763         if (test_opt(sb, BLOCK_VALIDITY) && !sbi->system_blks) {
5764                 err = ext4_setup_system_zone(sb);
5765                 if (err)
5766                         goto restore_opts;
5767         }
5768
5769         if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) {
5770                 err = ext4_commit_super(sb, 1);
5771                 if (err)
5772                         goto restore_opts;
5773         }
5774
5775 #ifdef CONFIG_QUOTA
5776         /* Release old quota file names */
5777         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5778                 kfree(old_opts.s_qf_names[i]);
5779         if (enable_quota) {
5780                 if (sb_any_quota_suspended(sb))
5781                         dquot_resume(sb, -1);
5782                 else if (ext4_has_feature_quota(sb)) {
5783                         err = ext4_enable_quotas(sb);
5784                         if (err)
5785                                 goto restore_opts;
5786                 }
5787         }
5788 #endif
5789         if (!test_opt(sb, BLOCK_VALIDITY) && sbi->system_blks)
5790                 ext4_release_system_zone(sb);
5791
5792         /*
5793          * Some options can be enabled by ext4 and/or by VFS mount flag
5794          * either way we need to make sure it matches in both *flags and
5795          * s_flags. Copy those selected flags from s_flags to *flags
5796          */
5797         *flags = (*flags & ~vfs_flags) | (sb->s_flags & vfs_flags);
5798
5799         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5800         kfree(orig_data);
5801         return 0;
5802
5803 restore_opts:
5804         sb->s_flags = old_sb_flags;
5805         sbi->s_mount_opt = old_opts.s_mount_opt;
5806         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5807         sbi->s_resuid = old_opts.s_resuid;
5808         sbi->s_resgid = old_opts.s_resgid;
5809         sbi->s_commit_interval = old_opts.s_commit_interval;
5810         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5811         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5812         if (!test_opt(sb, BLOCK_VALIDITY) && sbi->system_blks)
5813                 ext4_release_system_zone(sb);
5814 #ifdef CONFIG_QUOTA
5815         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5816         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5817                 to_free[i] = get_qf_name(sb, sbi, i);
5818                 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]);
5819         }
5820         synchronize_rcu();
5821         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5822                 kfree(to_free[i]);
5823 #endif
5824         kfree(orig_data);
5825         return err;
5826 }
5827
5828 #ifdef CONFIG_QUOTA
5829 static int ext4_statfs_project(struct super_block *sb,
5830                                kprojid_t projid, struct kstatfs *buf)
5831 {
5832         struct kqid qid;
5833         struct dquot *dquot;
5834         u64 limit;
5835         u64 curblock;
5836
5837         qid = make_kqid_projid(projid);
5838         dquot = dqget(sb, qid);
5839         if (IS_ERR(dquot))
5840                 return PTR_ERR(dquot);
5841         spin_lock(&dquot->dq_dqb_lock);
5842
5843         limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit,
5844                              dquot->dq_dqb.dqb_bhardlimit);
5845         limit >>= sb->s_blocksize_bits;
5846
5847         if (limit && buf->f_blocks > limit) {
5848                 curblock = (dquot->dq_dqb.dqb_curspace +
5849                             dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits;
5850                 buf->f_blocks = limit;
5851                 buf->f_bfree = buf->f_bavail =
5852                         (buf->f_blocks > curblock) ?
