fs/ext4/super.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/ext4/super.c
   4  *
   5  * Copyright (C) 1992, 1993, 1994, 1995
   6  * Remy Card (card@masi.ibp.fr)
   7  * Laboratoire MASI - Institut Blaise Pascal
   8  * Universite Pierre et Marie Curie (Paris VI)
   9  *
  10  *  from
  11  *
  12  *  linux/fs/minix/inode.c
  13  *
  14  *  Copyright (C) 1991, 1992  Linus Torvalds
  15  *
  16  *  Big-endian to little-endian byte-swapping/bitmaps by
  17  *        David S. Miller (davem@caip.rutgers.edu), 1995
  18  */
  19
  20 #include <linux/module.h>
  21 #include <linux/string.h>
  22 #include <linux/fs.h>
  23 #include <linux/time.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/backing-dev.h>
  29 #include <linux/parser.h>
  30 #include <linux/buffer_head.h>
  31 #include <linux/exportfs.h>
  32 #include <linux/vfs.h>
  33 #include <linux/random.h>
  34 #include <linux/mount.h>
  35 #include <linux/namei.h>
  36 #include <linux/quotaops.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/dax.h>
  42 #include <linux/cleancache.h>
  43 #include <linux/uaccess.h>
  44 #include <linux/iversion.h>
  45 #include <linux/unicode.h>
  46 #include <linux/part_stat.h>
  47 #include <linux/kthread.h>
  48 #include <linux/freezer.h>
  49
  50 #include "ext4.h"
  51 #include "ext4_extents.h"       /* Needed for trace points definition */
  52 #include "ext4_jbd2.h"
  53 #include "xattr.h"
  54 #include "acl.h"
  55 #include "mballoc.h"
  56 #include "fsmap.h"
  57
  58 #define CREATE_TRACE_POINTS
  59 #include <trace/events/ext4.h>
  60
  61 static struct ext4_lazy_init *ext4_li_info;
  62 static struct mutex ext4_li_mtx;
  63 static struct ratelimit_state ext4_mount_msg_ratelimit;
  64
  65 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  66                              unsigned long journal_devnum);
  67 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  68 static int ext4_commit_super(struct super_block *sb, int sync);
  69 static 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 static inline void __ext4_read_bh(struct buffer_head *bh, int op_flags,
 146                                   bh_end_io_t *end_io)
 147 {
 148         /*
 149          * buffer's verified bit is no longer valid after reading from
 150          * disk again due to write out error, clear it to make sure we
 151          * recheck the buffer contents.
 152          */
 153         clear_buffer_verified(bh);
 154
 155         bh->b_end_io = end_io ? end_io : end_buffer_read_sync;
 156         get_bh(bh);
 157         submit_bh(REQ_OP_READ, op_flags, bh);
 158 }
 159
 160 void ext4_read_bh_nowait(struct buffer_head *bh, int op_flags,
 161                          bh_end_io_t *end_io)
 162 {
 163         BUG_ON(!buffer_locked(bh));
 164
 165         if (ext4_buffer_uptodate(bh)) {
 166                 unlock_buffer(bh);
 167                 return;
 168         }
 169         __ext4_read_bh(bh, op_flags, end_io);
 170 }
 171
 172 int ext4_read_bh(struct buffer_head *bh, int op_flags, bh_end_io_t *end_io)
 173 {
 174         BUG_ON(!buffer_locked(bh));
 175
 176         if (ext4_buffer_uptodate(bh)) {
 177                 unlock_buffer(bh);
 178                 return 0;
 179         }
 180
 181         __ext4_read_bh(bh, op_flags, end_io);
 182
 183         wait_on_buffer(bh);
 184         if (buffer_uptodate(bh))
 185                 return 0;
 186         return -EIO;
 187 }
 188
 189 int ext4_read_bh_lock(struct buffer_head *bh, int op_flags, bool wait)
 190 {
 191         if (trylock_buffer(bh)) {
 192                 if (wait)
 193                         return ext4_read_bh(bh, op_flags, NULL);
 194                 ext4_read_bh_nowait(bh, op_flags, NULL);
 195                 return 0;
 196         }
 197         if (wait) {
 198                 wait_on_buffer(bh);
 199                 if (buffer_uptodate(bh))
 200                         return 0;
 201                 return -EIO;
 202         }
 203         return 0;
 204 }
 205
 206 /*
 207  * This works like sb_bread() except it uses ERR_PTR for error
 208  * returns.  Currently with sb_bread it's impossible to distinguish
 209  * between ENOMEM and EIO situations (since both result in a NULL
 210  * return.
 211  */
 212 struct buffer_head *
 213 ext4_sb_bread(struct super_block *sb, sector_t block, int op_flags)
 214 {
 215         struct buffer_head *bh = sb_getblk(sb, block);
 216
 217         if (bh == NULL)
 218                 return ERR_PTR(-ENOMEM);
 219         if (ext4_buffer_uptodate(bh))
 220                 return bh;
 221         clear_buffer_verified(bh);
 222         ll_rw_block(REQ_OP_READ, REQ_META | op_flags, 1, &bh);
 223         wait_on_buffer(bh);
 224         if (buffer_uptodate(bh))
 225                 return bh;
 226         put_bh(bh);
 227         return ERR_PTR(-EIO);
 228 }
 229
 230 static int ext4_verify_csum_type(struct super_block *sb,
 231                                  struct ext4_super_block *es)
 232 {
 233         if (!ext4_has_feature_metadata_csum(sb))
 234                 return 1;
 235
 236         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 237 }
 238
 239 static __le32 ext4_superblock_csum(struct super_block *sb,
 240                                    struct ext4_super_block *es)
 241 {
 242         struct ext4_sb_info *sbi = EXT4_SB(sb);
 243         int offset = offsetof(struct ext4_super_block, s_checksum);
 244         __u32 csum;
 245
 246         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 247
 248         return cpu_to_le32(csum);
 249 }
 250
 251 static int ext4_superblock_csum_verify(struct super_block *sb,
 252                                        struct ext4_super_block *es)
 253 {
 254         if (!ext4_has_metadata_csum(sb))
 255                 return 1;
 256
 257         return es->s_checksum == ext4_superblock_csum(sb, es);
 258 }
 259
 260 void ext4_superblock_csum_set(struct super_block *sb)
 261 {
 262         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 263
 264         if (!ext4_has_metadata_csum(sb))
 265                 return;
 266
 267         /*
 268          * Locking the superblock prevents the scenario
 269          * where:
 270          *  1) a first thread pauses during checksum calculation.
 271          *  2) a second thread updates the superblock, recalculates
 272          *     the checksum, and updates s_checksum
 273          *  3) the first thread resumes and finishes its checksum calculation
 274          *     and updates s_checksum with a potentially stale or torn value.
 275          */
 276         lock_buffer(EXT4_SB(sb)->s_sbh);
 277         es->s_checksum = ext4_superblock_csum(sb, es);
 278         unlock_buffer(EXT4_SB(sb)->s_sbh);
 279 }
 280
 281 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 282                                struct ext4_group_desc *bg)
 283 {
 284         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 285                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 286                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 287 }
 288
 289 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 290                                struct ext4_group_desc *bg)
 291 {
 292         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 293                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 294                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 295 }
 296
 297 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 298                               struct ext4_group_desc *bg)
 299 {
 300         return le32_to_cpu(bg->bg_inode_table_lo) |
 301                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 302                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 303 }
 304
 305 __u32 ext4_free_group_clusters(struct super_block *sb,
 306                                struct ext4_group_desc *bg)
 307 {
 308         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 309                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 310                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 311 }
 312
 313 __u32 ext4_free_inodes_count(struct super_block *sb,
 314                               struct ext4_group_desc *bg)
 315 {
 316         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 317                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 318                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 319 }
 320
 321 __u32 ext4_used_dirs_count(struct super_block *sb,
 322                               struct ext4_group_desc *bg)
 323 {
 324         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 325                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 326                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 327 }
 328
 329 __u32 ext4_itable_unused_count(struct super_block *sb,
 330                               struct ext4_group_desc *bg)
 331 {
 332         return le16_to_cpu(bg->bg_itable_unused_lo) |
 333                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 334                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 335 }
 336
 337 void ext4_block_bitmap_set(struct super_block *sb,
 338                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 339 {
 340         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 341         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 342                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 343 }
 344
 345 void ext4_inode_bitmap_set(struct super_block *sb,
 346                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 347 {
 348         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 349         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 350                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 351 }
 352
 353 void ext4_inode_table_set(struct super_block *sb,
 354                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 355 {
 356         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 357         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 358                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 359 }
 360
 361 void ext4_free_group_clusters_set(struct super_block *sb,
 362                                   struct ext4_group_desc *bg, __u32 count)
 363 {
 364         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 365         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 366                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 367 }
 368
 369 void ext4_free_inodes_set(struct super_block *sb,
 370                           struct ext4_group_desc *bg, __u32 count)
 371 {
 372         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 373         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 374                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 375 }
 376
 377 void ext4_used_dirs_set(struct super_block *sb,
 378                           struct ext4_group_desc *bg, __u32 count)
 379 {
 380         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 381         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 382                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 383 }
 384
 385 void ext4_itable_unused_set(struct super_block *sb,
 386                           struct ext4_group_desc *bg, __u32 count)
 387 {
 388         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 389         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 390                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 391 }
 392
 393 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi)
 394 {
 395         time64_t now = ktime_get_real_seconds();
 396
 397         now = clamp_val(now, 0, (1ull << 40) - 1);
 398
 399         *lo = cpu_to_le32(lower_32_bits(now));
 400         *hi = upper_32_bits(now);
 401 }
 402
 403 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi)
 404 {
 405         return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo);
 406 }
 407 #define ext4_update_tstamp(es, tstamp) \
 408         __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
 409 #define ext4_get_tstamp(es, tstamp) \
 410         __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
 411
 412 static void __save_error_info(struct super_block *sb, int error,
 413                               __u32 ino, __u64 block,
 414                               const char *func, unsigned int line)
 415 {
 416         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 417         int err;
 418
 419         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 420         if (bdev_read_only(sb->s_bdev))
 421                 return;
 422         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 423         ext4_update_tstamp(es, s_last_error_time);
 424         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 425         es->s_last_error_line = cpu_to_le32(line);
 426         es->s_last_error_ino = cpu_to_le32(ino);
 427         es->s_last_error_block = cpu_to_le64(block);
 428         switch (error) {
 429         case EIO:
 430                 err = EXT4_ERR_EIO;
 431                 break;
 432         case ENOMEM:
 433                 err = EXT4_ERR_ENOMEM;
 434                 break;
 435         case EFSBADCRC:
 436                 err = EXT4_ERR_EFSBADCRC;
 437                 break;
 438         case 0:
 439         case EFSCORRUPTED:
 440                 err = EXT4_ERR_EFSCORRUPTED;
 441                 break;
 442         case ENOSPC:
 443                 err = EXT4_ERR_ENOSPC;
 444                 break;
 445         case ENOKEY:
 446                 err = EXT4_ERR_ENOKEY;
 447                 break;
 448         case EROFS:
 449                 err = EXT4_ERR_EROFS;
 450                 break;
 451         case EFBIG:
 452                 err = EXT4_ERR_EFBIG;
 453                 break;
 454         case EEXIST:
 455                 err = EXT4_ERR_EEXIST;
 456                 break;
 457         case ERANGE:
 458                 err = EXT4_ERR_ERANGE;
 459                 break;
 460         case EOVERFLOW:
 461                 err = EXT4_ERR_EOVERFLOW;
 462                 break;
 463         case EBUSY:
 464                 err = EXT4_ERR_EBUSY;
 465                 break;
 466         case ENOTDIR:
 467                 err = EXT4_ERR_ENOTDIR;
 468                 break;
 469         case ENOTEMPTY:
 470                 err = EXT4_ERR_ENOTEMPTY;
 471                 break;
 472         case ESHUTDOWN:
 473                 err = EXT4_ERR_ESHUTDOWN;
 474                 break;
 475         case EFAULT:
 476                 err = EXT4_ERR_EFAULT;
 477                 break;
 478         default:
 479                 err = EXT4_ERR_UNKNOWN;
 480         }
 481         es->s_last_error_errcode = err;
 482         if (!es->s_first_error_time) {
 483                 es->s_first_error_time = es->s_last_error_time;
 484                 es->s_first_error_time_hi = es->s_last_error_time_hi;
 485                 strncpy(es->s_first_error_func, func,
 486                         sizeof(es->s_first_error_func));
 487                 es->s_first_error_line = cpu_to_le32(line);
 488                 es->s_first_error_ino = es->s_last_error_ino;
 489                 es->s_first_error_block = es->s_last_error_block;
 490                 es->s_first_error_errcode = es->s_last_error_errcode;
 491         }
 492         /*
 493          * Start the daily error reporting function if it hasn't been
 494          * started already
 495          */
 496         if (!es->s_error_count)
 497                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 498         le32_add_cpu(&es->s_error_count, 1);
 499 }
 500
 501 static void save_error_info(struct super_block *sb, int error,
 502                             __u32 ino, __u64 block,
 503                             const char *func, unsigned int line)
 504 {
 505         __save_error_info(sb, error, ino, block, func, line);
 506         if (!bdev_read_only(sb->s_bdev))
 507                 ext4_commit_super(sb, 1);
 508 }
 509
 510 /*
 511  * The del_gendisk() function uninitializes the disk-specific data
 512  * structures, including the bdi structure, without telling anyone
 513  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 514  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 515  * This is a kludge to prevent these oops until we can put in a proper
 516  * hook in del_gendisk() to inform the VFS and file system layers.
 517  */
 518 static int block_device_ejected(struct super_block *sb)
 519 {
 520         struct inode *bd_inode = sb->s_bdev->bd_inode;
 521         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 522
 523         return bdi->dev == NULL;
 524 }
 525
 526 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 527 {
 528         struct super_block              *sb = journal->j_private;
 529         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 530         int                             error = is_journal_aborted(journal);
 531         struct ext4_journal_cb_entry    *jce;
 532
 533         BUG_ON(txn->t_state == T_FINISHED);
 534
 535         ext4_process_freed_data(sb, txn->t_tid);
 536
 537         spin_lock(&sbi->s_md_lock);
 538         while (!list_empty(&txn->t_private_list)) {
 539                 jce = list_entry(txn->t_private_list.next,
 540                                  struct ext4_journal_cb_entry, jce_list);
 541                 list_del_init(&jce->jce_list);
 542                 spin_unlock(&sbi->s_md_lock);
 543                 jce->jce_func(sb, jce, error);
 544                 spin_lock(&sbi->s_md_lock);
 545         }
 546         spin_unlock(&sbi->s_md_lock);
 547 }
 548
 549 static bool system_going_down(void)
 550 {
 551         return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF
 552                 || system_state == SYSTEM_RESTART;
 553 }
 554
 555 /* Deal with the reporting of failure conditions on a filesystem such as
 556  * inconsistencies detected or read IO failures.
 557  *
 558  * On ext2, we can store the error state of the filesystem in the
 559  * superblock.  That is not possible on ext4, because we may have other
 560  * write ordering constraints on the superblock which prevent us from
 561  * writing it out straight away; and given that the journal is about to
 562  * be aborted, we can't rely on the current, or future, transactions to
 563  * write out the superblock safely.
 564  *
 565  * We'll just use the jbd2_journal_abort() error code to record an error in
 566  * the journal instead.  On recovery, the journal will complain about
 567  * that error until we've noted it down and cleared it.
 568  */
 569
 570 static void ext4_handle_error(struct super_block *sb)
 571 {
 572         if (test_opt(sb, WARN_ON_ERROR))
 573                 WARN_ON_ONCE(1);
 574
 575         if (sb_rdonly(sb))
 576                 return;
 577
 578         if (!test_opt(sb, ERRORS_CONT)) {
 579                 journal_t *journal = EXT4_SB(sb)->s_journal;
 580
 581                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 582                 if (journal)
 583                         jbd2_journal_abort(journal, -EIO);
 584         }
 585         /*
 586          * We force ERRORS_RO behavior when system is rebooting. Otherwise we
 587          * could panic during 'reboot -f' as the underlying device got already
 588          * disabled.
 589          */
 590         if (test_opt(sb, ERRORS_RO) || system_going_down()) {
 591                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 592                 /*
 593                  * Make sure updated value of ->s_mount_flags will be visible
 594                  * before ->s_flags update
 595                  */
 596                 smp_wmb();
 597                 sb->s_flags |= SB_RDONLY;
 598         } else if (test_opt(sb, ERRORS_PANIC)) {
 599                 panic("EXT4-fs (device %s): panic forced after error\n",
 600                         sb->s_id);
 601         }
 602 }
 603
 604 #define ext4_error_ratelimit(sb)                                        \
 605                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 606                              "EXT4-fs error")
 607
 608 void __ext4_error(struct super_block *sb, const char *function,
 609                   unsigned int line, int error, __u64 block,
 610                   const char *fmt, ...)
 611 {
 612         struct va_format vaf;
 613         va_list args;
 614
 615         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 616                 return;
 617
 618         trace_ext4_error(sb, function, line);
 619         if (ext4_error_ratelimit(sb)) {
 620                 va_start(args, fmt);
 621                 vaf.fmt = fmt;
 622                 vaf.va = &args;
 623                 printk(KERN_CRIT
 624                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 625                        sb->s_id, function, line, current->comm, &vaf);
 626                 va_end(args);
 627         }
 628         save_error_info(sb, error, 0, block, function, line);
 629         ext4_handle_error(sb);
 630 }
 631
 632 void __ext4_error_inode(struct inode *inode, const char *function,
 633                         unsigned int line, ext4_fsblk_t block, int error,
 634                         const char *fmt, ...)
 635 {
 636         va_list args;
 637         struct va_format vaf;
 638
 639         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 640                 return;
 641
 642         trace_ext4_error(inode->i_sb, function, line);
 643         if (ext4_error_ratelimit(inode->i_sb)) {
 644                 va_start(args, fmt);
 645                 vaf.fmt = fmt;
 646                 vaf.va = &args;
 647                 if (block)
 648                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 649                                "inode #%lu: block %llu: comm %s: %pV\n",
 650                                inode->i_sb->s_id, function, line, inode->i_ino,
 651                                block, current->comm, &vaf);
 652                 else
 653                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 654                                "inode #%lu: comm %s: %pV\n",
 655                                inode->i_sb->s_id, function, line, inode->i_ino,
 656                                current->comm, &vaf);
 657                 va_end(args);
 658         }
 659         save_error_info(inode->i_sb, error, inode->i_ino, block,
 660                         function, line);
 661         ext4_handle_error(inode->i_sb);
 662 }
 663
 664 void __ext4_error_file(struct file *file, const char *function,
 665                        unsigned int line, ext4_fsblk_t block,
 666                        const char *fmt, ...)
 667 {
 668         va_list args;
 669         struct va_format vaf;
 670         struct inode *inode = file_inode(file);
 671         char pathname[80], *path;
 672
 673         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 674                 return;
 675
 676         trace_ext4_error(inode->i_sb, function, line);
 677         if (ext4_error_ratelimit(inode->i_sb)) {
 678                 path = file_path(file, pathname, sizeof(pathname));
 679                 if (IS_ERR(path))
 680                         path = "(unknown)";
 681                 va_start(args, fmt);
 682                 vaf.fmt = fmt;
 683                 vaf.va = &args;
 684                 if (block)
 685                         printk(KERN_CRIT
 686                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 687                                "block %llu: comm %s: path %s: %pV\n",
 688                                inode->i_sb->s_id, function, line, inode->i_ino,
 689                                block, current->comm, path, &vaf);
 690                 else
 691                         printk(KERN_CRIT
 692                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 693                                "comm %s: path %s: %pV\n",
 694                                inode->i_sb->s_id, function, line, inode->i_ino,
 695                                current->comm, path, &vaf);
 696                 va_end(args);
 697         }
 698         save_error_info(inode->i_sb, EFSCORRUPTED, inode->i_ino, block,
 699                         function, line);
 700         ext4_handle_error(inode->i_sb);
 701 }
 702
 703 const char *ext4_decode_error(struct super_block *sb, int errno,
 704                               char nbuf[16])
 705 {
 706         char *errstr = NULL;
 707
 708         switch (errno) {
 709         case -EFSCORRUPTED:
 710                 errstr = "Corrupt filesystem";
 711                 break;
 712         case -EFSBADCRC:
 713                 errstr = "Filesystem failed CRC";
 714                 break;
 715         case -EIO:
 716                 errstr = "IO failure";
 717                 break;
 718         case -ENOMEM:
 719                 errstr = "Out of memory";
 720                 break;
 721         case -EROFS:
 722                 if (!sb || (EXT4_SB(sb)->s_journal &&
 723                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 724                         errstr = "Journal has aborted";
 725                 else
 726                         errstr = "Readonly filesystem";
 727                 break;
 728         default:
 729                 /* If the caller passed in an extra buffer for unknown
 730                  * errors, textualise them now.  Else we just return
 731                  * NULL. */
 732                 if (nbuf) {
 733                         /* Check for truncated error codes... */
 734                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 735                                 errstr = nbuf;
 736                 }
 737                 break;
 738         }
 739
 740         return errstr;
 741 }
 742
 743 /* __ext4_std_error decodes expected errors from journaling functions
 744  * automatically and invokes the appropriate error response.  */
 745
 746 void __ext4_std_error(struct super_block *sb, const char *function,
 747                       unsigned int line, int errno)
 748 {
 749         char nbuf[16];
 750         const char *errstr;
 751
 752         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 753                 return;
 754
 755         /* Special case: if the error is EROFS, and we're not already
 756          * inside a transaction, then there's really no point in logging
 757          * an error. */
 758         if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb))
 759                 return;
 760
 761         if (ext4_error_ratelimit(sb)) {
 762                 errstr = ext4_decode_error(sb, errno, nbuf);
 763                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 764                        sb->s_id, function, line, errstr);
 765         }
 766
 767         save_error_info(sb, -errno, 0, 0, function, line);
 768         ext4_handle_error(sb);
 769 }
 770
 771 /*
 772  * ext4_abort is a much stronger failure handler than ext4_error.  The
 773  * abort function may be used to deal with unrecoverable failures such
 774  * as journal IO errors or ENOMEM at a critical moment in log management.
 775  *
 776  * We unconditionally force the filesystem into an ABORT|READONLY state,
 777  * unless the error response on the fs has been set to panic in which
 778  * case we take the easy way out and panic immediately.
 779  */
 780
 781 void __ext4_abort(struct super_block *sb, const char *function,
 782                   unsigned int line, int error, const char *fmt, ...)
 783 {
 784         struct va_format vaf;
 785         va_list args;
 786
 787         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 788                 return;
 789
 790         save_error_info(sb, error, 0, 0, function, line);
 791         va_start(args, fmt);
 792         vaf.fmt = fmt;
 793         vaf.va = &args;
 794         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: %pV\n",
 795                sb->s_id, function, line, &vaf);
 796         va_end(args);
 797
 798         if (sb_rdonly(sb) == 0) {
 799                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 800                 if (EXT4_SB(sb)->s_journal)
 801                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 802
 803                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 804                 /*
 805                  * Make sure updated value of ->s_mount_flags will be visible
 806                  * before ->s_flags update
 807                  */
 808                 smp_wmb();
 809                 sb->s_flags |= SB_RDONLY;
 810         }
 811         if (test_opt(sb, ERRORS_PANIC) && !system_going_down())
 812                 panic("EXT4-fs panic from previous error\n");
 813 }
 814
 815 void __ext4_msg(struct super_block *sb,
 816                 const char *prefix, const char *fmt, ...)
 817 {
 818         struct va_format vaf;
 819         va_list args;
 820
 821         atomic_inc(&EXT4_SB(sb)->s_msg_count);
 822         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 823                 return;
 824
 825         va_start(args, fmt);
 826         vaf.fmt = fmt;
 827         vaf.va = &args;
 828         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 829         va_end(args);
 830 }
 831
 832 static int ext4_warning_ratelimit(struct super_block *sb)
 833 {
 834         atomic_inc(&EXT4_SB(sb)->s_warning_count);
 835         return ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state),
 836                             "EXT4-fs warning");
 837 }
 838
 839 void __ext4_warning(struct super_block *sb, const char *function,
 840                     unsigned int line, const char *fmt, ...)
 841 {
 842         struct va_format vaf;
 843         va_list args;
 844
 845         if (!ext4_warning_ratelimit(sb))
 846                 return;
 847
 848         va_start(args, fmt);
 849         vaf.fmt = fmt;
 850         vaf.va = &args;
 851         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 852                sb->s_id, function, line, &vaf);
 853         va_end(args);
 854 }
 855
 856 void __ext4_warning_inode(const struct inode *inode, const char *function,
 857                           unsigned int line, const char *fmt, ...)
 858 {
 859         struct va_format vaf;
 860         va_list args;
 861
 862         if (!ext4_warning_ratelimit(inode->i_sb))
 863                 return;
 864
 865         va_start(args, fmt);
 866         vaf.fmt = fmt;
 867         vaf.va = &args;
 868         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 869                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 870                function, line, inode->i_ino, current->comm, &vaf);
 871         va_end(args);
 872 }
 873
 874 void __ext4_grp_locked_error(const char *function, unsigned int line,
 875                              struct super_block *sb, ext4_group_t grp,
 876                              unsigned long ino, ext4_fsblk_t block,
 877                              const char *fmt, ...)
 878 __releases(bitlock)
 879 __acquires(bitlock)
 880 {
 881         struct va_format vaf;
 882         va_list args;
 883
 884         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 885                 return;
 886
 887         trace_ext4_error(sb, function, line);
 888         __save_error_info(sb, EFSCORRUPTED, ino, block, function, line);
 889
 890         if (ext4_error_ratelimit(sb)) {
 891                 va_start(args, fmt);
 892                 vaf.fmt = fmt;
 893                 vaf.va = &args;
 894                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 895                        sb->s_id, function, line, grp);
 896                 if (ino)
 897                         printk(KERN_CONT "inode %lu: ", ino);
 898                 if (block)
 899                         printk(KERN_CONT "block %llu:",
 900                                (unsigned long long) block);
 901                 printk(KERN_CONT "%pV\n", &vaf);
 902                 va_end(args);
 903         }
 904
 905         if (test_opt(sb, WARN_ON_ERROR))
 906                 WARN_ON_ONCE(1);
 907
 908         if (test_opt(sb, ERRORS_CONT)) {
 909                 ext4_commit_super(sb, 0);
 910                 return;
 911         }
 912
 913         ext4_unlock_group(sb, grp);
 914         ext4_commit_super(sb, 1);
 915         ext4_handle_error(sb);
 916         /*
 917          * We only get here in the ERRORS_RO case; relocking the group
 918          * may be dangerous, but nothing bad will happen since the
 919          * filesystem will have already been marked read/only and the
 920          * journal has been aborted.  We return 1 as a hint to callers
 921          * who might what to use the return value from
 922          * ext4_grp_locked_error() to distinguish between the
 923          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 924          * aggressively from the ext4 function in question, with a
 925          * more appropriate error code.
 926          */
 927         ext4_lock_group(sb, grp);
 928         return;
 929 }
 930
 931 void ext4_mark_group_bitmap_corrupted(struct super_block *sb,
 932                                      ext4_group_t group,
 933                                      unsigned int flags)
 934 {
 935         struct ext4_sb_info *sbi = EXT4_SB(sb);
 936         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
 937         struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
 938         int ret;
 939
 940         if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) {
 941                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT,
 942                                             &grp->bb_state);
 943                 if (!ret)
 944                         percpu_counter_sub(&sbi->s_freeclusters_counter,
 945                                            grp->bb_free);
 946         }
 947
 948         if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) {
 949                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT,
 950                                             &grp->bb_state);
 951                 if (!ret && gdp) {
 952                         int count;
 953
 954                         count = ext4_free_inodes_count(sb, gdp);
 955                         percpu_counter_sub(&sbi->s_freeinodes_counter,
 956                                            count);
 957                 }
 958         }
 959 }
 960
 961 void ext4_update_dynamic_rev(struct super_block *sb)
 962 {
 963         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 964
 965         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 966                 return;
 967
 968         ext4_warning(sb,
 969                      "updating to rev %d because of new feature flag, "
 970                      "running e2fsck is recommended",
 971                      EXT4_DYNAMIC_REV);
 972
 973         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 974         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 975         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 976         /* leave es->s_feature_*compat flags alone */
 977         /* es->s_uuid will be set by e2fsck if empty */
 978
 979         /*
 980          * The rest of the superblock fields should be zero, and if not it
 981          * means they are likely already in use, so leave them alone.  We
 982          * can leave it up to e2fsck to clean up any inconsistencies there.
 983          */
 984 }
 985
 986 /*
 987  * Open the external journal device
 988  */
 989 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 990 {
 991         struct block_device *bdev;
 992
 993         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 994         if (IS_ERR(bdev))
 995                 goto fail;
 996         return bdev;
 997
 998 fail:
 999         ext4_msg(sb, KERN_ERR,
1000                  "failed to open journal device unknown-block(%u,%u) %ld",
1001                  MAJOR(dev), MINOR(dev), PTR_ERR(bdev));
1002         return NULL;
1003 }
1004
1005 /*
1006  * Release the journal device
1007  */
1008 static void ext4_blkdev_put(struct block_device *bdev)
1009 {
1010         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1011 }
1012
1013 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
1014 {
1015         struct block_device *bdev;
1016         bdev = sbi->s_journal_bdev;
1017         if (bdev) {
1018                 ext4_blkdev_put(bdev);
1019                 sbi->s_journal_bdev = NULL;
1020         }
1021 }
1022
1023 static inline struct inode *orphan_list_entry(struct list_head *l)
1024 {
1025         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
1026 }
1027
1028 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
1029 {
1030         struct list_head *l;
1031
1032         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
1033                  le32_to_cpu(sbi->s_es->s_last_orphan));
1034
1035         printk(KERN_ERR "sb_info orphan list:\n");
1036         list_for_each(l, &sbi->s_orphan) {
1037                 struct inode *inode = orphan_list_entry(l);
1038                 printk(KERN_ERR "  "
1039                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
1040                        inode->i_sb->s_id, inode->i_ino, inode,
1041                        inode->i_mode, inode->i_nlink,
1042                        NEXT_ORPHAN(inode));
1043         }
1044 }
1045
1046 #ifdef CONFIG_QUOTA
1047 static int ext4_quota_off(struct super_block *sb, int type);
1048
1049 static inline void ext4_quota_off_umount(struct super_block *sb)
1050 {
1051         int type;
1052
1053         /* Use our quota_off function to clear inode flags etc. */
1054         for (type = 0; type < EXT4_MAXQUOTAS; type++)
1055                 ext4_quota_off(sb, type);
1056 }
1057
1058 /*
1059  * This is a helper function which is used in the mount/remount
1060  * codepaths (which holds s_umount) to fetch the quota file name.
1061  */
1062 static inline char *get_qf_name(struct super_block *sb,
1063                                 struct ext4_sb_info *sbi,
1064                                 int type)
1065 {
1066         return rcu_dereference_protected(sbi->s_qf_names[type],
1067                                          lockdep_is_held(&sb->s_umount));
1068 }
1069 #else
1070 static inline void ext4_quota_off_umount(struct super_block *sb)
1071 {
1072 }
1073 #endif
1074
1075 static void ext4_put_super(struct super_block *sb)
1076 {
1077         struct ext4_sb_info *sbi = EXT4_SB(sb);
1078         struct ext4_super_block *es = sbi->s_es;
1079         struct buffer_head **group_desc;
1080         struct flex_groups **flex_groups;
1081         int aborted = 0;
1082         int i, err;
1083
1084         ext4_unregister_li_request(sb);
1085         ext4_quota_off_umount(sb);
1086
1087         destroy_workqueue(sbi->rsv_conversion_wq);
1088
1089         /*
1090          * Unregister sysfs before destroying jbd2 journal.
1091          * Since we could still access attr_journal_task attribute via sysfs
1092          * path which could have sbi->s_journal->j_task as NULL
1093          */
1094         ext4_unregister_sysfs(sb);
1095
1096         if (sbi->s_journal) {
1097                 aborted = is_journal_aborted(sbi->s_journal);
1098                 err = jbd2_journal_destroy(sbi->s_journal);
1099                 sbi->s_journal = NULL;
1100                 if ((err < 0) && !aborted) {
1101                         ext4_abort(sb, -err, "Couldn't clean up the journal");
1102                 }
1103         }
1104
1105         ext4_es_unregister_shrinker(sbi);
1106         del_timer_sync(&sbi->s_err_report);
1107         ext4_release_system_zone(sb);
1108         ext4_mb_release(sb);
1109         ext4_ext_release(sb);
1110
1111         if (!sb_rdonly(sb) && !aborted) {
1112                 ext4_clear_feature_journal_needs_recovery(sb);
1113                 es->s_state = cpu_to_le16(sbi->s_mount_state);
1114         }
1115         if (!sb_rdonly(sb))
1116                 ext4_commit_super(sb, 1);
1117
1118         rcu_read_lock();
1119         group_desc = rcu_dereference(sbi->s_group_desc);
1120         for (i = 0; i < sbi->s_gdb_count; i++)
1121                 brelse(group_desc[i]);
1122         kvfree(group_desc);
1123         flex_groups = rcu_dereference(sbi->s_flex_groups);
1124         if (flex_groups) {
1125                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
1126                         kvfree(flex_groups[i]);
1127                 kvfree(flex_groups);
1128         }
1129         rcu_read_unlock();
1130         percpu_counter_destroy(&sbi->s_freeclusters_counter);
1131         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1132         percpu_counter_destroy(&sbi->s_dirs_counter);
1133         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
1134         percpu_free_rwsem(&sbi->s_writepages_rwsem);
1135 #ifdef CONFIG_QUOTA
1136         for (i = 0; i < EXT4_MAXQUOTAS; i++)
1137                 kfree(get_qf_name(sb, sbi, i));
1138 #endif
1139
1140         /* Debugging code just in case the in-memory inode orphan list
1141          * isn't empty.  The on-disk one can be non-empty if we've
1142          * detected an error and taken the fs readonly, but the
1143          * in-memory list had better be clean by this point. */
1144         if (!list_empty(&sbi->s_orphan))
1145                 dump_orphan_list(sb, sbi);
1146         J_ASSERT(list_empty(&sbi->s_orphan));
1147
1148         sync_blockdev(sb->s_bdev);
1149         invalidate_bdev(sb->s_bdev);
1150         if (sbi->s_journal_bdev && sbi->s_journal_bdev != sb->s_bdev) {
1151                 /*
1152                  * Invalidate the journal device's buffers.  We don't want them
1153                  * floating about in memory - the physical journal device may
1154                  * hotswapped, and it breaks the `ro-after' testing code.
1155                  */
1156                 sync_blockdev(sbi->s_journal_bdev);
1157                 invalidate_bdev(sbi->s_journal_bdev);
1158                 ext4_blkdev_remove(sbi);
1159         }
1160
1161         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
1162         sbi->s_ea_inode_cache = NULL;
1163
1164         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
1165         sbi->s_ea_block_cache = NULL;
1166
1167         if (sbi->s_mmp_tsk)
1168                 kthread_stop(sbi->s_mmp_tsk);
1169         brelse(sbi->s_sbh);
1170         sb->s_fs_info = NULL;
1171         /*
1172          * Now that we are completely done shutting down the
1173          * superblock, we need to actually destroy the kobject.
1174          */
1175         kobject_put(&sbi->s_kobj);
1176         wait_for_completion(&sbi->s_kobj_unregister);
1177         if (sbi->s_chksum_driver)
1178                 crypto_free_shash(sbi->s_chksum_driver);
1179         kfree(sbi->s_blockgroup_lock);
1180         fs_put_dax(sbi->s_daxdev);
1181         fscrypt_free_dummy_context(&sbi->s_dummy_enc_ctx);
1182 #ifdef CONFIG_UNICODE
1183         utf8_unload(sbi->s_encoding);
1184 #endif
1185         kfree(sbi);
1186 }
1187
1188 static struct kmem_cache *ext4_inode_cachep;
1189
1190 /*
1191  * Called inside transaction, so use GFP_NOFS
1192  */
1193 static struct inode *ext4_alloc_inode(struct super_block *sb)
1194 {
1195         struct ext4_inode_info *ei;
1196
1197         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
1198         if (!ei)
1199                 return NULL;
1200
1201         inode_set_iversion(&ei->vfs_inode, 1);
1202         spin_lock_init(&ei->i_raw_lock);
1203         INIT_LIST_HEAD(&ei->i_prealloc_list);
1204         atomic_set(&ei->i_prealloc_active, 0);
1205         spin_lock_init(&ei->i_prealloc_lock);
1206         ext4_es_init_tree(&ei->i_es_tree);
1207         rwlock_init(&ei->i_es_lock);
1208         INIT_LIST_HEAD(&ei->i_es_list);
1209         ei->i_es_all_nr = 0;
1210         ei->i_es_shk_nr = 0;
1211         ei->i_es_shrink_lblk = 0;
1212         ei->i_reserved_data_blocks = 0;
1213         spin_lock_init(&(ei->i_block_reservation_lock));
1214         ext4_init_pending_tree(&ei->i_pending_tree);
1215 #ifdef CONFIG_QUOTA
1216         ei->i_reserved_quota = 0;
1217         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
1218 #endif
1219         ei->jinode = NULL;
1220         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
1221         spin_lock_init(&ei->i_completed_io_lock);
1222         ei->i_sync_tid = 0;
1223         ei->i_datasync_tid = 0;
1224         atomic_set(&ei->i_unwritten, 0);
1225         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
1226         return &ei->vfs_inode;
1227 }
1228
1229 static int ext4_drop_inode(struct inode *inode)
1230 {
1231         int drop = generic_drop_inode(inode);
1232
1233         if (!drop)
1234                 drop = fscrypt_drop_inode(inode);
1235
1236         trace_ext4_drop_inode(inode, drop);
1237         return drop;
1238 }
1239
1240 static void ext4_free_in_core_inode(struct inode *inode)
1241 {
1242         fscrypt_free_inode(inode);
1243         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1244 }
1245
1246 static void ext4_destroy_inode(struct inode *inode)
1247 {
1248         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1249                 ext4_msg(inode->i_sb, KERN_ERR,
1250                          "Inode %lu (%p): orphan list check failed!",
1251                          inode->i_ino, EXT4_I(inode));
1252                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1253                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1254                                 true);
1255                 dump_stack();
1256         }
1257 }
1258
1259 static void init_once(void *foo)
1260 {
1261         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1262
1263         INIT_LIST_HEAD(&ei->i_orphan);
1264         init_rwsem(&ei->xattr_sem);
1265         init_rwsem(&ei->i_data_sem);
1266         init_rwsem(&ei->i_mmap_sem);
1267         inode_init_once(&ei->vfs_inode);
1268 }
1269
1270 static int __init init_inodecache(void)
1271 {
1272         ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache",
1273                                 sizeof(struct ext4_inode_info), 0,
1274                                 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
1275                                         SLAB_ACCOUNT),
1276                                 offsetof(struct ext4_inode_info, i_data),
1277                                 sizeof_field(struct ext4_inode_info, i_data),
1278                                 init_once);
1279         if (ext4_inode_cachep == NULL)
1280                 return -ENOMEM;
1281         return 0;
1282 }
1283
1284 static void destroy_inodecache(void)
1285 {
1286         /*
1287          * Make sure all delayed rcu free inodes are flushed before we
1288          * destroy cache.