5853                          (buf->f_blocks - curblock) : 0;
5854         }
5855
5856         limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit,
5857                              dquot->dq_dqb.dqb_ihardlimit);
5858         if (limit && buf->f_files > limit) {
5859                 buf->f_files = limit;
5860                 buf->f_ffree =
5861                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
5862                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
5863         }
5864
5865         spin_unlock(&dquot->dq_dqb_lock);
5866         dqput(dquot);
5867         return 0;
5868 }
5869 #endif
5870
5871 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5872 {
5873         struct super_block *sb = dentry->d_sb;
5874         struct ext4_sb_info *sbi = EXT4_SB(sb);
5875         struct ext4_super_block *es = sbi->s_es;
5876         ext4_fsblk_t overhead = 0, resv_blocks;
5877         u64 fsid;
5878         s64 bfree;
5879         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5880
5881         if (!test_opt(sb, MINIX_DF))
5882                 overhead = sbi->s_overhead;
5883
5884         buf->f_type = EXT4_SUPER_MAGIC;
5885         buf->f_bsize = sb->s_blocksize;
5886         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5887         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5888                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5889         /* prevent underflow in case that few free space is available */
5890         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5891         buf->f_bavail = buf->f_bfree -
5892                         (ext4_r_blocks_count(es) + resv_blocks);
5893         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5894                 buf->f_bavail = 0;
5895         buf->f_files = le32_to_cpu(es->s_inodes_count);
5896         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5897         buf->f_namelen = EXT4_NAME_LEN;
5898         fsid = le64_to_cpup((void *)es->s_uuid) ^
5899                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5900         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5901         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5902
5903 #ifdef CONFIG_QUOTA
5904         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
5905             sb_has_quota_limits_enabled(sb, PRJQUOTA))
5906                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
5907 #endif
5908         return 0;
5909 }
5910
5911
5912 #ifdef CONFIG_QUOTA
5913
5914 /*
5915  * Helper functions so that transaction is started before we acquire dqio_sem
5916  * to keep correct lock ordering of transaction > dqio_sem
5917  */
5918 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5919 {
5920         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5921 }
5922
5923 static int ext4_write_dquot(struct dquot *dquot)
5924 {
5925         int ret, err;
5926         handle_t *handle;
5927         struct inode *inode;
5928
5929         inode = dquot_to_inode(dquot);
5930         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5931                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5932         if (IS_ERR(handle))
5933                 return PTR_ERR(handle);
5934         ret = dquot_commit(dquot);
5935         err = ext4_journal_stop(handle);
5936         if (!ret)
5937                 ret = err;
5938         return ret;
5939 }
5940
5941 static int ext4_acquire_dquot(struct dquot *dquot)
5942 {
5943         int ret, err;
5944         handle_t *handle;
5945
5946         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5947                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5948         if (IS_ERR(handle))
5949                 return PTR_ERR(handle);
5950         ret = dquot_acquire(dquot);
5951         err = ext4_journal_stop(handle);
5952         if (!ret)
5953                 ret = err;
5954         return ret;
5955 }
5956
5957 static int ext4_release_dquot(struct dquot *dquot)
5958 {
5959         int ret, err;
5960         handle_t *handle;
5961
5962         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5963                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5964         if (IS_ERR(handle)) {
5965                 /* Release dquot anyway to avoid endless cycle in dqput() */
5966                 dquot_release(dquot);
5967                 return PTR_ERR(handle);
5968         }
5969         ret = dquot_release(dquot);
5970         err = ext4_journal_stop(handle);
5971         if (!ret)
5972                 ret = err;
5973         return ret;
5974 }
5975
5976 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5977 {
5978         struct super_block *sb = dquot->dq_sb;
5979         struct ext4_sb_info *sbi = EXT4_SB(sb);
5980
5981         /* Are we journaling quotas? */
5982         if (ext4_has_feature_quota(sb) ||
5983             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5984                 dquot_mark_dquot_dirty(dquot);
5985                 return ext4_write_dquot(dquot);
5986         } else {
5987                 return dquot_mark_dquot_dirty(dquot);
5988         }
5989 }
5990
5991 static int ext4_write_info(struct super_block *sb, int type)
5992 {
5993         int ret, err;
5994         handle_t *handle;
5995
5996         /* Data block + inode block */
5997         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5998         if (IS_ERR(handle))
5999                 return PTR_ERR(handle);
6000         ret = dquot_commit_info(sb, type);
6001         err = ext4_journal_stop(handle);
6002         if (!ret)
6003                 ret = err;
6004         return ret;
6005 }
6006
6007 /*
6008  * Turn on quotas during mount time - we need to find
6009  * the quota file and such...