1289          */
1290         rcu_barrier();
1291         kmem_cache_destroy(ext4_inode_cachep);
1292 }
1293
1294 void ext4_clear_inode(struct inode *inode)
1295 {
1296         invalidate_inode_buffers(inode);
1297         clear_inode(inode);
1298         ext4_discard_preallocations(inode, 0);
1299         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1300         dquot_drop(inode);
1301         if (EXT4_I(inode)->jinode) {
1302                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1303                                                EXT4_I(inode)->jinode);
1304                 jbd2_free_inode(EXT4_I(inode)->jinode);
1305                 EXT4_I(inode)->jinode = NULL;
1306         }
1307         fscrypt_put_encryption_info(inode);
1308         fsverity_cleanup_inode(inode);
1309 }
1310
1311 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1312                                         u64 ino, u32 generation)
1313 {
1314         struct inode *inode;
1315
1316         /*
1317          * Currently we don't know the generation for parent directory, so
1318          * a generation of 0 means "accept any"
1319          */
1320         inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE);
1321         if (IS_ERR(inode))
1322                 return ERR_CAST(inode);
1323         if (generation && inode->i_generation != generation) {
1324                 iput(inode);
1325                 return ERR_PTR(-ESTALE);
1326         }
1327
1328         return inode;
1329 }
1330
1331 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1332                                         int fh_len, int fh_type)
1333 {
1334         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1335                                     ext4_nfs_get_inode);
1336 }
1337
1338 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1339                                         int fh_len, int fh_type)
1340 {
1341         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1342                                     ext4_nfs_get_inode);
1343 }
1344
1345 static int ext4_nfs_commit_metadata(struct inode *inode)
1346 {
1347         struct writeback_control wbc = {
1348                 .sync_mode = WB_SYNC_ALL
1349         };
1350
1351         trace_ext4_nfs_commit_metadata(inode);
1352         return ext4_write_inode(inode, &wbc);
1353 }
1354
1355 /*
1356  * Try to release metadata pages (indirect blocks, directories) which are
1357  * mapped via the block device.  Since these pages could have journal heads
1358  * which would prevent try_to_free_buffers() from freeing them, we must use
1359  * jbd2 layer's try_to_free_buffers() function to release them.
1360  */
1361 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1362                                  gfp_t wait)
1363 {
1364         journal_t *journal = EXT4_SB(sb)->s_journal;
1365
1366         WARN_ON(PageChecked(page));
1367         if (!page_has_buffers(page))
1368                 return 0;
1369         if (journal)
1370                 return jbd2_journal_try_to_free_buffers(journal, page);
1371
1372         return try_to_free_buffers(page);
1373 }
1374
1375 #ifdef CONFIG_FS_ENCRYPTION
1376 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1377 {
1378         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1379                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1380 }
1381
1382 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1383                                                         void *fs_data)
1384 {
1385         handle_t *handle = fs_data;
1386         int res, res2, credits, retries = 0;
1387
1388         /*
1389          * Encrypting the root directory is not allowed because e2fsck expects
1390          * lost+found to exist and be unencrypted, and encrypting the root
1391          * directory would imply encrypting the lost+found directory as well as
1392          * the filename "lost+found" itself.
1393          */
1394         if (inode->i_ino == EXT4_ROOT_INO)
1395                 return -EPERM;
1396
1397         if (WARN_ON_ONCE(IS_DAX(inode) && i_size_read(inode)))
1398                 return -EINVAL;
1399
1400         if (ext4_test_inode_flag(inode, EXT4_INODE_DAX))
1401                 return -EOPNOTSUPP;
1402
1403         res = ext4_convert_inline_data(inode);
1404         if (res)
1405                 return res;
1406
1407         /*
1408          * If a journal handle was specified, then the encryption context is
1409          * being set on a new inode via inheritance and is part of a larger
1410          * transaction to create the inode.  Otherwise the encryption context is
1411          * being set on an existing inode in its own transaction.  Only in the
1412          * latter case should the "retry on ENOSPC" logic be used.
1413          */
1414
1415         if (handle) {
1416                 res = ext4_xattr_set_handle(handle, inode,
1417                                             EXT4_XATTR_INDEX_ENCRYPTION,
1418                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1419                                             ctx, len, 0);
1420                 if (!res) {
1421                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1422                         ext4_clear_inode_state(inode,
1423                                         EXT4_STATE_MAY_INLINE_DATA);
1424                         /*
1425                          * Update inode->i_flags - S_ENCRYPTED will be enabled,
1426                          * S_DAX may be disabled
1427                          */
1428                         ext4_set_inode_flags(inode, false);
1429                 }
1430                 return res;
1431         }
1432
1433         res = dquot_initialize(inode);
1434         if (res)
1435                 return res;
1436 retry:
1437         res = ext4_xattr_set_credits(inode, len, false /* is_create */,
1438                                      &credits);
1439         if (res)
1440                 return res;
1441
1442         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
1443         if (IS_ERR(handle))
1444                 return PTR_ERR(handle);
1445
1446         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1447                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1448                                     ctx, len, 0);
1449         if (!res) {
1450                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1451                 /*
1452                  * Update inode->i_flags - S_ENCRYPTED will be enabled,
1453                  * S_DAX may be disabled
1454                  */
1455                 ext4_set_inode_flags(inode, false);
1456                 res = ext4_mark_inode_dirty(handle, inode);
1457                 if (res)
1458                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1459         }
1460         res2 = ext4_journal_stop(handle);
1461
1462         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1463                 goto retry;
1464         if (!res)
1465                 res = res2;
1466         return res;
1467 }
1468
1469 static const union fscrypt_context *
1470 ext4_get_dummy_context(struct super_block *sb)
1471 {
1472         return EXT4_SB(sb)->s_dummy_enc_ctx.ctx;
1473 }
1474
1475 static bool ext4_has_stable_inodes(struct super_block *sb)
1476 {
1477         return ext4_has_feature_stable_inodes(sb);
1478 }
1479
1480 static void ext4_get_ino_and_lblk_bits(struct super_block *sb,
1481                                        int *ino_bits_ret, int *lblk_bits_ret)
1482 {
1483         *ino_bits_ret = 8 * sizeof(EXT4_SB(sb)->s_es->s_inodes_count);
1484         *lblk_bits_ret = 8 * sizeof(ext4_lblk_t);
1485 }
1486
1487 static const struct fscrypt_operations ext4_cryptops = {
1488         .key_prefix             = "ext4:",
1489         .get_context            = ext4_get_context,
1490         .set_context            = ext4_set_context,
1491         .get_dummy_context      = ext4_get_dummy_context,
1492         .empty_dir              = ext4_empty_dir,
1493         .max_namelen            = EXT4_NAME_LEN,
1494         .has_stable_inodes      = ext4_has_stable_inodes,
1495         .get_ino_and_lblk_bits  = ext4_get_ino_and_lblk_bits,
1496 };
1497 #endif
1498
1499 #ifdef CONFIG_QUOTA
1500 static const char * const quotatypes[] = INITQFNAMES;
1501 #define QTYPE2NAME(t) (quotatypes[t])
1502
1503 static int ext4_write_dquot(struct dquot *dquot);
1504 static int ext4_acquire_dquot(struct dquot *dquot);
1505 static int ext4_release_dquot(struct dquot *dquot);
1506 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1507 static int ext4_write_info(struct super_block *sb, int type);
1508 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1509                          const struct path *path);
1510 static int ext4_quota_on_mount(struct super_block *sb, int type);
1511 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1512                                size_t len, loff_t off);
1513 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1514                                 const char *data, size_t len, loff_t off);
1515 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1516                              unsigned int flags);
1517 static int ext4_enable_quotas(struct super_block *sb);
1518
1519 static struct dquot **ext4_get_dquots(struct inode *inode)
1520 {
1521         return EXT4_I(inode)->i_dquot;
1522 }
1523
1524 static const struct dquot_operations ext4_quota_operations = {
1525         .get_reserved_space     = ext4_get_reserved_space,
1526         .write_dquot            = ext4_write_dquot,
1527         .acquire_dquot          = ext4_acquire_dquot,
1528         .release_dquot          = ext4_release_dquot,
1529         .mark_dirty             = ext4_mark_dquot_dirty,
1530         .write_info             = ext4_write_info,
1531         .alloc_dquot            = dquot_alloc,
1532         .destroy_dquot          = dquot_destroy,
1533         .get_projid             = ext4_get_projid,
1534         .get_inode_usage        = ext4_get_inode_usage,
1535         .get_next_id            = dquot_get_next_id,
1536 };
1537
1538 static const struct quotactl_ops ext4_qctl_operations = {
1539         .quota_on       = ext4_quota_on,
1540         .quota_off      = ext4_quota_off,
1541         .quota_sync     = dquot_quota_sync,
1542         .get_state      = dquot_get_state,
1543         .set_info       = dquot_set_dqinfo,
1544         .get_dqblk      = dquot_get_dqblk,
1545         .set_dqblk      = dquot_set_dqblk,
1546         .get_nextdqblk  = dquot_get_next_dqblk,
1547 };
1548 #endif
1549
1550 static const struct super_operations ext4_sops = {
1551         .alloc_inode    = ext4_alloc_inode,
1552         .free_inode     = ext4_free_in_core_inode,
1553         .destroy_inode  = ext4_destroy_inode,
1554         .write_inode    = ext4_write_inode,
1555         .dirty_inode    = ext4_dirty_inode,
1556         .drop_inode     = ext4_drop_inode,
1557         .evict_inode    = ext4_evict_inode,
1558         .put_super      = ext4_put_super,
1559         .sync_fs        = ext4_sync_fs,
1560         .freeze_fs      = ext4_freeze,
1561         .unfreeze_fs    = ext4_unfreeze,
1562         .statfs         = ext4_statfs,
1563         .remount_fs     = ext4_remount,
1564         .show_options   = ext4_show_options,
1565 #ifdef CONFIG_QUOTA
1566         .quota_read     = ext4_quota_read,
1567         .quota_write    = ext4_quota_write,
1568         .get_dquots     = ext4_get_dquots,
1569 #endif
1570         .bdev_try_to_free_page = bdev_try_to_free_page,
1571 };
1572
1573 static const struct export_operations ext4_export_ops = {
1574         .fh_to_dentry = ext4_fh_to_dentry,
1575         .fh_to_parent = ext4_fh_to_parent,
1576         .get_parent = ext4_get_parent,
1577         .commit_metadata = ext4_nfs_commit_metadata,
1578 };
1579
1580 enum {
1581         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1582         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1583         Opt_nouid32, Opt_debug, Opt_removed,
1584         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1585         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1586         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1587         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1588         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1589         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1590         Opt_inlinecrypt,
1591         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1592         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1593         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1594         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version,
1595         Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never,
1596         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
1597         Opt_nowarn_on_error, Opt_mblk_io_submit,
1598         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1599         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1600         Opt_inode_readahead_blks, Opt_journal_ioprio,
1601         Opt_dioread_nolock, Opt_dioread_lock,
1602         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1603         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1604         Opt_prefetch_block_bitmaps,
1605 };
1606
1607 static const match_table_t tokens = {
1608         {Opt_bsd_df, "bsddf"},
1609         {Opt_minix_df, "minixdf"},
1610         {Opt_grpid, "grpid"},
1611         {Opt_grpid, "bsdgroups"},
1612         {Opt_nogrpid, "nogrpid"},
1613         {Opt_nogrpid, "sysvgroups"},
1614         {Opt_resgid, "resgid=%u"},
1615         {Opt_resuid, "resuid=%u"},
1616         {Opt_sb, "sb=%u"},
1617         {Opt_err_cont, "errors=continue"},
1618         {Opt_err_panic, "errors=panic"},
1619         {Opt_err_ro, "errors=remount-ro"},
1620         {Opt_nouid32, "nouid32"},
1621         {Opt_debug, "debug"},
1622         {Opt_removed, "oldalloc"},
1623         {Opt_removed, "orlov"},
1624         {Opt_user_xattr, "user_xattr"},
1625         {Opt_nouser_xattr, "nouser_xattr"},
1626         {Opt_acl, "acl"},
1627         {Opt_noacl, "noacl"},
1628         {Opt_noload, "norecovery"},
1629         {Opt_noload, "noload"},
1630         {Opt_removed, "nobh"},
1631         {Opt_removed, "bh"},
1632         {Opt_commit, "commit=%u"},
1633         {Opt_min_batch_time, "min_batch_time=%u"},
1634         {Opt_max_batch_time, "max_batch_time=%u"},
1635         {Opt_journal_dev, "journal_dev=%u"},
1636         {Opt_journal_path, "journal_path=%s"},
1637         {Opt_journal_checksum, "journal_checksum"},
1638         {Opt_nojournal_checksum, "nojournal_checksum"},
1639         {Opt_journal_async_commit, "journal_async_commit"},
1640         {Opt_abort, "abort"},
1641         {Opt_data_journal, "data=journal"},
1642         {Opt_data_ordered, "data=ordered"},
1643         {Opt_data_writeback, "data=writeback"},
1644         {Opt_data_err_abort, "data_err=abort"},
1645         {Opt_data_err_ignore, "data_err=ignore"},
1646         {Opt_offusrjquota, "usrjquota="},
1647         {Opt_usrjquota, "usrjquota=%s"},
1648         {Opt_offgrpjquota, "grpjquota="},
1649         {Opt_grpjquota, "grpjquota=%s"},
1650         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1651         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1652         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1653         {Opt_grpquota, "grpquota"},
1654         {Opt_noquota, "noquota"},
1655         {Opt_quota, "quota"},
1656         {Opt_usrquota, "usrquota"},
1657         {Opt_prjquota, "prjquota"},
1658         {Opt_barrier, "barrier=%u"},
1659         {Opt_barrier, "barrier"},
1660         {Opt_nobarrier, "nobarrier"},
1661         {Opt_i_version, "i_version"},
1662         {Opt_dax, "dax"},
1663         {Opt_dax_always, "dax=always"},
1664         {Opt_dax_inode, "dax=inode"},
1665         {Opt_dax_never, "dax=never"},
1666         {Opt_stripe, "stripe=%u"},
1667         {Opt_delalloc, "delalloc"},
1668         {Opt_warn_on_error, "warn_on_error"},
1669         {Opt_nowarn_on_error, "nowarn_on_error"},
1670         {Opt_lazytime, "lazytime"},
1671         {Opt_nolazytime, "nolazytime"},
1672         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1673         {Opt_nodelalloc, "nodelalloc"},
1674         {Opt_removed, "mblk_io_submit"},
1675         {Opt_removed, "nomblk_io_submit"},
1676         {Opt_block_validity, "block_validity"},
1677         {Opt_noblock_validity, "noblock_validity"},
1678         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1679         {Opt_journal_ioprio, "journal_ioprio=%u"},
1680         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1681         {Opt_auto_da_alloc, "auto_da_alloc"},
1682         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1683         {Opt_dioread_nolock, "dioread_nolock"},
1684         {Opt_dioread_lock, "nodioread_nolock"},
1685         {Opt_dioread_lock, "dioread_lock"},
1686         {Opt_discard, "discard"},
1687         {Opt_nodiscard, "nodiscard"},
1688         {Opt_init_itable, "init_itable=%u"},
1689         {Opt_init_itable, "init_itable"},
1690         {Opt_noinit_itable, "noinit_itable"},
1691         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1692         {Opt_test_dummy_encryption, "test_dummy_encryption=%s"},
1693         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1694         {Opt_inlinecrypt, "inlinecrypt"},
1695         {Opt_nombcache, "nombcache"},
1696         {Opt_nombcache, "no_mbcache"},  /* for backward compatibility */
1697         {Opt_prefetch_block_bitmaps, "prefetch_block_bitmaps"},
1698         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1699         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1700         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1701         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1702         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1703         {Opt_err, NULL},
1704 };
1705
1706 static ext4_fsblk_t get_sb_block(void **data)
1707 {
1708         ext4_fsblk_t    sb_block;
1709         char            *options = (char *) *data;
1710
1711         if (!options || strncmp(options, "sb=", 3) != 0)
1712                 return 1;       /* Default location */
1713
1714         options += 3;
1715         /* TODO: use simple_strtoll with >32bit ext4 */
1716         sb_block = simple_strtoul(options, &options, 0);
1717         if (*options && *options != ',') {
1718                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1719                        (char *) *data);
1720                 return 1;
1721         }
1722         if (*options == ',')
1723                 options++;
1724         *data = (void *) options;
1725
1726         return sb_block;
1727 }
1728
1729 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1730 static const char deprecated_msg[] =
1731         "Mount option \"%s\" will be removed by %s\n"
1732         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1733
1734 #ifdef CONFIG_QUOTA
1735 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1736 {
1737         struct ext4_sb_info *sbi = EXT4_SB(sb);
1738         char *qname, *old_qname = get_qf_name(sb, sbi, qtype);
1739         int ret = -1;
1740
1741         if (sb_any_quota_loaded(sb) && !old_qname) {
1742                 ext4_msg(sb, KERN_ERR,
1743                         "Cannot change journaled "
1744                         "quota options when quota turned on");
1745                 return -1;
1746         }
1747         if (ext4_has_feature_quota(sb)) {
1748                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1749                          "ignored when QUOTA feature is enabled");
1750                 return 1;
1751         }
1752         qname = match_strdup(args);
1753         if (!qname) {
1754                 ext4_msg(sb, KERN_ERR,
1755                         "Not enough memory for storing quotafile name");
1756                 return -1;
1757         }
1758         if (old_qname) {
1759                 if (strcmp(old_qname, qname) == 0)
1760                         ret = 1;
1761                 else
1762                         ext4_msg(sb, KERN_ERR,
1763                                  "%s quota file already specified",
1764                                  QTYPE2NAME(qtype));
1765                 goto errout;
1766         }
1767         if (strchr(qname, '/')) {
1768                 ext4_msg(sb, KERN_ERR,
1769                         "quotafile must be on filesystem root");
1770                 goto errout;
1771         }
1772         rcu_assign_pointer(sbi->s_qf_names[qtype], qname);
1773         set_opt(sb, QUOTA);
1774         return 1;
1775 errout:
1776         kfree(qname);
1777         return ret;
1778 }
1779
1780 static int clear_qf_name(struct super_block *sb, int qtype)
1781 {
1782
1783         struct ext4_sb_info *sbi = EXT4_SB(sb);
1784         char *old_qname = get_qf_name(sb, sbi, qtype);
1785
1786         if (sb_any_quota_loaded(sb) && old_qname) {
1787                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1788                         " when quota turned on");
1789                 return -1;
1790         }
1791         rcu_assign_pointer(sbi->s_qf_names[qtype], NULL);
1792         synchronize_rcu();
1793         kfree(old_qname);
1794         return 1;
1795 }
1796 #endif
1797
1798 #define MOPT_SET        0x0001
1799 #define MOPT_CLEAR      0x0002
1800 #define MOPT_NOSUPPORT  0x0004
1801 #define MOPT_EXPLICIT   0x0008
1802 #define MOPT_CLEAR_ERR  0x0010
1803 #define MOPT_GTE0       0x0020
1804 #ifdef CONFIG_QUOTA
1805 #define MOPT_Q          0
1806 #define MOPT_QFMT       0x0040
1807 #else
1808 #define MOPT_Q          MOPT_NOSUPPORT
1809 #define MOPT_QFMT       MOPT_NOSUPPORT
1810 #endif
1811 #define MOPT_DATAJ      0x0080
1812 #define MOPT_NO_EXT2    0x0100
1813 #define MOPT_NO_EXT3    0x0200
1814 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1815 #define MOPT_STRING     0x0400
1816 #define MOPT_SKIP       0x0800
1817
1818 static const struct mount_opts {
1819         int     token;
1820         int     mount_opt;
1821         int     flags;
1822 } ext4_mount_opts[] = {
1823         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1824         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1825         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1826         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1827         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1828         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1829         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1830          MOPT_EXT4_ONLY | MOPT_SET},
1831         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1832          MOPT_EXT4_ONLY | MOPT_CLEAR},
1833         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1834         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1835         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1836          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1837         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1838          MOPT_EXT4_ONLY | MOPT_CLEAR},
1839         {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET},
1840         {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR},
1841         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1842          MOPT_EXT4_ONLY | MOPT_CLEAR},
1843         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1844          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1845         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1846                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1847          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1848         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1849         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1850         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1851         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1852         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1853          MOPT_NO_EXT2},
1854         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1855          MOPT_NO_EXT2},
1856         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1857         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1858         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1859         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1860         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1861         {Opt_commit, 0, MOPT_GTE0},
1862         {Opt_max_batch_time, 0, MOPT_GTE0},
1863         {Opt_min_batch_time, 0, MOPT_GTE0},
1864         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1865         {Opt_init_itable, 0, MOPT_GTE0},
1866         {Opt_dax, EXT4_MOUNT_DAX_ALWAYS, MOPT_SET | MOPT_SKIP},
1867         {Opt_dax_always, EXT4_MOUNT_DAX_ALWAYS,
1868                 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP},
1869         {Opt_dax_inode, EXT4_MOUNT2_DAX_INODE,
1870                 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP},
1871         {Opt_dax_never, EXT4_MOUNT2_DAX_NEVER,
1872                 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP},
1873         {Opt_stripe, 0, MOPT_GTE0},
1874         {Opt_resuid, 0, MOPT_GTE0},
1875         {Opt_resgid, 0, MOPT_GTE0},
1876         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1877         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1878         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1879         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1880         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1881         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1882          MOPT_NO_EXT2 | MOPT_DATAJ},
1883         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1884         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1885 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1886         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1887         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1888 #else
1889         {Opt_acl, 0, MOPT_NOSUPPORT},
1890         {Opt_noacl, 0, MOPT_NOSUPPORT},
1891 #endif
1892         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1893         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1894         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1895         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1896         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1897                                                         MOPT_SET | MOPT_Q},
1898         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1899                                                         MOPT_SET | MOPT_Q},
1900         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1901                                                         MOPT_SET | MOPT_Q},
1902         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1903                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1904                                                         MOPT_CLEAR | MOPT_Q},
1905         {Opt_usrjquota, 0, MOPT_Q},
1906         {Opt_grpjquota, 0, MOPT_Q},
1907         {Opt_offusrjquota, 0, MOPT_Q},
1908         {Opt_offgrpjquota, 0, MOPT_Q},
1909         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1910         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1911         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1912         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1913         {Opt_test_dummy_encryption, 0, MOPT_STRING},
1914         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1915         {Opt_prefetch_block_bitmaps, EXT4_MOUNT_PREFETCH_BLOCK_BITMAPS,
1916          MOPT_SET},
1917         {Opt_err, 0, 0}
1918 };
1919
1920 #ifdef CONFIG_UNICODE
1921 static const struct ext4_sb_encodings {
1922         __u16 magic;
1923         char *name;
1924         char *version;
1925 } ext4_sb_encoding_map[] = {
1926         {EXT4_ENC_UTF8_12_1, "utf8", "12.1.0"},
1927 };
1928
1929 static int ext4_sb_read_encoding(const struct ext4_super_block *es,
1930                                  const struct ext4_sb_encodings **encoding,
1931                                  __u16 *flags)
1932 {
1933         __u16 magic = le16_to_cpu(es->s_encoding);
1934         int i;
1935
1936         for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++)
1937                 if (magic == ext4_sb_encoding_map[i].magic)
1938                         break;
1939
1940         if (i >= ARRAY_SIZE(ext4_sb_encoding_map))
1941                 return -EINVAL;
1942
1943         *encoding = &ext4_sb_encoding_map[i];
1944         *flags = le16_to_cpu(es->s_encoding_flags);
1945
1946         return 0;
1947 }
1948 #endif
1949
1950 static int ext4_set_test_dummy_encryption(struct super_block *sb,
1951                                           const char *opt,
1952                                           const substring_t *arg,
1953                                           bool is_remount)
1954 {
1955 #ifdef CONFIG_FS_ENCRYPTION
1956         struct ext4_sb_info *sbi = EXT4_SB(sb);
1957         int err;
1958
1959         /*
1960          * This mount option is just for testing, and it's not worthwhile to
1961          * implement the extra complexity (e.g. RCU protection) that would be
1962          * needed to allow it to be set or changed during remount.  We do allow
1963          * it to be specified during remount, but only if there is no change.
1964          */
1965         if (is_remount && !sbi->s_dummy_enc_ctx.ctx) {
1966                 ext4_msg(sb, KERN_WARNING,
1967                          "Can't set test_dummy_encryption on remount");
1968                 return -1;
1969         }
1970         err = fscrypt_set_test_dummy_encryption(sb, arg, &sbi->s_dummy_enc_ctx);
1971         if (err) {
1972                 if (err == -EEXIST)
1973                         ext4_msg(sb, KERN_WARNING,
1974                                  "Can't change test_dummy_encryption on remount");
1975                 else if (err == -EINVAL)
1976                         ext4_msg(sb, KERN_WARNING,
1977                                  "Value of option \"%s\" is unrecognized", opt);
1978                 else
1979                         ext4_msg(sb, KERN_WARNING,
1980                                  "Error processing option \"%s\" [%d]",
1981                                  opt, err);
1982                 return -1;
1983         }
1984         ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled");
1985 #else
1986         ext4_msg(sb, KERN_WARNING,
1987                  "Test dummy encryption mount option ignored");
1988 #endif
1989         return 1;
1990 }
1991
1992 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1993                             substring_t *args, unsigned long *journal_devnum,
1994                             unsigned int *journal_ioprio, int is_remount)
1995 {
1996         struct ext4_sb_info *sbi = EXT4_SB(sb);
1997         const struct mount_opts *m;
1998         kuid_t uid;
1999         kgid_t gid;
2000         int arg = 0;
2001
2002 #ifdef CONFIG_QUOTA
2003         if (token == Opt_usrjquota)
2004                 return set_qf_name(sb, USRQUOTA, &args[0]);
2005         else if (token == Opt_grpjquota)
2006                 return set_qf_name(sb, GRPQUOTA, &args[0]);
2007         else if (token == Opt_offusrjquota)
2008                 return clear_qf_name(sb, USRQUOTA);
2009         else if (token == Opt_offgrpjquota)
2010                 return clear_qf_name(sb, GRPQUOTA);
2011 #endif
2012         switch (token) {
2013         case Opt_noacl:
2014         case Opt_nouser_xattr:
2015                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
2016                 break;
2017         case Opt_sb:
2018                 return 1;       /* handled by get_sb_block() */
2019         case Opt_removed:
2020                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
2021                 return 1;
2022         case Opt_abort:
2023                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
2024                 return 1;
2025         case Opt_i_version:
2026                 sb->s_flags |= SB_I_VERSION;
2027                 return 1;
2028         case Opt_lazytime:
2029                 sb->s_flags |= SB_LAZYTIME;
2030                 return 1;
2031         case Opt_nolazytime:
2032                 sb->s_flags &= ~SB_LAZYTIME;
2033                 return 1;
2034         case Opt_inlinecrypt:
2035 #ifdef CONFIG_FS_ENCRYPTION_INLINE_CRYPT
2036                 sb->s_flags |= SB_INLINECRYPT;
2037 #else
2038                 ext4_msg(sb, KERN_ERR, "inline encryption not supported");
2039 #endif
2040                 return 1;
2041         }
2042
2043         for (m = ext4_mount_opts; m->token != Opt_err; m++)
2044                 if (token == m->token)
2045                         break;
2046
2047         if (m->token == Opt_err) {
2048                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
2049                          "or missing value", opt);
2050                 return -1;
2051         }
2052
2053         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
2054                 ext4_msg(sb, KERN_ERR,
2055                          "Mount option \"%s\" incompatible with ext2", opt);
2056                 return -1;
2057         }
2058         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
2059                 ext4_msg(sb, KERN_ERR,
2060                          "Mount option \"%s\" incompatible with ext3", opt);
2061                 return -1;
2062         }
2063
2064         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
2065                 return -1;
2066         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
2067                 return -1;
2068         if (m->flags & MOPT_EXPLICIT) {
2069                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
2070                         set_opt2(sb, EXPLICIT_DELALLOC);
2071                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
2072                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
2073                 } else
2074                         return -1;
2075         }
2076         if (m->flags & MOPT_CLEAR_ERR)
2077                 clear_opt(sb, ERRORS_MASK);
2078         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
2079                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
2080                          "options when quota turned on");
2081                 return -1;
2082         }
2083
2084         if (m->flags & MOPT_NOSUPPORT) {
2085                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
2086         } else if (token == Opt_commit) {
2087                 if (arg == 0)
2088                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
2089                 else if (arg > INT_MAX / HZ) {
2090                         ext4_msg(sb, KERN_ERR,
2091                                  "Invalid commit interval %d, "
2092                                  "must be smaller than %d",
2093                                  arg, INT_MAX / HZ);
2094                         return -1;
2095                 }
2096                 sbi->s_commit_interval = HZ * arg;
2097         } else if (token == Opt_debug_want_extra_isize) {
2098                 if ((arg & 1) ||
2099                     (arg < 4) ||
2100                     (arg > (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE))) {
2101                         ext4_msg(sb, KERN_ERR,
2102                                  "Invalid want_extra_isize %d", arg);
2103                         return -1;
2104                 }
2105                 sbi->s_want_extra_isize = arg;
2106         } else if (token == Opt_max_batch_time) {
2107                 sbi->s_max_batch_time = arg;
2108         } else if (token == Opt_min_batch_time) {
2109                 sbi->s_min_batch_time = arg;
2110         } else if (token == Opt_inode_readahead_blks) {
2111                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
2112                         ext4_msg(sb, KERN_ERR,
2113                                  "EXT4-fs: inode_readahead_blks must be "
2114                                  "0 or a power of 2 smaller than 2^31");
2115                         return -1;
2116                 }
2117                 sbi->s_inode_readahead_blks = arg;
2118         } else if (token == Opt_init_itable) {
2119                 set_opt(sb, INIT_INODE_TABLE);
2120                 if (!args->from)
2121                         arg = EXT4_DEF_LI_WAIT_MULT;
2122                 sbi->s_li_wait_mult = arg;
2123         } else if (token == Opt_max_dir_size_kb) {
2124                 sbi->s_max_dir_size_kb = arg;
2125         } else if (token == Opt_stripe) {
2126                 sbi->s_stripe = arg;
2127         } else if (token == Opt_resuid) {
2128                 uid = make_kuid(current_user_ns(), arg);
2129                 if (!uid_valid(uid)) {
2130                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
2131                         return -1;
2132                 }
2133                 sbi->s_resuid = uid;
2134         } else if (token == Opt_resgid) {
2135                 gid = make_kgid(current_user_ns(), arg);
2136                 if (!gid_valid(gid)) {
2137                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
2138                         return -1;
2139                 }
2140                 sbi->s_resgid = gid;
2141         } else if (token == Opt_journal_dev) {
2142                 if (is_remount) {
2143                         ext4_msg(sb, KERN_ERR,
2144                                  "Cannot specify journal on remount");
2145                         return -1;
2146                 }
2147                 *journal_devnum = arg;
2148         } else if (token == Opt_journal_path) {
2149                 char *journal_path;
2150                 struct inode *journal_inode;
2151                 struct path path;
2152                 int error;
2153
2154                 if (is_remount) {
2155                         ext4_msg(sb, KERN_ERR,
2156                                  "Cannot specify journal on remount");
2157                         return -1;
2158                 }
2159                 journal_path = match_strdup(&args[0]);
2160                 if (!journal_path) {
2161                         ext4_msg(sb, KERN_ERR, "error: could not dup "
2162                                 "journal device string");
2163                         return -1;
2164                 }
2165
2166                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
2167                 if (error) {
2168                         ext4_msg(sb, KERN_ERR, "error: could not find "
2169                                 "journal device path: error %d", error);
2170                         kfree(journal_path);
2171                         return -1;
2172                 }
2173
2174                 journal_inode = d_inode(path.dentry);
2175                 if (!S_ISBLK(journal_inode->i_mode)) {
2176                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
2177                                 "is not a block device", journal_path);
2178                         path_put(&path);
2179                         kfree(journal_path);
2180                         return -1;
2181                 }
2182
2183                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
2184                 path_put(&path);
2185                 kfree(journal_path);
2186         } else if (token == Opt_journal_ioprio) {
2187                 if (arg > 7) {
2188                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
2189                                  " (must be 0-7)");
2190                         return -1;
2191                 }
2192                 *journal_ioprio =
2193                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
2194         } else if (token == Opt_test_dummy_encryption) {
2195                 return ext4_set_test_dummy_encryption(sb, opt, &args[0],
2196                                                       is_remount);
2197         } else if (m->flags & MOPT_DATAJ) {
2198                 if (is_remount) {
2199                         if (!sbi->s_journal)
2200                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
2201                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
2202                                 ext4_msg(sb, KERN_ERR,
2203                                          "Cannot change data mode on remount");
2204                                 return -1;
2205                         }
2206                 } else {
2207                         clear_opt(sb, DATA_FLAGS);
2208                         sbi->s_mount_opt |= m->mount_opt;
2209                 }
2210 #ifdef CONFIG_QUOTA
2211         } else if (m->flags & MOPT_QFMT) {
2212                 if (sb_any_quota_loaded(sb) &&
2213                     sbi->s_jquota_fmt != m->mount_opt) {
2214                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
2215                                  "quota options when quota turned on");
2216                         return -1;
2217                 }
2218                 if (ext4_has_feature_quota(sb)) {
2219                         ext4_msg(sb, KERN_INFO,
2220                                  "Quota format mount options ignored "
2221                                  "when QUOTA feature is enabled");
2222                         return 1;
2223                 }
2224                 sbi->s_jquota_fmt = m->mount_opt;
2225 #endif
2226         } else if (token == Opt_dax || token == Opt_dax_always ||
2227                    token == Opt_dax_inode || token == Opt_dax_never) {
2228 #ifdef CONFIG_FS_DAX
2229                 switch (token) {
2230                 case Opt_dax:
2231                 case Opt_dax_always:
2232                         if (is_remount &&
2233                             (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2234                              (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) {
2235                         fail_dax_change_remount:
2236                                 ext4_msg(sb, KERN_ERR, "can't change "
2237                                          "dax mount option while remounting");
2238                                 return -1;
2239                         }
2240                         if (is_remount &&
2241                             (test_opt(sb, DATA_FLAGS) ==
2242                              EXT4_MOUNT_JOURNAL_DATA)) {
2243                                     ext4_msg(sb, KERN_ERR, "can't mount with "
2244                                              "both data=journal and dax");
2245                                     return -1;
2246                         }
2247                         ext4_msg(sb, KERN_WARNING,
2248                                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
2249                         sbi->s_mount_opt |= EXT4_MOUNT_DAX_ALWAYS;
2250                         sbi->s_mount_opt2 &= ~EXT4_MOUNT2_DAX_NEVER;
2251                         break;
2252                 case Opt_dax_never:
2253                         if (is_remount &&
2254                             (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2255                              (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS)))
2256                                 goto fail_dax_change_remount;
2257                         sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_NEVER;
2258                         sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS;
2259                         break;
2260                 case Opt_dax_inode:
2261                         if (is_remount &&
2262                             ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2263                              (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2264                              !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE)))
2265                                 goto fail_dax_change_remount;
2266                         sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS;
2267                         sbi->s_mount_opt2 &= ~EXT4_MOUNT2_DAX_NEVER;
2268                         /* Strictly for printing options */
2269                         sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_INODE;
2270                         break;
2271                 }
2272 #else
2273                 ext4_msg(sb, KERN_INFO, "dax option not supported");
2274                 sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_NEVER;
2275                 sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS;
2276                 return -1;
2277 #endif
2278         } else if (token == Opt_data_err_abort) {
2279                 sbi->s_mount_opt |= m->mount_opt;
2280         } else if (token == Opt_data_err_ignore) {
2281                 sbi->s_mount_opt &= ~m->mount_opt;
2282         } else {
2283                 if (!args->from)
2284                         arg = 1;
2285                 if (m->flags & MOPT_CLEAR)
2286                         arg = !arg;
2287                 else if (unlikely(!(m->flags & MOPT_SET))) {
2288                         ext4_msg(sb, KERN_WARNING,
2289                                  "buggy handling of option %s", opt);
2290                         WARN_ON(1);
2291                         return -1;
2292                 }
2293                 if (arg != 0)
2294                         sbi->s_mount_opt |= m->mount_opt;
2295                 else
2296                         sbi->s_mount_opt &= ~m->mount_opt;
2297         }
2298         return 1;
2299 }
2300
2301 static int parse_options(char *options, struct super_block *sb,
2302                          unsigned long *journal_devnum,
2303                          unsigned int *journal_ioprio,
2304                          int is_remount)
2305 {
2306         struct ext4_sb_info __maybe_unused *sbi = EXT4_SB(sb);
2307         char *p, __maybe_unused *usr_qf_name, __maybe_unused *grp_qf_name;
2308         substring_t args[MAX_OPT_ARGS];
2309         int token;
2310
2311         if (!options)
2312                 return 1;
2313
2314         while ((p = strsep(&options, ",")) != NULL) {
2315                 if (!*p)
2316                         continue;
2317                 /*
2318                  * Initialize args struct so we know whether arg was
2319                  * found; some options take optional arguments.
2320                  */
2321                 args[0].to = args[0].from = NULL;
2322                 token = match_token(p, tokens, args);
2323                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
2324                                      journal_ioprio, is_remount) < 0)
2325                         return 0;
2326         }
2327 #ifdef CONFIG_QUOTA
2328         /*
2329          * We do the test below only for project quotas. 'usrquota' and
2330          * 'grpquota' mount options are allowed even without quota feature
2331          * to support legacy quotas in quota files.