6010  */
6011 static int ext4_quota_on_mount(struct super_block *sb, int type)
6012 {
6013         return dquot_quota_on_mount(sb, get_qf_name(sb, EXT4_SB(sb), type),
6014                                         EXT4_SB(sb)->s_jquota_fmt, type);
6015 }
6016
6017 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
6018 {
6019         struct ext4_inode_info *ei = EXT4_I(inode);
6020
6021         /* The first argument of lockdep_set_subclass has to be
6022          * *exactly* the same as the argument to init_rwsem() --- in
6023          * this case, in init_once() --- or lockdep gets unhappy
6024          * because the name of the lock is set using the
6025          * stringification of the argument to init_rwsem().
6026          */
6027         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
6028         lockdep_set_subclass(&ei->i_data_sem, subclass);
6029 }
6030
6031 /*
6032  * Standard function to be called on quota_on
6033  */
6034 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
6035                          const struct path *path)
6036 {
6037         int err;
6038
6039         if (!test_opt(sb, QUOTA))
6040                 return -EINVAL;
6041
6042         /* Quotafile not on the same filesystem? */
6043         if (path->dentry->d_sb != sb)
6044                 return -EXDEV;
6045         /* Journaling quota? */
6046         if (EXT4_SB(sb)->s_qf_names[type]) {
6047                 /* Quotafile not in fs root? */
6048                 if (path->dentry->d_parent != sb->s_root)
6049                         ext4_msg(sb, KERN_WARNING,
6050                                 "Quota file not on filesystem root. "
6051                                 "Journaled quota will not work");
6052                 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
6053         } else {
6054                 /*
6055                  * Clear the flag just in case mount options changed since
6056                  * last time.
6057                  */
6058                 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
6059         }
6060
6061         /*
6062          * When we journal data on quota file, we have to flush journal to see
6063          * all updates to the file when we bypass pagecache...
6064          */
6065         if (EXT4_SB(sb)->s_journal &&
6066             ext4_should_journal_data(d_inode(path->dentry))) {
6067                 /*
6068                  * We don't need to lock updates but journal_flush() could
6069                  * otherwise be livelocked...
6070                  */
6071                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
6072                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
6073                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
6074                 if (err)
6075                         return err;
6076         }
6077
6078         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
6079         err = dquot_quota_on(sb, type, format_id, path);
6080         if (err) {
6081                 lockdep_set_quota_inode(path->dentry->d_inode,
6082                                              I_DATA_SEM_NORMAL);
6083         } else {
6084                 struct inode *inode = d_inode(path->dentry);
6085                 handle_t *handle;
6086
6087                 /*
6088                  * Set inode flags to prevent userspace from messing with quota
6089                  * files. If this fails, we return success anyway since quotas
6090                  * are already enabled and this is not a hard failure.