2332          */
2333         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
2334                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
2335                          "Cannot enable project quota enforcement.");
2336                 return 0;
2337         }
2338         usr_qf_name = get_qf_name(sb, sbi, USRQUOTA);
2339         grp_qf_name = get_qf_name(sb, sbi, GRPQUOTA);
2340         if (usr_qf_name || grp_qf_name) {
2341                 if (test_opt(sb, USRQUOTA) && usr_qf_name)
2342                         clear_opt(sb, USRQUOTA);
2343
2344                 if (test_opt(sb, GRPQUOTA) && grp_qf_name)
2345                         clear_opt(sb, GRPQUOTA);
2346
2347                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
2348                         ext4_msg(sb, KERN_ERR, "old and new quota "
2349                                         "format mixing");
2350                         return 0;
2351                 }
2352
2353                 if (!sbi->s_jquota_fmt) {
2354                         ext4_msg(sb, KERN_ERR, "journaled quota format "
2355                                         "not specified");
2356                         return 0;
2357                 }
2358         }
2359 #endif
2360         if (test_opt(sb, DIOREAD_NOLOCK)) {
2361                 int blocksize =
2362                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
2363                 if (blocksize < PAGE_SIZE)
2364                         ext4_msg(sb, KERN_WARNING, "Warning: mounting with an "
2365                                  "experimental mount option 'dioread_nolock' "
2366                                  "for blocksize < PAGE_SIZE");
2367         }
2368         return 1;
2369 }
2370
2371 static inline void ext4_show_quota_options(struct seq_file *seq,
2372                                            struct super_block *sb)
2373 {
2374 #if defined(CONFIG_QUOTA)
2375         struct ext4_sb_info *sbi = EXT4_SB(sb);
2376         char *usr_qf_name, *grp_qf_name;
2377
2378         if (sbi->s_jquota_fmt) {
2379                 char *fmtname = "";
2380
2381                 switch (sbi->s_jquota_fmt) {
2382                 case QFMT_VFS_OLD:
2383                         fmtname = "vfsold";
2384                         break;
2385                 case QFMT_VFS_V0:
2386                         fmtname = "vfsv0";
2387                         break;
2388                 case QFMT_VFS_V1:
2389                         fmtname = "vfsv1";
2390                         break;
2391                 }
2392                 seq_printf(seq, ",jqfmt=%s", fmtname);
2393         }
2394
2395         rcu_read_lock();
2396         usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]);
2397         grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]);
2398         if (usr_qf_name)
2399                 seq_show_option(seq, "usrjquota", usr_qf_name);
2400         if (grp_qf_name)
2401                 seq_show_option(seq, "grpjquota", grp_qf_name);
2402         rcu_read_unlock();
2403 #endif
2404 }
2405
2406 static const char *token2str(int token)
2407 {
2408         const struct match_token *t;
2409
2410         for (t = tokens; t->token != Opt_err; t++)
2411                 if (t->token == token && !strchr(t->pattern, '='))
2412                         break;
2413         return t->pattern;
2414 }
2415
2416 /*
2417  * Show an option if
2418  *  - it's set to a non-default value OR
2419  *  - if the per-sb default is different from the global default
2420  */
2421 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2422                               int nodefs)
2423 {
2424         struct ext4_sb_info *sbi = EXT4_SB(sb);
2425         struct ext4_super_block *es = sbi->s_es;
2426         int def_errors, def_mount_opt = sbi->s_def_mount_opt;
2427         const struct mount_opts *m;
2428         char sep = nodefs ? '\n' : ',';
2429
2430 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2431 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2432
2433         if (sbi->s_sb_block != 1)
2434                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2435
2436         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2437                 int want_set = m->flags & MOPT_SET;
2438                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2439                     (m->flags & MOPT_CLEAR_ERR) || m->flags & MOPT_SKIP)
2440                         continue;
2441                 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2442                         continue; /* skip if same as the default */
2443                 if ((want_set &&
2444                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2445                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2446                         continue; /* select Opt_noFoo vs Opt_Foo */
2447                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2448         }
2449
2450         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2451             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2452                 SEQ_OPTS_PRINT("resuid=%u",
2453                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2454         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2455             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2456                 SEQ_OPTS_PRINT("resgid=%u",
2457                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2458         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2459         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2460                 SEQ_OPTS_PUTS("errors=remount-ro");
2461         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2462                 SEQ_OPTS_PUTS("errors=continue");
2463         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2464                 SEQ_OPTS_PUTS("errors=panic");
2465         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2466                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2467         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2468                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2469         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2470                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2471         if (sb->s_flags & SB_I_VERSION)
2472                 SEQ_OPTS_PUTS("i_version");
2473         if (nodefs || sbi->s_stripe)
2474                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2475         if (nodefs || EXT4_MOUNT_DATA_FLAGS &
2476                         (sbi->s_mount_opt ^ def_mount_opt)) {
2477                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2478                         SEQ_OPTS_PUTS("data=journal");
2479                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2480                         SEQ_OPTS_PUTS("data=ordered");
2481                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2482                         SEQ_OPTS_PUTS("data=writeback");
2483         }
2484         if (nodefs ||
2485             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2486                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2487                                sbi->s_inode_readahead_blks);
2488
2489         if (test_opt(sb, INIT_INODE_TABLE) && (nodefs ||
2490                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2491                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2492         if (nodefs || sbi->s_max_dir_size_kb)
2493                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2494         if (test_opt(sb, DATA_ERR_ABORT))
2495                 SEQ_OPTS_PUTS("data_err=abort");
2496
2497         fscrypt_show_test_dummy_encryption(seq, sep, sb);
2498
2499         if (sb->s_flags & SB_INLINECRYPT)
2500                 SEQ_OPTS_PUTS("inlinecrypt");
2501
2502         if (test_opt(sb, DAX_ALWAYS)) {
2503                 if (IS_EXT2_SB(sb))
2504                         SEQ_OPTS_PUTS("dax");
2505                 else
2506                         SEQ_OPTS_PUTS("dax=always");
2507         } else if (test_opt2(sb, DAX_NEVER)) {
2508                 SEQ_OPTS_PUTS("dax=never");
2509         } else if (test_opt2(sb, DAX_INODE)) {
2510                 SEQ_OPTS_PUTS("dax=inode");
2511         }
2512
2513         ext4_show_quota_options(seq, sb);
2514         return 0;
2515 }
2516
2517 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2518 {
2519         return _ext4_show_options(seq, root->d_sb, 0);
2520 }
2521
2522 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2523 {
2524         struct super_block *sb = seq->private;
2525         int rc;
2526
2527         seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
2528         rc = _ext4_show_options(seq, sb, 1);
2529         seq_puts(seq, "\n");
2530         return rc;
2531 }
2532
2533 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2534                             int read_only)
2535 {
2536         struct ext4_sb_info *sbi = EXT4_SB(sb);
2537         int err = 0;
2538
2539         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2540                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2541                          "forcing read-only mode");
2542                 err = -EROFS;
2543                 goto done;
2544         }
2545         if (read_only)
2546                 goto done;
2547         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2548                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2549                          "running e2fsck is recommended");
2550         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2551                 ext4_msg(sb, KERN_WARNING,
2552                          "warning: mounting fs with errors, "
2553                          "running e2fsck is recommended");
2554         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2555                  le16_to_cpu(es->s_mnt_count) >=
2556                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2557                 ext4_msg(sb, KERN_WARNING,
2558                          "warning: maximal mount count reached, "
2559                          "running e2fsck is recommended");
2560         else if (le32_to_cpu(es->s_checkinterval) &&
2561                  (ext4_get_tstamp(es, s_lastcheck) +
2562                   le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds()))
2563                 ext4_msg(sb, KERN_WARNING,
2564                          "warning: checktime reached, "
2565                          "running e2fsck is recommended");
2566         if (!sbi->s_journal)
2567                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2568         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2569                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2570         le16_add_cpu(&es->s_mnt_count, 1);
2571         ext4_update_tstamp(es, s_mtime);
2572         if (sbi->s_journal)
2573                 ext4_set_feature_journal_needs_recovery(sb);
2574
2575         err = ext4_commit_super(sb, 1);
2576 done:
2577         if (test_opt(sb, DEBUG))
2578                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2579                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2580                         sb->s_blocksize,
2581                         sbi->s_groups_count,
2582                         EXT4_BLOCKS_PER_GROUP(sb),
2583                         EXT4_INODES_PER_GROUP(sb),
2584                         sbi->s_mount_opt, sbi->s_mount_opt2);
2585
2586         cleancache_init_fs(sb);
2587         return err;
2588 }
2589
2590 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2591 {
2592         struct ext4_sb_info *sbi = EXT4_SB(sb);
2593         struct flex_groups **old_groups, **new_groups;
2594         int size, i, j;
2595
2596         if (!sbi->s_log_groups_per_flex)
2597                 return 0;
2598
2599         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2600         if (size <= sbi->s_flex_groups_allocated)
2601                 return 0;
2602
2603         new_groups = kvzalloc(roundup_pow_of_two(size *
2604                               sizeof(*sbi->s_flex_groups)), GFP_KERNEL);
2605         if (!new_groups) {
2606                 ext4_msg(sb, KERN_ERR,
2607                          "not enough memory for %d flex group pointers", size);
2608                 return -ENOMEM;
2609         }
2610         for (i = sbi->s_flex_groups_allocated; i < size; i++) {
2611                 new_groups[i] = kvzalloc(roundup_pow_of_two(
2612                                          sizeof(struct flex_groups)),
2613                                          GFP_KERNEL);
2614                 if (!new_groups[i]) {
2615                         for (j = sbi->s_flex_groups_allocated; j < i; j++)
2616                                 kvfree(new_groups[j]);
2617                         kvfree(new_groups);
2618                         ext4_msg(sb, KERN_ERR,
2619                                  "not enough memory for %d flex groups", size);
2620                         return -ENOMEM;
2621                 }
2622         }
2623         rcu_read_lock();
2624         old_groups = rcu_dereference(sbi->s_flex_groups);
2625         if (old_groups)
2626                 memcpy(new_groups, old_groups,
2627                        (sbi->s_flex_groups_allocated *
2628                         sizeof(struct flex_groups *)));
2629         rcu_read_unlock();
2630         rcu_assign_pointer(sbi->s_flex_groups, new_groups);
2631         sbi->s_flex_groups_allocated = size;
2632         if (old_groups)
2633                 ext4_kvfree_array_rcu(old_groups);
2634         return 0;
2635 }
2636
2637 static int ext4_fill_flex_info(struct super_block *sb)
2638 {
2639         struct ext4_sb_info *sbi = EXT4_SB(sb);
2640         struct ext4_group_desc *gdp = NULL;
2641         struct flex_groups *fg;
2642         ext4_group_t flex_group;
2643         int i, err;
2644
2645         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2646         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2647                 sbi->s_log_groups_per_flex = 0;
2648                 return 1;
2649         }
2650
2651         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2652         if (err)
2653                 goto failed;
2654
2655         for (i = 0; i < sbi->s_groups_count; i++) {
2656                 gdp = ext4_get_group_desc(sb, i, NULL);
2657
2658                 flex_group = ext4_flex_group(sbi, i);
2659                 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
2660                 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes);
2661                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2662                              &fg->free_clusters);
2663                 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs);
2664         }
2665
2666         return 1;
2667 failed:
2668         return 0;
2669 }
2670
2671 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2672                                    struct ext4_group_desc *gdp)
2673 {
2674         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2675         __u16 crc = 0;
2676         __le32 le_group = cpu_to_le32(block_group);
2677         struct ext4_sb_info *sbi = EXT4_SB(sb);
2678
2679         if (ext4_has_metadata_csum(sbi->s_sb)) {
2680                 /* Use new metadata_csum algorithm */
2681                 __u32 csum32;
2682                 __u16 dummy_csum = 0;
2683
2684                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2685                                      sizeof(le_group));
2686                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2687                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2688                                      sizeof(dummy_csum));
2689                 offset += sizeof(dummy_csum);
2690                 if (offset < sbi->s_desc_size)
2691                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2692                                              sbi->s_desc_size - offset);
2693
2694                 crc = csum32 & 0xFFFF;
2695                 goto out;
2696         }
2697
2698         /* old crc16 code */
2699         if (!ext4_has_feature_gdt_csum(sb))
2700                 return 0;
2701
2702         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2703         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2704         crc = crc16(crc, (__u8 *)gdp, offset);
2705         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2706         /* for checksum of struct ext4_group_desc do the rest...*/
2707         if (ext4_has_feature_64bit(sb) &&
2708             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2709                 crc = crc16(crc, (__u8 *)gdp + offset,
2710                             le16_to_cpu(sbi->s_es->s_desc_size) -
2711                                 offset);
2712
2713 out:
2714         return cpu_to_le16(crc);
2715 }
2716
2717 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2718                                 struct ext4_group_desc *gdp)
2719 {
2720         if (ext4_has_group_desc_csum(sb) &&
2721             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2722                 return 0;
2723
2724         return 1;
2725 }
2726
2727 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2728                               struct ext4_group_desc *gdp)
2729 {
2730         if (!ext4_has_group_desc_csum(sb))
2731                 return;
2732         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2733 }
2734
2735 /* Called at mount-time, super-block is locked */
2736 static int ext4_check_descriptors(struct super_block *sb,
2737                                   ext4_fsblk_t sb_block,
2738                                   ext4_group_t *first_not_zeroed)
2739 {
2740         struct ext4_sb_info *sbi = EXT4_SB(sb);
2741         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2742         ext4_fsblk_t last_block;
2743         ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0);
2744         ext4_fsblk_t block_bitmap;
2745         ext4_fsblk_t inode_bitmap;
2746         ext4_fsblk_t inode_table;
2747         int flexbg_flag = 0;
2748         ext4_group_t i, grp = sbi->s_groups_count;
2749
2750         if (ext4_has_feature_flex_bg(sb))
2751                 flexbg_flag = 1;
2752
2753         ext4_debug("Checking group descriptors");
2754
2755         for (i = 0; i < sbi->s_groups_count; i++) {
2756                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2757
2758                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2759                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2760                 else
2761                         last_block = first_block +
2762                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2763
2764                 if ((grp == sbi->s_groups_count) &&
2765                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2766                         grp = i;
2767
2768                 block_bitmap = ext4_block_bitmap(sb, gdp);
2769                 if (block_bitmap == sb_block) {
2770                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2771                                  "Block bitmap for group %u overlaps "
2772                                  "superblock", i);
2773                         if (!sb_rdonly(sb))
2774                                 return 0;
2775                 }
2776                 if (block_bitmap >= sb_block + 1 &&
2777                     block_bitmap <= last_bg_block) {
2778                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2779                                  "Block bitmap for group %u overlaps "
2780                                  "block group descriptors", i);
2781                         if (!sb_rdonly(sb))
2782                                 return 0;
2783                 }
2784                 if (block_bitmap < first_block || block_bitmap > last_block) {
2785                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2786                                "Block bitmap for group %u not in group "
2787                                "(block %llu)!", i, block_bitmap);
2788                         return 0;
2789                 }
2790                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2791                 if (inode_bitmap == sb_block) {
2792                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2793                                  "Inode bitmap for group %u overlaps "
2794                                  "superblock", i);
2795                         if (!sb_rdonly(sb))
2796                                 return 0;
2797                 }
2798                 if (inode_bitmap >= sb_block + 1 &&
2799                     inode_bitmap <= last_bg_block) {
2800                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2801                                  "Inode bitmap for group %u overlaps "
2802                                  "block group descriptors", i);
2803                         if (!sb_rdonly(sb))
2804                                 return 0;
2805                 }
2806                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2807                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2808                                "Inode bitmap for group %u not in group "
2809                                "(block %llu)!", i, inode_bitmap);
2810                         return 0;
2811                 }
2812                 inode_table = ext4_inode_table(sb, gdp);
2813                 if (inode_table == sb_block) {
2814                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2815                                  "Inode table for group %u overlaps "
2816                                  "superblock", i);
2817                         if (!sb_rdonly(sb))
2818                                 return 0;
2819                 }
2820                 if (inode_table >= sb_block + 1 &&
2821                     inode_table <= last_bg_block) {
2822                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2823                                  "Inode table for group %u overlaps "
2824                                  "block group descriptors", i);
2825                         if (!sb_rdonly(sb))
2826                                 return 0;
2827                 }
2828                 if (inode_table < first_block ||
2829                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2830                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2831                                "Inode table for group %u not in group "
2832                                "(block %llu)!", i, inode_table);
2833                         return 0;
2834                 }
2835                 ext4_lock_group(sb, i);
2836                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2837                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2838                                  "Checksum for group %u failed (%u!=%u)",
2839                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2840                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2841                         if (!sb_rdonly(sb)) {
2842                                 ext4_unlock_group(sb, i);
2843                                 return 0;
2844                         }
2845                 }
2846                 ext4_unlock_group(sb, i);
2847                 if (!flexbg_flag)
2848                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2849         }
2850         if (NULL != first_not_zeroed)
2851                 *first_not_zeroed = grp;
2852         return 1;
2853 }
2854
2855 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2856  * the superblock) which were deleted from all directories, but held open by
2857  * a process at the time of a crash.  We walk the list and try to delete these
2858  * inodes at recovery time (only with a read-write filesystem).
2859  *
2860  * In order to keep the orphan inode chain consistent during traversal (in
2861  * case of crash during recovery), we link each inode into the superblock
2862  * orphan list_head and handle it the same way as an inode deletion during
2863  * normal operation (which journals the operations for us).
2864  *
2865  * We only do an iget() and an iput() on each inode, which is very safe if we
2866  * accidentally point at an in-use or already deleted inode.  The worst that
2867  * can happen in this case is that we get a "bit already cleared" message from
2868  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2869  * e2fsck was run on this filesystem, and it must have already done the orphan
2870  * inode cleanup for us, so we can safely abort without any further action.
2871  */
2872 static void ext4_orphan_cleanup(struct super_block *sb,
2873                                 struct ext4_super_block *es)
2874 {
2875         unsigned int s_flags = sb->s_flags;
2876         int ret, nr_orphans = 0, nr_truncates = 0;
2877 #ifdef CONFIG_QUOTA
2878         int quota_update = 0;
2879         int i;
2880 #endif
2881         if (!es->s_last_orphan) {
2882                 jbd_debug(4, "no orphan inodes to clean up\n");
2883                 return;
2884         }
2885
2886         if (bdev_read_only(sb->s_bdev)) {
2887                 ext4_msg(sb, KERN_ERR, "write access "
2888                         "unavailable, skipping orphan cleanup");
2889                 return;
2890         }
2891
2892         /* Check if feature set would not allow a r/w mount */
2893         if (!ext4_feature_set_ok(sb, 0)) {
2894                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2895                          "unknown ROCOMPAT features");
2896                 return;
2897         }
2898
2899         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2900                 /* don't clear list on RO mount w/ errors */
2901                 if (es->s_last_orphan && !(s_flags & SB_RDONLY)) {
2902                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2903                                   "clearing orphan list.\n");
2904                         es->s_last_orphan = 0;
2905                 }
2906                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2907                 return;
2908         }
2909
2910         if (s_flags & SB_RDONLY) {
2911                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2912                 sb->s_flags &= ~SB_RDONLY;
2913         }
2914 #ifdef CONFIG_QUOTA
2915         /* Needed for iput() to work correctly and not trash data */
2916         sb->s_flags |= SB_ACTIVE;
2917
2918         /*
2919          * Turn on quotas which were not enabled for read-only mounts if
2920          * filesystem has quota feature, so that they are updated correctly.