6091                  */
6092                 inode_lock(inode);
6093                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
6094                 if (IS_ERR(handle))
6095                         goto unlock_inode;
6096                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
6097                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
6098                                 S_NOATIME | S_IMMUTABLE);
6099                 err = ext4_mark_inode_dirty(handle, inode);
6100                 ext4_journal_stop(handle);
6101         unlock_inode:
6102                 inode_unlock(inode);
6103         }
6104         return err;
6105 }
6106
6107 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
6108                              unsigned int flags)
6109 {
6110         int err;
6111         struct inode *qf_inode;
6112         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
6113                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
6114                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
6115                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
6116         };
6117
6118         BUG_ON(!ext4_has_feature_quota(sb));
6119
6120         if (!qf_inums[type])
6121                 return -EPERM;
6122
6123         qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL);
6124         if (IS_ERR(qf_inode)) {
6125                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
6126                 return PTR_ERR(qf_inode);
6127         }
6128
6129         /* Don't account quota for quota files to avoid recursion */
6130         qf_inode->i_flags |= S_NOQUOTA;
6131         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
6132         err = dquot_load_quota_inode(qf_inode, type, format_id, flags);
6133         if (err)
6134                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
6135         iput(qf_inode);
6136
6137         return err;
6138 }
6139
6140 /* Enable usage tracking for all quota types. */
6141 static int ext4_enable_quotas(struct super_block *sb)
6142 {
6143         int type, err = 0;
6144         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
6145                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
6146                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
6147                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
6148         };
6149         bool quota_mopt[EXT4_MAXQUOTAS] = {
6150                 test_opt(sb, USRQUOTA),
6151                 test_opt(sb, GRPQUOTA),
6152                 test_opt(sb, PRJQUOTA),
6153         };
6154
6155         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
6156         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
6157                 if (qf_inums[type]) {
6158                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
6159                                 DQUOT_USAGE_ENABLED |
6160                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
6161                         if (err) {
6162                                 ext4_warning(sb,
6163                                         "Failed to enable quota tracking "
6164                                         "(type=%d, err=%d). Please run "
6165                                         "e2fsck to fix.", type, err);
6166                                 for (type--; type >= 0; type--)
6167                                         dquot_quota_off(sb, type);
6168
6169                                 return err;
6170                         }
6171                 }
6172         }
6173         return 0;
6174 }
6175
6176 static int ext4_quota_off(struct super_block *sb, int type)
6177 {
6178         struct inode *inode = sb_dqopt(sb)->files[type];
6179         handle_t *handle;
6180         int err;
6181
6182         /* Force all delayed allocation blocks to be allocated.
6183          * Caller already holds s_umount sem */
6184         if (test_opt(sb, DELALLOC))
6185                 sync_filesystem(sb);
6186
6187         if (!inode || !igrab(inode))
6188                 goto out;
6189
6190         err = dquot_quota_off(sb, type);
6191         if (err || ext4_has_feature_quota(sb))
6192                 goto out_put;
6193
6194         inode_lock(inode);
6195         /*
6196          * Update modification times of quota files when userspace can
6197          * start looking at them. If we fail, we return success anyway since
6198          * this is not a hard failure and quotas are already disabled.
6199          */
6200         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
6201         if (IS_ERR(handle)) {
6202                 err = PTR_ERR(handle);
6203                 goto out_unlock;
6204         }
6205         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
6206         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
6207         inode->i_mtime = inode->i_ctime = current_time(inode);
6208         err = ext4_mark_inode_dirty(handle, inode);
6209         ext4_journal_stop(handle);
6210 out_unlock:
6211         inode_unlock(inode);
6212 out_put:
6213         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
6214         iput(inode);
6215         return err;
6216 out:
6217         return dquot_quota_off(sb, type);
6218 }
6219
6220 /* Read data from quotafile - avoid pagecache and such because we cannot afford
6221  * acquiring the locks... As quota files are never truncated and quota code
6222  * itself serializes the operations (and no one else should touch the files)
6223  * we don't have to be afraid of races */
6224 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
6225                                size_t len, loff_t off)
6226 {
6227         struct inode *inode = sb_dqopt(sb)->files[type];
6228         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
6229         int offset = off & (sb->s_blocksize - 1);
6230         int tocopy;
6231         size_t toread;
6232         struct buffer_head *bh;
6233         loff_t i_size = i_size_read(inode);
6234
6235         if (off > i_size)
6236                 return 0;
6237         if (off+len > i_size)
6238                 len = i_size-off;
6239         toread = len;
6240         while (toread > 0) {
6241                 tocopy = sb->s_blocksize - offset < toread ?