2921          */
2922         if (ext4_has_feature_quota(sb) && (s_flags & SB_RDONLY)) {
2923                 int ret = ext4_enable_quotas(sb);
2924
2925                 if (!ret)
2926                         quota_update = 1;
2927                 else
2928                         ext4_msg(sb, KERN_ERR,
2929                                 "Cannot turn on quotas: error %d", ret);
2930         }
2931
2932         /* Turn on journaled quotas used for old sytle */
2933         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2934                 if (EXT4_SB(sb)->s_qf_names[i]) {
2935                         int ret = ext4_quota_on_mount(sb, i);
2936
2937                         if (!ret)
2938                                 quota_update = 1;
2939                         else
2940                                 ext4_msg(sb, KERN_ERR,
2941                                         "Cannot turn on journaled "
2942                                         "quota: type %d: error %d", i, ret);
2943                 }
2944         }
2945 #endif
2946
2947         while (es->s_last_orphan) {
2948                 struct inode *inode;
2949
2950                 /*
2951                  * We may have encountered an error during cleanup; if
2952                  * so, skip the rest.
2953                  */
2954                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2955                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2956                         es->s_last_orphan = 0;
2957                         break;
2958                 }
2959
2960                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2961                 if (IS_ERR(inode)) {
2962                         es->s_last_orphan = 0;
2963                         break;
2964                 }
2965
2966                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2967                 dquot_initialize(inode);
2968                 if (inode->i_nlink) {
2969                         if (test_opt(sb, DEBUG))
2970                                 ext4_msg(sb, KERN_DEBUG,
2971                                         "%s: truncating inode %lu to %lld bytes",
2972                                         __func__, inode->i_ino, inode->i_size);
2973                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2974                                   inode->i_ino, inode->i_size);
2975                         inode_lock(inode);
2976                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2977                         ret = ext4_truncate(inode);
2978                         if (ret)
2979                                 ext4_std_error(inode->i_sb, ret);
2980                         inode_unlock(inode);
2981                         nr_truncates++;
2982                 } else {
2983                         if (test_opt(sb, DEBUG))
2984                                 ext4_msg(sb, KERN_DEBUG,
2985                                         "%s: deleting unreferenced inode %lu",
2986                                         __func__, inode->i_ino);
2987                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2988                                   inode->i_ino);
2989                         nr_orphans++;
2990                 }
2991                 iput(inode);  /* The delete magic happens here! */
2992         }
2993
2994 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2995
2996         if (nr_orphans)
2997                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2998                        PLURAL(nr_orphans));
2999         if (nr_truncates)
3000                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
3001                        PLURAL(nr_truncates));
3002 #ifdef CONFIG_QUOTA
3003         /* Turn off quotas if they were enabled for orphan cleanup */
3004         if (quota_update) {
3005                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
3006                         if (sb_dqopt(sb)->files[i])
3007                                 dquot_quota_off(sb, i);
3008                 }
3009         }
3010 #endif
3011         sb->s_flags = s_flags; /* Restore SB_RDONLY status */
3012 }
3013
3014 /*
3015  * Maximal extent format file size.
3016  * Resulting logical blkno at s_maxbytes must fit in our on-disk
3017  * extent format containers, within a sector_t, and within i_blocks
3018  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
3019  * so that won't be a limiting factor.
3020  *
3021  * However there is other limiting factor. We do store extents in the form
3022  * of starting block and length, hence the resulting length of the extent
3023  * covering maximum file size must fit into on-disk format containers as
3024  * well. Given that length is always by 1 unit bigger than max unit (because
3025  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
3026  *
3027  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
3028  */
3029 static loff_t ext4_max_size(int blkbits, int has_huge_files)
3030 {
3031         loff_t res;
3032         loff_t upper_limit = MAX_LFS_FILESIZE;
3033
3034         BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64));
3035
3036         if (!has_huge_files) {
3037                 upper_limit = (1LL << 32) - 1;
3038
3039                 /* total blocks in file system block size */
3040                 upper_limit >>= (blkbits - 9);
3041                 upper_limit <<= blkbits;
3042         }
3043
3044         /*
3045          * 32-bit extent-start container, ee_block. We lower the maxbytes
3046          * by one fs block, so ee_len can cover the extent of maximum file
3047          * size
3048          */
3049         res = (1LL << 32) - 1;
3050         res <<= blkbits;
3051
3052         /* Sanity check against vm- & vfs- imposed limits */
3053         if (res > upper_limit)
3054                 res = upper_limit;
3055
3056         return res;
3057 }
3058
3059 /*
3060  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
3061  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
3062  * We need to be 1 filesystem block less than the 2^48 sector limit.
3063  */
3064 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
3065 {
3066         loff_t res = EXT4_NDIR_BLOCKS;
3067         int meta_blocks;
3068         loff_t upper_limit;
3069         /* This is calculated to be the largest file size for a dense, block
3070          * mapped file such that the file's total number of 512-byte sectors,
3071          * including data and all indirect blocks, does not exceed (2^48 - 1).
3072          *
3073          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
3074          * number of 512-byte sectors of the file.
3075          */
3076
3077         if (!has_huge_files) {
3078                 /*
3079                  * !has_huge_files or implies that the inode i_block field
3080                  * represents total file blocks in 2^32 512-byte sectors ==
3081                  * size of vfs inode i_blocks * 8
3082                  */
3083                 upper_limit = (1LL << 32) - 1;
3084
3085                 /* total blocks in file system block size */
3086                 upper_limit >>= (bits - 9);
3087
3088         } else {
3089                 /*
3090                  * We use 48 bit ext4_inode i_blocks
3091                  * With EXT4_HUGE_FILE_FL set the i_blocks
3092                  * represent total number of blocks in
3093                  * file system block size
3094                  */
3095                 upper_limit = (1LL << 48) - 1;
3096
3097         }
3098
3099         /* indirect blocks */
3100         meta_blocks = 1;
3101         /* double indirect blocks */
3102         meta_blocks += 1 + (1LL << (bits-2));
3103         /* tripple indirect blocks */
3104         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
3105
3106         upper_limit -= meta_blocks;
3107         upper_limit <<= bits;
3108
3109         res += 1LL << (bits-2);
3110         res += 1LL << (2*(bits-2));
3111         res += 1LL << (3*(bits-2));
3112         res <<= bits;
3113         if (res > upper_limit)
3114                 res = upper_limit;
3115
3116         if (res > MAX_LFS_FILESIZE)
3117                 res = MAX_LFS_FILESIZE;
3118
3119         return res;
3120 }
3121
3122 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
3123                                    ext4_fsblk_t logical_sb_block, int nr)
3124 {
3125         struct ext4_sb_info *sbi = EXT4_SB(sb);
3126         ext4_group_t bg, first_meta_bg;
3127         int has_super = 0;
3128
3129         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
3130
3131         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
3132                 return logical_sb_block + nr + 1;
3133         bg = sbi->s_desc_per_block * nr;
3134         if (ext4_bg_has_super(sb, bg))
3135                 has_super = 1;
3136
3137         /*
3138          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
3139          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
3140          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
3141          * compensate.
3142          */
3143         if (sb->s_blocksize == 1024 && nr == 0 &&
3144             le32_to_cpu(sbi->s_es->s_first_data_block) == 0)
3145                 has_super++;
3146
3147         return (has_super + ext4_group_first_block_no(sb, bg));
3148 }
3149
3150 /**
3151  * ext4_get_stripe_size: Get the stripe size.
3152  * @sbi: In memory super block info
3153  *
3154  * If we have specified it via mount option, then
3155  * use the mount option value. If the value specified at mount time is
3156  * greater than the blocks per group use the super block value.
3157  * If the super block value is greater than blocks per group return 0.
3158  * Allocator needs it be less than blocks per group.
3159  *
3160  */
3161 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
3162 {
3163         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
3164         unsigned long stripe_width =
3165                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
3166         int ret;
3167
3168         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
3169                 ret = sbi->s_stripe;
3170         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
3171                 ret = stripe_width;
3172         else if (stride && stride <= sbi->s_blocks_per_group)
3173                 ret = stride;
3174         else
3175                 ret = 0;
3176
3177         /*
3178          * If the stripe width is 1, this makes no sense and
3179          * we set it to 0 to turn off stripe handling code.
3180          */
3181         if (ret <= 1)
3182                 ret = 0;
3183
3184         return ret;
3185 }
3186
3187 /*
3188  * Check whether this filesystem can be mounted based on
3189  * the features present and the RDONLY/RDWR mount requested.
3190  * Returns 1 if this filesystem can be mounted as requested,
3191  * 0 if it cannot be.
3192  */
3193 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
3194 {
3195         if (ext4_has_unknown_ext4_incompat_features(sb)) {
3196                 ext4_msg(sb, KERN_ERR,
3197                         "Couldn't mount because of "
3198                         "unsupported optional features (%x)",
3199                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
3200                         ~EXT4_FEATURE_INCOMPAT_SUPP));
3201                 return 0;
3202         }
3203
3204 #ifndef CONFIG_UNICODE
3205         if (ext4_has_feature_casefold(sb)) {
3206                 ext4_msg(sb, KERN_ERR,
3207                          "Filesystem with casefold feature cannot be "
3208                          "mounted without CONFIG_UNICODE");
3209                 return 0;
3210         }
3211 #endif
3212
3213         if (readonly)
3214                 return 1;
3215
3216         if (ext4_has_feature_readonly(sb)) {
3217                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
3218                 sb->s_flags |= SB_RDONLY;
3219                 return 1;
3220         }
3221
3222         /* Check that feature set is OK for a read-write mount */
3223         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
3224                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
3225                          "unsupported optional features (%x)",
3226                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
3227                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
3228                 return 0;
3229         }
3230         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
3231                 ext4_msg(sb, KERN_ERR,
3232                          "Can't support bigalloc feature without "
3233                          "extents feature\n");
3234                 return 0;
3235         }
3236
3237 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2)
3238         if (!readonly && (ext4_has_feature_quota(sb) ||
3239                           ext4_has_feature_project(sb))) {
3240                 ext4_msg(sb, KERN_ERR,
3241                          "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2");
3242                 return 0;
3243         }
3244 #endif  /* CONFIG_QUOTA */
3245         return 1;
3246 }
3247
3248 /*
3249  * This function is called once a day if we have errors logged
3250  * on the file system
3251  */
3252 static void print_daily_error_info(struct timer_list *t)
3253 {
3254         struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report);
3255         struct super_block *sb = sbi->s_sb;
3256         struct ext4_super_block *es = sbi->s_es;
3257
3258         if (es->s_error_count)
3259                 /* fsck newer than v1.41.13 is needed to clean this condition. */
3260                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
3261                          le32_to_cpu(es->s_error_count));
3262         if (es->s_first_error_time) {
3263                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d",
3264                        sb->s_id,
3265                        ext4_get_tstamp(es, s_first_error_time),
3266                        (int) sizeof(es->s_first_error_func),
3267                        es->s_first_error_func,
3268                        le32_to_cpu(es->s_first_error_line));
3269                 if (es->s_first_error_ino)
3270                         printk(KERN_CONT ": inode %u",
3271                                le32_to_cpu(es->s_first_error_ino));
3272                 if (es->s_first_error_block)
3273                         printk(KERN_CONT ": block %llu", (unsigned long long)
3274                                le64_to_cpu(es->s_first_error_block));
3275                 printk(KERN_CONT "\n");
3276         }
3277         if (es->s_last_error_time) {
3278                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d",
3279                        sb->s_id,
3280                        ext4_get_tstamp(es, s_last_error_time),
3281                        (int) sizeof(es->s_last_error_func),
3282                        es->s_last_error_func,
3283                        le32_to_cpu(es->s_last_error_line));
3284                 if (es->s_last_error_ino)
3285                         printk(KERN_CONT ": inode %u",
3286                                le32_to_cpu(es->s_last_error_ino));
3287                 if (es->s_last_error_block)
3288                         printk(KERN_CONT ": block %llu", (unsigned long long)
3289                                le64_to_cpu(es->s_last_error_block));
3290                 printk(KERN_CONT "\n");
3291         }
3292         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
3293 }
3294
3295 /* Find next suitable group and run ext4_init_inode_table */
3296 static int ext4_run_li_request(struct ext4_li_request *elr)
3297 {
3298         struct ext4_group_desc *gdp = NULL;
3299         struct super_block *sb = elr->lr_super;
3300         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3301         ext4_group_t group = elr->lr_next_group;
3302         unsigned long timeout = 0;
3303         unsigned int prefetch_ios = 0;
3304         int ret = 0;
3305
3306         if (elr->lr_mode == EXT4_LI_MODE_PREFETCH_BBITMAP) {
3307                 elr->lr_next_group = ext4_mb_prefetch(sb, group,
3308                                 EXT4_SB(sb)->s_mb_prefetch, &prefetch_ios);
3309                 if (prefetch_ios)
3310                         ext4_mb_prefetch_fini(sb, elr->lr_next_group,
3311                                               prefetch_ios);
3312                 trace_ext4_prefetch_bitmaps(sb, group, elr->lr_next_group,
3313                                             prefetch_ios);
3314                 if (group >= elr->lr_next_group) {
3315                         ret = 1;
3316                         if (elr->lr_first_not_zeroed != ngroups &&
3317                             !sb_rdonly(sb) && test_opt(sb, INIT_INODE_TABLE)) {
3318                                 elr->lr_next_group = elr->lr_first_not_zeroed;
3319                                 elr->lr_mode = EXT4_LI_MODE_ITABLE;
3320                                 ret = 0;
3321                         }
3322                 }
3323                 return ret;
3324         }
3325
3326         for (; group < ngroups; group++) {
3327                 gdp = ext4_get_group_desc(sb, group, NULL);
3328                 if (!gdp) {
3329                         ret = 1;
3330                         break;
3331                 }
3332
3333                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3334                         break;
3335         }
3336
3337         if (group >= ngroups)
3338                 ret = 1;
3339
3340         if (!ret) {
3341                 timeout = jiffies;
3342                 ret = ext4_init_inode_table(sb, group,
3343                                             elr->lr_timeout ? 0 : 1);
3344                 trace_ext4_lazy_itable_init(sb, group);
3345                 if (elr->lr_timeout == 0) {
3346                         timeout = (jiffies - timeout) *
3347                                 EXT4_SB(elr->lr_super)->s_li_wait_mult;
3348                         elr->lr_timeout = timeout;
3349                 }
3350                 elr->lr_next_sched = jiffies + elr->lr_timeout;
3351                 elr->lr_next_group = group + 1;
3352         }
3353         return ret;
3354 }
3355
3356 /*
3357  * Remove lr_request from the list_request and free the
3358  * request structure. Should be called with li_list_mtx held
3359  */
3360 static void ext4_remove_li_request(struct ext4_li_request *elr)
3361 {
3362         if (!elr)
3363                 return;
3364
3365         list_del(&elr->lr_request);
3366         EXT4_SB(elr->lr_super)->s_li_request = NULL;
3367         kfree(elr);
3368 }
3369
3370 static void ext4_unregister_li_request(struct super_block *sb)
3371 {
3372         mutex_lock(&ext4_li_mtx);
3373         if (!ext4_li_info) {
3374                 mutex_unlock(&ext4_li_mtx);
3375                 return;
3376         }
3377
3378         mutex_lock(&ext4_li_info->li_list_mtx);
3379         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
3380         mutex_unlock(&ext4_li_info->li_list_mtx);
3381         mutex_unlock(&ext4_li_mtx);
3382 }
3383
3384 static struct task_struct *ext4_lazyinit_task;
3385
3386 /*
3387  * This is the function where ext4lazyinit thread lives. It walks
3388  * through the request list searching for next scheduled filesystem.
3389  * When such a fs is found, run the lazy initialization request
3390  * (ext4_rn_li_request) and keep track of the time spend in this
3391  * function. Based on that time we compute next schedule time of
3392  * the request. When walking through the list is complete, compute
3393  * next waking time and put itself into sleep.
3394  */
3395 static int ext4_lazyinit_thread(void *arg)
3396 {
3397         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
3398         struct list_head *pos, *n;
3399         struct ext4_li_request *elr;
3400         unsigned long next_wakeup, cur;
3401
3402         BUG_ON(NULL == eli);
3403
3404 cont_thread:
3405         while (true) {
3406                 next_wakeup = MAX_JIFFY_OFFSET;
3407
3408                 mutex_lock(&eli->li_list_mtx);
3409                 if (list_empty(&eli->li_request_list)) {
3410                         mutex_unlock(&eli->li_list_mtx);
3411                         goto exit_thread;
3412                 }
3413                 list_for_each_safe(pos, n, &eli->li_request_list) {
3414                         int err = 0;
3415                         int progress = 0;
3416                         elr = list_entry(pos, struct ext4_li_request,
3417                                          lr_request);
3418
3419                         if (time_before(jiffies, elr->lr_next_sched)) {
3420                                 if (time_before(elr->lr_next_sched, next_wakeup))
3421                                         next_wakeup = elr->lr_next_sched;
3422                                 continue;
3423                         }
3424                         if (down_read_trylock(&elr->lr_super->s_umount)) {
3425                                 if (sb_start_write_trylock(elr->lr_super)) {
3426                                         progress = 1;
3427                                         /*
3428                                          * We hold sb->s_umount, sb can not
3429                                          * be removed from the list, it is
3430                                          * now safe to drop li_list_mtx
3431                                          */
3432                                         mutex_unlock(&eli->li_list_mtx);
3433                                         err = ext4_run_li_request(elr);
3434                                         sb_end_write(elr->lr_super);
3435                                         mutex_lock(&eli->li_list_mtx);
3436                                         n = pos->next;
3437                                 }
3438                                 up_read((&elr->lr_super->s_umount));
3439                         }
3440                         /* error, remove the lazy_init job */
3441                         if (err) {
3442                                 ext4_remove_li_request(elr);
3443                                 continue;
3444                         }
3445                         if (!progress) {
3446                                 elr->lr_next_sched = jiffies +
3447                                         (prandom_u32()
3448                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3449                         }
3450                         if (time_before(elr->lr_next_sched, next_wakeup))
3451                                 next_wakeup = elr->lr_next_sched;
3452                 }
3453                 mutex_unlock(&eli->li_list_mtx);
3454
3455                 try_to_freeze();
3456
3457                 cur = jiffies;
3458                 if ((time_after_eq(cur, next_wakeup)) ||
3459                     (MAX_JIFFY_OFFSET == next_wakeup)) {
3460                         cond_resched();
3461                         continue;
3462                 }
3463
3464                 schedule_timeout_interruptible(next_wakeup - cur);
3465
3466                 if (kthread_should_stop()) {
3467                         ext4_clear_request_list();
3468                         goto exit_thread;
3469                 }
3470         }
3471
3472 exit_thread:
3473         /*
3474          * It looks like the request list is empty, but we need
3475          * to check it under the li_list_mtx lock, to prevent any
3476          * additions into it, and of course we should lock ext4_li_mtx
3477          * to atomically free the list and ext4_li_info, because at
3478          * this point another ext4 filesystem could be registering
3479          * new one.
3480          */
3481         mutex_lock(&ext4_li_mtx);
3482         mutex_lock(&eli->li_list_mtx);
3483         if (!list_empty(&eli->li_request_list)) {
3484                 mutex_unlock(&eli->li_list_mtx);
3485                 mutex_unlock(&ext4_li_mtx);
3486                 goto cont_thread;
3487         }
3488         mutex_unlock(&eli->li_list_mtx);
3489         kfree(ext4_li_info);
3490         ext4_li_info = NULL;
3491         mutex_unlock(&ext4_li_mtx);
3492
3493         return 0;
3494 }
3495
3496 static void ext4_clear_request_list(void)
3497 {
3498         struct list_head *pos, *n;
3499         struct ext4_li_request *elr;
3500
3501         mutex_lock(&ext4_li_info->li_list_mtx);
3502         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3503                 elr = list_entry(pos, struct ext4_li_request,
3504                                  lr_request);
3505                 ext4_remove_li_request(elr);
3506         }
3507         mutex_unlock(&ext4_li_info->li_list_mtx);
3508 }
3509
3510 static int ext4_run_lazyinit_thread(void)
3511 {
3512         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3513                                          ext4_li_info, "ext4lazyinit");
3514         if (IS_ERR(ext4_lazyinit_task)) {
3515                 int err = PTR_ERR(ext4_lazyinit_task);
3516                 ext4_clear_request_list();
3517                 kfree(ext4_li_info);
3518                 ext4_li_info = NULL;
3519                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3520                                  "initialization thread\n",
3521                                  err);
3522                 return err;
3523         }
3524         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3525         return 0;
3526 }
3527
3528 /*
3529  * Check whether it make sense to run itable init. thread or not.
3530  * If there is at least one uninitialized inode table, return
3531  * corresponding group number, else the loop goes through all
3532  * groups and return total number of groups.
3533  */
3534 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3535 {
3536         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3537         struct ext4_group_desc *gdp = NULL;
3538
3539         if (!ext4_has_group_desc_csum(sb))
3540                 return ngroups;
3541
3542         for (group = 0; group < ngroups; group++) {
3543                 gdp = ext4_get_group_desc(sb, group, NULL);
3544                 if (!gdp)
3545                         continue;
3546
3547                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3548                         break;
3549         }
3550
3551         return group;
3552 }
3553
3554 static int ext4_li_info_new(void)
3555 {
3556         struct ext4_lazy_init *eli = NULL;
3557
3558         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3559         if (!eli)
3560                 return -ENOMEM;
3561
3562         INIT_LIST_HEAD(&eli->li_request_list);
3563         mutex_init(&eli->li_list_mtx);
3564
3565         eli->li_state |= EXT4_LAZYINIT_QUIT;
3566
3567         ext4_li_info = eli;
3568
3569         return 0;
3570 }
3571
3572 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3573                                             ext4_group_t start)
3574 {
3575         struct ext4_li_request *elr;
3576
3577         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3578         if (!elr)
3579                 return NULL;
3580
3581         elr->lr_super = sb;
3582         elr->lr_first_not_zeroed = start;
3583         if (test_opt(sb, PREFETCH_BLOCK_BITMAPS))
3584                 elr->lr_mode = EXT4_LI_MODE_PREFETCH_BBITMAP;
3585         else {
3586                 elr->lr_mode = EXT4_LI_MODE_ITABLE;
3587                 elr->lr_next_group = start;
3588         }
3589
3590         /*
3591          * Randomize first schedule time of the request to
3592          * spread the inode table initialization requests
3593          * better.
3594          */
3595         elr->lr_next_sched = jiffies + (prandom_u32() %
3596                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3597         return elr;
3598 }
3599
3600 int ext4_register_li_request(struct super_block *sb,
3601                              ext4_group_t first_not_zeroed)
3602 {
3603         struct ext4_sb_info *sbi = EXT4_SB(sb);
3604         struct ext4_li_request *elr = NULL;
3605         ext4_group_t ngroups = sbi->s_groups_count;
3606         int ret = 0;
3607
3608         mutex_lock(&ext4_li_mtx);
3609         if (sbi->s_li_request != NULL) {
3610                 /*
3611                  * Reset timeout so it can be computed again, because
3612                  * s_li_wait_mult might have changed.
3613                  */
3614                 sbi->s_li_request->lr_timeout = 0;
3615                 goto out;
3616         }
3617
3618         if (!test_opt(sb, PREFETCH_BLOCK_BITMAPS) &&
3619             (first_not_zeroed == ngroups || sb_rdonly(sb) ||
3620              !test_opt(sb, INIT_INODE_TABLE)))
3621                 goto out;
3622
3623         elr = ext4_li_request_new(sb, first_not_zeroed);
3624         if (!elr) {
3625                 ret = -ENOMEM;
3626                 goto out;
3627         }
3628
3629         if (NULL == ext4_li_info) {
3630                 ret = ext4_li_info_new();
3631                 if (ret)
3632                         goto out;
3633         }
3634
3635         mutex_lock(&ext4_li_info->li_list_mtx);
3636         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3637         mutex_unlock(&ext4_li_info->li_list_mtx);
3638
3639         sbi->s_li_request = elr;
3640         /*
3641          * set elr to NULL here since it has been inserted to
3642          * the request_list and the removal and free of it is
3643          * handled by ext4_clear_request_list from now on.
3644          */
3645         elr = NULL;
3646
3647         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3648                 ret = ext4_run_lazyinit_thread();
3649                 if (ret)
3650                         goto out;
3651         }
3652 out:
3653         mutex_unlock(&ext4_li_mtx);
3654         if (ret)
3655                 kfree(elr);
3656         return ret;
3657 }
3658
3659 /*
3660  * We do not need to lock anything since this is called on
3661  * module unload.
3662  */
3663 static void ext4_destroy_lazyinit_thread(void)
3664 {
3665         /*
3666          * If thread exited earlier
3667          * there's nothing to be done.
3668          */
3669         if (!ext4_li_info || !ext4_lazyinit_task)
3670                 return;
3671
3672         kthread_stop(ext4_lazyinit_task);
3673 }
3674
3675 static int set_journal_csum_feature_set(struct super_block *sb)
3676 {
3677         int ret = 1;
3678         int compat, incompat;
3679         struct ext4_sb_info *sbi = EXT4_SB(sb);
3680
3681         if (ext4_has_metadata_csum(sb)) {
3682                 /* journal checksum v3 */
3683                 compat = 0;
3684                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3685         } else {
3686                 /* journal checksum v1 */
3687                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3688                 incompat = 0;
3689         }
3690
3691         jbd2_journal_clear_features(sbi->s_journal,
3692                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3693                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3694                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3695         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3696                 ret = jbd2_journal_set_features(sbi->s_journal,
3697                                 compat, 0,
3698                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3699                                 incompat);
3700         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3701                 ret = jbd2_journal_set_features(sbi->s_journal,
3702                                 compat, 0,
3703                                 incompat);
3704                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3705                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3706         } else {
3707                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3708                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3709         }
3710
3711         return ret;
3712 }
3713
3714 /*
3715  * Note: calculating the overhead so we can be compatible with
3716  * historical BSD practice is quite difficult in the face of
3717  * clusters/bigalloc.  This is because multiple metadata blocks from
3718  * different block group can end up in the same allocation cluster.
3719  * Calculating the exact overhead in the face of clustered allocation
3720  * requires either O(all block bitmaps) in memory or O(number of block
3721  * groups**2) in time.  We will still calculate the superblock for
3722  * older file systems --- and if we come across with a bigalloc file
3723  * system with zero in s_overhead_clusters the estimate will be close to
3724  * correct especially for very large cluster sizes --- but for newer
3725  * file systems, it's better to calculate this figure once at mkfs
3726  * time, and store it in the superblock.  If the superblock value is
3727  * present (even for non-bigalloc file systems), we will use it.
3728  */
3729 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3730                           char *buf)
3731 {
3732         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3733         struct ext4_group_desc  *gdp;
3734         ext4_fsblk_t            first_block, last_block, b;
3735         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3736         int                     s, j, count = 0;
3737
3738         if (!ext4_has_feature_bigalloc(sb))
3739                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3740                         sbi->s_itb_per_group + 2);
3741
3742         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3743                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3744         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3745         for (i = 0; i < ngroups; i++) {
3746                 gdp = ext4_get_group_desc(sb, i, NULL);
3747                 b = ext4_block_bitmap(sb, gdp);
3748                 if (b >= first_block && b <= last_block) {
3749                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3750                         count++;
3751                 }
3752                 b = ext4_inode_bitmap(sb, gdp);
3753                 if (b >= first_block && b <= last_block) {
3754                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3755                         count++;
3756                 }
3757                 b = ext4_inode_table(sb, gdp);
3758                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3759                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3760                                 int c = EXT4_B2C(sbi, b - first_block);
3761                                 ext4_set_bit(c, buf);
3762                                 count++;
3763                         }
3764                 if (i != grp)
3765                         continue;
3766                 s = 0;
3767                 if (ext4_bg_has_super(sb, grp)) {
3768                         ext4_set_bit(s++, buf);
3769                         count++;
3770                 }
3771                 j = ext4_bg_num_gdb(sb, grp);
3772                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3773                         ext4_error(sb, "Invalid number of block group "
3774                                    "descriptor blocks: %d", j);
3775                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3776                 }
3777                 count += j;
3778                 for (; j > 0; j--)
3779                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3780         }
3781         if (!count)
3782                 return 0;
3783         return EXT4_CLUSTERS_PER_GROUP(sb) -
3784                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3785 }
3786
3787 /*
3788  * Compute the overhead and stash it in sbi->s_overhead
3789  */
3790 int ext4_calculate_overhead(struct super_block *sb)
3791 {
3792         struct ext4_sb_info *sbi = EXT4_SB(sb);
3793         struct ext4_super_block *es = sbi->s_es;
3794         struct inode *j_inode;
3795         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3796         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3797         ext4_fsblk_t overhead = 0;
3798         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3799
3800         if (!buf)
3801                 return -ENOMEM;
3802
3803         /*
3804          * Compute the overhead (FS structures).  This is constant
3805          * for a given filesystem unless the number of block groups
3806          * changes so we cache the previous value until it does.
3807          */
3808
3809         /*
3810          * All of the blocks before first_data_block are overhead
3811          */
3812         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3813
3814         /*
3815          * Add the overhead found in each block group
3816          */
3817         for (i = 0; i < ngroups; i++) {
3818                 int blks;
3819
3820                 blks = count_overhead(sb, i, buf);
3821                 overhead += blks;
3822                 if (blks)
3823                         memset(buf, 0, PAGE_SIZE);
3824                 cond_resched();
3825         }
3826
3827         /*
3828          * Add the internal journal blocks whether the journal has been
3829          * loaded or not
3830          */
3831         if (sbi->s_journal && !sbi->s_journal_bdev)
3832                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3833         else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) {
3834                 /* j_inum for internal journal is non-zero */
3835                 j_inode = ext4_get_journal_inode(sb, j_inum);
3836                 if (j_inode) {
3837                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3838                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
3839                         iput(j_inode);
3840                 } else {
3841                         ext4_msg(sb, KERN_ERR, "can't get journal size");
3842                 }
3843         }
3844         sbi->s_overhead = overhead;
3845         smp_wmb();
3846         free_page((unsigned long) buf);
3847         return 0;
3848 }
3849
3850 static void ext4_set_resv_clusters(struct super_block *sb)
3851 {
3852         ext4_fsblk_t resv_clusters;
3853         struct ext4_sb_info *sbi = EXT4_SB(sb);
3854
3855         /*
3856          * There's no need to reserve anything when we aren't using extents.
3857          * The space estimates are exact, there are no unwritten extents,
3858          * hole punching doesn't need new metadata... This is needed especially
3859          * to keep ext2/3 backward compatibility.
3860          */
3861         if (!ext4_has_feature_extents(sb))
3862                 return;
3863         /*
3864          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3865          * This should cover the situations where we can not afford to run
3866          * out of space like for example punch hole, or converting
3867          * unwritten extents in delalloc path. In most cases such
3868          * allocation would require 1, or 2 blocks, higher numbers are
3869          * very rare.
3870          */
3871         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3872                          sbi->s_cluster_bits);
3873
3874         do_div(resv_clusters, 50);
3875         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3876
3877         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3878 }
3879
3880 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3881 {
3882         struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev);
3883         char *orig_data = kstrdup(data, GFP_KERNEL);
3884         struct buffer_head *bh, **group_desc;
3885         struct ext4_super_block *es = NULL;
3886         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3887         struct flex_groups **flex_groups;
3888         ext4_fsblk_t block;
3889         ext4_fsblk_t sb_block = get_sb_block(&data);
3890         ext4_fsblk_t logical_sb_block;
3891         unsigned long offset = 0;
3892         unsigned long journal_devnum = 0;
3893         unsigned long def_mount_opts;
3894         struct inode *root;
3895         const char *descr;
3896         int ret = -ENOMEM;
3897         int blocksize, clustersize;
3898         unsigned int db_count;
3899         unsigned int i;
3900         int needs_recovery, has_huge_files;
3901         __u64 blocks_count;
3902         int err = 0;
3903         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3904         ext4_group_t first_not_zeroed;
3905
3906         if ((data && !orig_data) || !sbi)
3907                 goto out_free_base;
3908
3909         sbi->s_daxdev = dax_dev;
3910         sbi->s_blockgroup_lock =
3911                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3912         if (!sbi->s_blockgroup_lock)
3913                 goto out_free_base;
3914
3915         sb->s_fs_info = sbi;
3916         sbi->s_sb = sb;
3917         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3918         sbi->s_sb_block = sb_block;
3919         if (sb->s_bdev->bd_part)
3920                 sbi->s_sectors_written_start =
3921                         part_stat_read(sb->s_bdev->bd_part, sectors[STAT_WRITE]);
3922
3923         /* Cleanup superblock name */
3924         strreplace(sb->s_id, '/', '!');
3925
3926         /* -EINVAL is default */
3927         ret = -EINVAL;
3928         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3929         if (!blocksize) {
3930                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3931                 goto out_fail;
3932         }
3933
3934         /*
3935          * The ext4 superblock will not be buffer aligned for other than 1kB
3936          * block sizes.  We need to calculate the offset from buffer start.
3937          */
3938         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3939                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3940                 offset = do_div(logical_sb_block, blocksize);
3941         } else {
3942                 logical_sb_block = sb_block;
3943         }
3944
3945         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3946                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3947                 goto out_fail;
3948         }
3949         /*
3950          * Note: s_es must be initialized as soon as possible because
3951          *       some ext4 macro-instructions depend on its value
3952          */
3953         es = (struct ext4_super_block *) (bh->b_data + offset);
3954         sbi->s_es = es;
3955         sb->s_magic = le16_to_cpu(es->s_magic);
3956         if (sb->s_magic != EXT4_SUPER_MAGIC)
3957                 goto cantfind_ext4;
3958         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3959
3960         /* Warn if metadata_csum and gdt_csum are both set. */
3961         if (ext4_has_feature_metadata_csum(sb) &&
3962             ext4_has_feature_gdt_csum(sb))
3963                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3964                              "redundant flags; please run fsck.");
3965
3966         /* Check for a known checksum algorithm */
3967         if (!ext4_verify_csum_type(sb, es)) {
3968                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3969                          "unknown checksum algorithm.");
3970                 silent = 1;
3971                 goto cantfind_ext4;
3972         }
3973
3974         /* Load the checksum driver */
3975         sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3976         if (IS_ERR(sbi->s_chksum_driver)) {
3977                 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3978                 ret = PTR_ERR(sbi->s_chksum_driver);
3979                 sbi->s_chksum_driver = NULL;
3980                 goto failed_mount;
3981         }
3982
3983         /* Check superblock checksum */
3984         if (!ext4_superblock_csum_verify(sb, es)) {
3985                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3986                          "invalid superblock checksum.  Run e2fsck?");
3987                 silent = 1;
3988                 ret = -EFSBADCRC;
3989                 goto cantfind_ext4;
3990         }
3991
3992         /* Precompute checksum seed for all metadata */
3993         if (ext4_has_feature_csum_seed(sb))
3994                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3995         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
3996                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3997                                                sizeof(es->s_uuid));
3998
3999         /* Set defaults before we parse the mount options */
4000         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
4001         set_opt(sb, INIT_INODE_TABLE);
4002         if (def_mount_opts & EXT4_DEFM_DEBUG)
4003                 set_opt(sb, DEBUG);
4004         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
4005                 set_opt(sb, GRPID);
4006         if (def_mount_opts & EXT4_DEFM_UID16)
4007                 set_opt(sb, NO_UID32);
4008         /* xattr user namespace & acls are now defaulted on */
4009         set_opt(sb, XATTR_USER);
4010 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4011         set_opt(sb, POSIX_ACL);
4012 #endif
4013         /* don't forget to enable journal_csum when metadata_csum is enabled. */
4014         if (ext4_has_metadata_csum(sb))
4015                 set_opt(sb, JOURNAL_CHECKSUM);
4016
4017         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
4018                 set_opt(sb, JOURNAL_DATA);
4019         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
4020                 set_opt(sb, ORDERED_DATA);
4021         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
4022                 set_opt(sb, WRITEBACK_DATA);
4023
4024         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
4025                 set_opt(sb, ERRORS_PANIC);
4026         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
4027                 set_opt(sb, ERRORS_CONT);
4028         else
4029                 set_opt(sb, ERRORS_RO);
4030         /* block_validity enabled by default; disable with noblock_validity */
4031         set_opt(sb, BLOCK_VALIDITY);
4032         if (def_mount_opts & EXT4_DEFM_DISCARD)
4033                 set_opt(sb, DISCARD);
4034
4035         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
4036         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
4037         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
4038         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
4039         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
4040
4041         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
4042                 set_opt(sb, BARRIER);
4043
4044         /*
4045          * enable delayed allocation by default
4046          * Use -o nodelalloc to turn it off
4047          */
4048         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
4049             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
4050                 set_opt(sb, DELALLOC);
4051
4052         /*
4053          * set default s_li_wait_mult for lazyinit, for the case there is
4054          * no mount option specified.
4055          */
4056         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
4057
4058         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
4059
4060         if (blocksize == PAGE_SIZE)
4061                 set_opt(sb, DIOREAD_NOLOCK);
4062
4063         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
4064             blocksize > EXT4_MAX_BLOCK_SIZE) {
4065                 ext4_msg(sb, KERN_ERR,
4066                        "Unsupported filesystem blocksize %d (%d log_block_size)",
4067                          blocksize, le32_to_cpu(es->s_log_block_size));
4068                 goto failed_mount;
4069         }
4070
4071         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
4072                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
4073                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
4074         } else {
4075                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
4076                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
4077                 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) {
4078                         ext4_msg(sb, KERN_ERR, "invalid first ino: %u",
4079                                  sbi->s_first_ino);
4080                         goto failed_mount;
4081                 }
4082                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
4083                     (!is_power_of_2(sbi->s_inode_size)) ||
4084                     (sbi->s_inode_size > blocksize)) {
4085                         ext4_msg(sb, KERN_ERR,
4086                                "unsupported inode size: %d",
4087                                sbi->s_inode_size);
4088                         ext4_msg(sb, KERN_ERR, "blocksize: %d", blocksize);
4089                         goto failed_mount;
4090                 }
4091                 /*
4092                  * i_atime_extra is the last extra field available for
4093                  * [acm]times in struct ext4_inode. Checking for that
4094                  * field should suffice to ensure we have extra space
4095                  * for all three.