6242                                 sb->s_blocksize - offset : toread;
6243                 bh = ext4_bread(NULL, inode, blk, 0);
6244                 if (IS_ERR(bh))
6245                         return PTR_ERR(bh);
6246                 if (!bh)        /* A hole? */
6247                         memset(data, 0, tocopy);
6248                 else
6249                         memcpy(data, bh->b_data+offset, tocopy);
6250                 brelse(bh);
6251                 offset = 0;
6252                 toread -= tocopy;
6253                 data += tocopy;
6254                 blk++;
6255         }
6256         return len;
6257 }
6258
6259 /* Write to quotafile (we know the transaction is already started and has
6260  * enough credits) */
6261 static ssize_t ext4_quota_write(struct super_block *sb, int type,
6262                                 const char *data, size_t len, loff_t off)
6263 {
6264         struct inode *inode = sb_dqopt(sb)->files[type];
6265         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
6266         int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1);
6267         int retries = 0;
6268         struct buffer_head *bh;
6269         handle_t *handle = journal_current_handle();
6270
6271         if (EXT4_SB(sb)->s_journal && !handle) {
6272                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
6273                         " cancelled because transaction is not started",
6274                         (unsigned long long)off, (unsigned long long)len);
6275                 return -EIO;
6276         }
6277         /*
6278          * Since we account only one data block in transaction credits,
6279          * then it is impossible to cross a block boundary.
6280          */
6281         if (sb->s_blocksize - offset < len) {
6282                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
6283                         " cancelled because not block aligned",
6284                         (unsigned long long)off, (unsigned long long)len);
6285                 return -EIO;
6286         }
6287
6288         do {
6289                 bh = ext4_bread(handle, inode, blk,
6290                                 EXT4_GET_BLOCKS_CREATE |
6291                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
6292         } while (PTR_ERR(bh) == -ENOSPC &&
6293                  ext4_should_retry_alloc(inode->i_sb, &retries));
6294         if (IS_ERR(bh))
6295                 return PTR_ERR(bh);
6296         if (!bh)
6297                 goto out;
6298         BUFFER_TRACE(bh, "get write access");
6299         err = ext4_journal_get_write_access(handle, bh);
6300         if (err) {
6301                 brelse(bh);
6302                 return err;
6303         }
6304         lock_buffer(bh);
6305         memcpy(bh->b_data+offset, data, len);
6306         flush_dcache_page(bh->b_page);
6307         unlock_buffer(bh);
6308         err = ext4_handle_dirty_metadata(handle, NULL, bh);
6309         brelse(bh);
6310 out:
6311         if (inode->i_size < off + len) {
6312                 i_size_write(inode, off + len);
6313                 EXT4_I(inode)->i_disksize = inode->i_size;
6314                 err2 = ext4_mark_inode_dirty(handle, inode);
6315                 if (unlikely(err2 && !err))
6316                         err = err2;
6317         }
6318         return err ? err : len;
6319 }
6320 #endif
6321
6322 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
6323                        const char *dev_name, void *data)
6324 {
6325         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
6326 }
6327
6328 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
6329 static inline void register_as_ext2(void)
6330 {
6331         int err = register_filesystem(&ext2_fs_type);
6332         if (err)
6333                 printk(KERN_WARNING
6334                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
6335 }
6336
6337 static inline void unregister_as_ext2(void)
6338 {
6339         unregister_filesystem(&ext2_fs_type);
6340 }
6341
6342 static inline int ext2_feature_set_ok(struct super_block *sb)
6343 {
6344         if (ext4_has_unknown_ext2_incompat_features(sb))
6345                 return 0;
6346         if (sb_rdonly(sb))
6347                 return 1;
6348         if (ext4_has_unknown_ext2_ro_compat_features(sb))
6349                 return 0;
6350         return 1;
6351 }
6352 #else
6353 static inline void register_as_ext2(void) { }
6354 static inline void unregister_as_ext2(void) { }
6355 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
6356 #endif
6357
6358 static inline void register_as_ext3(void)
6359 {
6360         int err = register_filesystem(&ext3_fs_type);
6361         if (err)
6362                 printk(KERN_WARNING
6363                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
6364 }
6365
6366 static inline void unregister_as_ext3(void)
6367 {
6368         unregister_filesystem(&ext3_fs_type);
6369 }
6370
6371 static inline int ext3_feature_set_ok(struct super_block *sb)
6372 {
6373         if (ext4_has_unknown_ext3_incompat_features(sb))
6374                 return 0;
6375         if (!ext4_has_feature_journal(sb))
6376                 return 0;
6377         if (sb_rdonly(sb))
6378                 return 1;
6379         if (ext4_has_unknown_ext3_ro_compat_features(sb))
6380                 return 0;
6381         return 1;
6382 }
6383
6384 static struct file_system_type ext4_fs_type = {
6385         .owner          = THIS_MODULE,
6386         .name           = "ext4",
6387         .mount          = ext4_mount,
6388         .kill_sb        = kill_block_super,
6389         .fs_flags       = FS_REQUIRES_DEV,
6390 };
6391 MODULE_ALIAS_FS("ext4");
6392
6393 /* Shared across all ext4 file systems */
6394 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
6395
6396 static int __init ext4_init_fs(void)
6397 {
6398         int i, err;
6399
6400         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
6401         ext4_li_info = NULL;
6402         mutex_init(&ext4_li_mtx);
6403
6404         /* Build-time check for flags consistency */
6405         ext4_check_flag_values();
6406
6407         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
6408                 init_waitqueue_head(&ext4__ioend_wq[i]);
6409
6410         err = ext4_init_es();
6411         if (err)
6412                 return err;
6413
6414         err = ext4_init_pending();
6415         if (err)
6416                 goto out7;
6417
6418         err = ext4_init_post_read_processing();
6419         if (err)
6420                 goto out6;
6421
6422         err = ext4_init_pageio();
6423         if (err)
6424                 goto out5;
6425
6426         err = ext4_init_system_zone();
6427         if (err)
6428                 goto out4;
6429
6430         err = ext4_init_sysfs();
6431         if (err)
6432                 goto out3;
6433
6434         err = ext4_init_mballoc();
6435         if (err)
6436                 goto out2;
6437         err = init_inodecache();
6438         if (err)
6439                 goto out1;
6440         register_as_ext3();
6441         register_as_ext2();
6442         err = register_filesystem(&ext4_fs_type);
6443         if (err)
6444                 goto out;
6445
6446         return 0;
6447 out:
6448         unregister_as_ext2();
6449         unregister_as_ext3();
6450         destroy_inodecache();
6451 out1:
6452         ext4_exit_mballoc();
6453 out2:
6454         ext4_exit_sysfs();
6455 out3:
6456         ext4_exit_system_zone();
6457 out4:
6458         ext4_exit_pageio();
6459 out5:
6460         ext4_exit_post_read_processing();
6461 out6:
6462         ext4_exit_pending();
6463 out7:
6464         ext4_exit_es();
6465
6466         return err;
6467 }
6468
6469 static void __exit ext4_exit_fs(void)
6470 {
6471         ext4_destroy_lazyinit_thread();
6472         unregister_as_ext2();
6473         unregister_as_ext3();
6474         unregister_filesystem(&ext4_fs_type);
6475         destroy_inodecache();
6476         ext4_exit_mballoc();
6477         ext4_exit_sysfs();
6478         ext4_exit_system_zone();
6479         ext4_exit_pageio();
6480         ext4_exit_post_read_processing();
6481         ext4_exit_es();
6482         ext4_exit_pending();
6483 }
6484
6485 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
6486 MODULE_DESCRIPTION("Fourth Extended Filesystem");
6487 MODULE_LICENSE("GPL");
6488 MODULE_SOFTDEP("pre: crc32c");
6489 module_init(ext4_init_fs)
6490 module_exit(ext4_exit_fs)