4096                  */
4097                 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) +
4098                         sizeof(((struct ext4_inode *)0)->i_atime_extra)) {
4099                         sb->s_time_gran = 1;
4100                         sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX;
4101                 } else {
4102                         sb->s_time_gran = NSEC_PER_SEC;
4103                         sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX;
4104                 }
4105                 sb->s_time_min = EXT4_TIMESTAMP_MIN;
4106         }
4107         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
4108                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4109                         EXT4_GOOD_OLD_INODE_SIZE;
4110                 if (ext4_has_feature_extra_isize(sb)) {
4111                         unsigned v, max = (sbi->s_inode_size -
4112                                            EXT4_GOOD_OLD_INODE_SIZE);
4113
4114                         v = le16_to_cpu(es->s_want_extra_isize);
4115                         if (v > max) {
4116                                 ext4_msg(sb, KERN_ERR,
4117                                          "bad s_want_extra_isize: %d", v);
4118                                 goto failed_mount;
4119                         }
4120                         if (sbi->s_want_extra_isize < v)
4121                                 sbi->s_want_extra_isize = v;
4122
4123                         v = le16_to_cpu(es->s_min_extra_isize);
4124                         if (v > max) {
4125                                 ext4_msg(sb, KERN_ERR,
4126                                          "bad s_min_extra_isize: %d", v);
4127                                 goto failed_mount;
4128                         }
4129                         if (sbi->s_want_extra_isize < v)
4130                                 sbi->s_want_extra_isize = v;
4131                 }
4132         }
4133
4134         if (sbi->s_es->s_mount_opts[0]) {
4135                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
4136                                               sizeof(sbi->s_es->s_mount_opts),
4137                                               GFP_KERNEL);
4138                 if (!s_mount_opts)
4139                         goto failed_mount;
4140                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
4141                                    &journal_ioprio, 0)) {
4142                         ext4_msg(sb, KERN_WARNING,
4143                                  "failed to parse options in superblock: %s",
4144                                  s_mount_opts);
4145                 }
4146                 kfree(s_mount_opts);
4147         }
4148         sbi->s_def_mount_opt = sbi->s_mount_opt;
4149         if (!parse_options((char *) data, sb, &journal_devnum,
4150                            &journal_ioprio, 0))
4151                 goto failed_mount;
4152
4153 #ifdef CONFIG_UNICODE
4154         if (ext4_has_feature_casefold(sb) && !sbi->s_encoding) {
4155                 const struct ext4_sb_encodings *encoding_info;
4156                 struct unicode_map *encoding;
4157                 __u16 encoding_flags;
4158
4159                 if (ext4_has_feature_encrypt(sb)) {
4160                         ext4_msg(sb, KERN_ERR,
4161                                  "Can't mount with encoding and encryption");
4162                         goto failed_mount;
4163                 }
4164
4165                 if (ext4_sb_read_encoding(es, &encoding_info,
4166                                           &encoding_flags)) {
4167                         ext4_msg(sb, KERN_ERR,
4168                                  "Encoding requested by superblock is unknown");
4169                         goto failed_mount;
4170                 }
4171
4172                 encoding = utf8_load(encoding_info->version);
4173                 if (IS_ERR(encoding)) {
4174                         ext4_msg(sb, KERN_ERR,
4175                                  "can't mount with superblock charset: %s-%s "
4176                                  "not supported by the kernel. flags: 0x%x.",
4177                                  encoding_info->name, encoding_info->version,
4178                                  encoding_flags);
4179                         goto failed_mount;
4180                 }
4181                 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: "
4182                          "%s-%s with flags 0x%hx", encoding_info->name,
4183                          encoding_info->version?:"\b", encoding_flags);
4184
4185                 sbi->s_encoding = encoding;
4186                 sbi->s_encoding_flags = encoding_flags;
4187         }
4188 #endif
4189
4190         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4191                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with data=journal disables delayed allocation, dioread_nolock, and O_DIRECT support!\n");
4192                 /* can't mount with both data=journal and dioread_nolock. */
4193                 clear_opt(sb, DIOREAD_NOLOCK);
4194                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4195                         ext4_msg(sb, KERN_ERR, "can't mount with "
4196                                  "both data=journal and delalloc");
4197                         goto failed_mount;
4198                 }
4199                 if (test_opt(sb, DAX_ALWAYS)) {
4200                         ext4_msg(sb, KERN_ERR, "can't mount with "
4201                                  "both data=journal and dax");
4202                         goto failed_mount;
4203                 }
4204                 if (ext4_has_feature_encrypt(sb)) {
4205                         ext4_msg(sb, KERN_WARNING,
4206                                  "encrypted files will use data=ordered "
4207                                  "instead of data journaling mode");
4208                 }
4209                 if (test_opt(sb, DELALLOC))
4210                         clear_opt(sb, DELALLOC);
4211         } else {
4212                 sb->s_iflags |= SB_I_CGROUPWB;
4213         }
4214
4215         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
4216                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
4217
4218         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
4219             (ext4_has_compat_features(sb) ||
4220              ext4_has_ro_compat_features(sb) ||
4221              ext4_has_incompat_features(sb)))
4222                 ext4_msg(sb, KERN_WARNING,
4223                        "feature flags set on rev 0 fs, "
4224                        "running e2fsck is recommended");
4225
4226         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
4227                 set_opt2(sb, HURD_COMPAT);
4228                 if (ext4_has_feature_64bit(sb)) {
4229                         ext4_msg(sb, KERN_ERR,
4230                                  "The Hurd can't support 64-bit file systems");
4231                         goto failed_mount;
4232                 }
4233
4234                 /*
4235                  * ea_inode feature uses l_i_version field which is not
4236                  * available in HURD_COMPAT mode.
4237                  */
4238                 if (ext4_has_feature_ea_inode(sb)) {
4239                         ext4_msg(sb, KERN_ERR,
4240                                  "ea_inode feature is not supported for Hurd");
4241                         goto failed_mount;
4242                 }
4243         }
4244
4245         if (IS_EXT2_SB(sb)) {
4246                 if (ext2_feature_set_ok(sb))
4247                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
4248                                  "using the ext4 subsystem");
4249                 else {
4250                         /*
4251                          * If we're probing be silent, if this looks like
4252                          * it's actually an ext[34] filesystem.
4253                          */
4254                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4255                                 goto failed_mount;
4256                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
4257                                  "to feature incompatibilities");
4258                         goto failed_mount;
4259                 }
4260         }
4261
4262         if (IS_EXT3_SB(sb)) {
4263                 if (ext3_feature_set_ok(sb))
4264                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
4265                                  "using the ext4 subsystem");
4266                 else {
4267                         /*
4268                          * If we're probing be silent, if this looks like
4269                          * it's actually an ext4 filesystem.
4270                          */
4271                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4272                                 goto failed_mount;
4273                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
4274                                  "to feature incompatibilities");
4275                         goto failed_mount;
4276                 }
4277         }
4278
4279         /*
4280          * Check feature flags regardless of the revision level, since we
4281          * previously didn't change the revision level when setting the flags,
4282          * so there is a chance incompat flags are set on a rev 0 filesystem.
4283          */
4284         if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
4285                 goto failed_mount;
4286
4287         if (le32_to_cpu(es->s_log_block_size) >
4288             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4289                 ext4_msg(sb, KERN_ERR,
4290                          "Invalid log block size: %u",
4291                          le32_to_cpu(es->s_log_block_size));
4292                 goto failed_mount;
4293         }
4294         if (le32_to_cpu(es->s_log_cluster_size) >
4295             (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4296                 ext4_msg(sb, KERN_ERR,
4297                          "Invalid log cluster size: %u",
4298                          le32_to_cpu(es->s_log_cluster_size));
4299                 goto failed_mount;
4300         }
4301
4302         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
4303                 ext4_msg(sb, KERN_ERR,
4304                          "Number of reserved GDT blocks insanely large: %d",
4305                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
4306                 goto failed_mount;
4307         }
4308
4309         if (bdev_dax_supported(sb->s_bdev, blocksize))
4310                 set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags);
4311
4312         if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) {
4313                 if (ext4_has_feature_inline_data(sb)) {
4314                         ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
4315                                         " that may contain inline data");
4316                         goto failed_mount;
4317                 }
4318                 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) {
4319                         ext4_msg(sb, KERN_ERR,
4320                                 "DAX unsupported by block device.");
4321                         goto failed_mount;
4322                 }
4323         }
4324
4325         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
4326                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
4327                          es->s_encryption_level);
4328                 goto failed_mount;
4329         }
4330
4331         if (sb->s_blocksize != blocksize) {
4332                 /* Validate the filesystem blocksize */
4333                 if (!sb_set_blocksize(sb, blocksize)) {
4334                         ext4_msg(sb, KERN_ERR, "bad block size %d",
4335                                         blocksize);
4336                         goto failed_mount;
4337                 }
4338
4339                 brelse(bh);
4340                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
4341                 offset = do_div(logical_sb_block, blocksize);
4342                 bh = sb_bread_unmovable(sb, logical_sb_block);
4343                 if (!bh) {
4344                         ext4_msg(sb, KERN_ERR,
4345                                "Can't read superblock on 2nd try");
4346                         goto failed_mount;
4347                 }
4348                 es = (struct ext4_super_block *)(bh->b_data + offset);
4349                 sbi->s_es = es;
4350                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
4351                         ext4_msg(sb, KERN_ERR,
4352                                "Magic mismatch, very weird!");
4353                         goto failed_mount;
4354                 }
4355         }
4356
4357         has_huge_files = ext4_has_feature_huge_file(sb);
4358         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
4359                                                       has_huge_files);
4360         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
4361
4362         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
4363         if (ext4_has_feature_64bit(sb)) {
4364                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
4365                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
4366                     !is_power_of_2(sbi->s_desc_size)) {
4367                         ext4_msg(sb, KERN_ERR,
4368                                "unsupported descriptor size %lu",
4369                                sbi->s_desc_size);
4370                         goto failed_mount;
4371                 }
4372         } else
4373                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
4374
4375         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
4376         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
4377
4378         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
4379         if (sbi->s_inodes_per_block == 0)
4380                 goto cantfind_ext4;
4381         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
4382             sbi->s_inodes_per_group > blocksize * 8) {
4383                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
4384                          sbi->s_inodes_per_group);
4385                 goto failed_mount;
4386         }
4387         sbi->s_itb_per_group = sbi->s_inodes_per_group /
4388                                         sbi->s_inodes_per_block;
4389         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
4390         sbi->s_sbh = bh;
4391         sbi->s_mount_state = le16_to_cpu(es->s_state);
4392         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
4393         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
4394
4395         for (i = 0; i < 4; i++)
4396                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
4397         sbi->s_def_hash_version = es->s_def_hash_version;
4398         if (ext4_has_feature_dir_index(sb)) {
4399                 i = le32_to_cpu(es->s_flags);
4400                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
4401                         sbi->s_hash_unsigned = 3;
4402                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
4403 #ifdef __CHAR_UNSIGNED__
4404                         if (!sb_rdonly(sb))
4405                                 es->s_flags |=
4406                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
4407                         sbi->s_hash_unsigned = 3;
4408 #else
4409                         if (!sb_rdonly(sb))
4410                                 es->s_flags |=
4411                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
4412 #endif
4413                 }
4414         }
4415
4416         /* Handle clustersize */
4417         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
4418         if (ext4_has_feature_bigalloc(sb)) {
4419                 if (clustersize < blocksize) {
4420                         ext4_msg(sb, KERN_ERR,
4421                                  "cluster size (%d) smaller than "
4422                                  "block size (%d)", clustersize, blocksize);
4423                         goto failed_mount;
4424                 }
4425                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
4426                         le32_to_cpu(es->s_log_block_size);
4427                 sbi->s_clusters_per_group =
4428                         le32_to_cpu(es->s_clusters_per_group);
4429                 if (sbi->s_clusters_per_group > blocksize * 8) {
4430                         ext4_msg(sb, KERN_ERR,
4431                                  "#clusters per group too big: %lu",
4432                                  sbi->s_clusters_per_group);
4433                         goto failed_mount;
4434                 }
4435                 if (sbi->s_blocks_per_group !=
4436                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
4437                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
4438                                  "clusters per group (%lu) inconsistent",
4439                                  sbi->s_blocks_per_group,
4440                                  sbi->s_clusters_per_group);
4441                         goto failed_mount;
4442                 }
4443         } else {
4444                 if (clustersize != blocksize) {
4445                         ext4_msg(sb, KERN_ERR,
4446                                  "fragment/cluster size (%d) != "
4447                                  "block size (%d)", clustersize, blocksize);
4448                         goto failed_mount;
4449                 }
4450                 if (sbi->s_blocks_per_group > blocksize * 8) {
4451                         ext4_msg(sb, KERN_ERR,
4452                                  "#blocks per group too big: %lu",
4453                                  sbi->s_blocks_per_group);
4454                         goto failed_mount;
4455                 }
4456                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
4457                 sbi->s_cluster_bits = 0;
4458         }
4459         sbi->s_cluster_ratio = clustersize / blocksize;
4460
4461         /* Do we have standard group size of clustersize * 8 blocks ? */
4462         if (sbi->s_blocks_per_group == clustersize << 3)
4463                 set_opt2(sb, STD_GROUP_SIZE);
4464
4465         /*
4466          * Test whether we have more sectors than will fit in sector_t,
4467          * and whether the max offset is addressable by the page cache.
4468          */
4469         err = generic_check_addressable(sb->s_blocksize_bits,
4470                                         ext4_blocks_count(es));
4471         if (err) {
4472                 ext4_msg(sb, KERN_ERR, "filesystem"
4473                          " too large to mount safely on this system");
4474                 goto failed_mount;
4475         }
4476
4477         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
4478                 goto cantfind_ext4;
4479
4480         /* check blocks count against device size */
4481         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
4482         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
4483                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
4484                        "exceeds size of device (%llu blocks)",
4485                        ext4_blocks_count(es), blocks_count);
4486                 goto failed_mount;
4487         }
4488
4489         /*
4490          * It makes no sense for the first data block to be beyond the end
4491          * of the filesystem.
4492          */
4493         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
4494                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4495                          "block %u is beyond end of filesystem (%llu)",
4496                          le32_to_cpu(es->s_first_data_block),
4497                          ext4_blocks_count(es));
4498                 goto failed_mount;
4499         }
4500         if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) &&
4501             (sbi->s_cluster_ratio == 1)) {
4502                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4503                          "block is 0 with a 1k block and cluster size");
4504                 goto failed_mount;
4505         }
4506
4507         blocks_count = (ext4_blocks_count(es) -
4508                         le32_to_cpu(es->s_first_data_block) +
4509                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
4510         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
4511         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
4512                 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu "
4513                        "(block count %llu, first data block %u, "
4514                        "blocks per group %lu)", blocks_count,
4515                        ext4_blocks_count(es),
4516                        le32_to_cpu(es->s_first_data_block),
4517                        EXT4_BLOCKS_PER_GROUP(sb));
4518                 goto failed_mount;
4519         }
4520         sbi->s_groups_count = blocks_count;
4521         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
4522                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
4523         if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) !=
4524             le32_to_cpu(es->s_inodes_count)) {
4525                 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu",
4526                          le32_to_cpu(es->s_inodes_count),
4527                          ((u64)sbi->s_groups_count * sbi->s_inodes_per_group));
4528                 ret = -EINVAL;
4529                 goto failed_mount;
4530         }
4531         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
4532                    EXT4_DESC_PER_BLOCK(sb);
4533         if (ext4_has_feature_meta_bg(sb)) {
4534                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
4535                         ext4_msg(sb, KERN_WARNING,
4536                                  "first meta block group too large: %u "
4537                                  "(group descriptor block count %u)",
4538                                  le32_to_cpu(es->s_first_meta_bg), db_count);
4539                         goto failed_mount;
4540                 }
4541         }
4542         rcu_assign_pointer(sbi->s_group_desc,
4543                            kvmalloc_array(db_count,
4544                                           sizeof(struct buffer_head *),
4545                                           GFP_KERNEL));
4546         if (sbi->s_group_desc == NULL) {
4547                 ext4_msg(sb, KERN_ERR, "not enough memory");
4548                 ret = -ENOMEM;
4549                 goto failed_mount;
4550         }
4551
4552         bgl_lock_init(sbi->s_blockgroup_lock);
4553
4554         /* Pre-read the descriptors into the buffer cache */
4555         for (i = 0; i < db_count; i++) {
4556                 block = descriptor_loc(sb, logical_sb_block, i);
4557                 sb_breadahead_unmovable(sb, block);
4558         }
4559
4560         for (i = 0; i < db_count; i++) {
4561                 struct buffer_head *bh;
4562
4563                 block = descriptor_loc(sb, logical_sb_block, i);
4564                 bh = sb_bread_unmovable(sb, block);
4565                 if (!bh) {
4566                         ext4_msg(sb, KERN_ERR,
4567                                "can't read group descriptor %d", i);
4568                         db_count = i;
4569                         goto failed_mount2;
4570                 }
4571                 rcu_read_lock();
4572                 rcu_dereference(sbi->s_group_desc)[i] = bh;
4573                 rcu_read_unlock();
4574         }
4575         sbi->s_gdb_count = db_count;
4576         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
4577                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
4578                 ret = -EFSCORRUPTED;
4579                 goto failed_mount2;
4580         }
4581
4582         timer_setup(&sbi->s_err_report, print_daily_error_info, 0);
4583
4584         /* Register extent status tree shrinker */
4585         if (ext4_es_register_shrinker(sbi))
4586                 goto failed_mount3;
4587
4588         sbi->s_stripe = ext4_get_stripe_size(sbi);
4589         sbi->s_extent_max_zeroout_kb = 32;
4590
4591         /*
4592          * set up enough so that it can read an inode
4593          */
4594         sb->s_op = &ext4_sops;
4595         sb->s_export_op = &ext4_export_ops;
4596         sb->s_xattr = ext4_xattr_handlers;
4597 #ifdef CONFIG_FS_ENCRYPTION
4598         sb->s_cop = &ext4_cryptops;
4599 #endif
4600 #ifdef CONFIG_FS_VERITY
4601         sb->s_vop = &ext4_verityops;
4602 #endif
4603 #ifdef CONFIG_QUOTA
4604         sb->dq_op = &ext4_quota_operations;
4605         if (ext4_has_feature_quota(sb))
4606                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4607         else
4608                 sb->s_qcop = &ext4_qctl_operations;
4609         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
4610 #endif
4611         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
4612
4613         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
4614         mutex_init(&sbi->s_orphan_lock);
4615
4616         sb->s_root = NULL;
4617
4618         needs_recovery = (es->s_last_orphan != 0 ||
4619                           ext4_has_feature_journal_needs_recovery(sb));
4620
4621         if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb))
4622                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
4623                         goto failed_mount3a;
4624
4625         /*
4626          * The first inode we look at is the journal inode.  Don't try
4627          * root first: it may be modified in the journal!
4628          */
4629         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
4630                 err = ext4_load_journal(sb, es, journal_devnum);
4631                 if (err)
4632                         goto failed_mount3a;
4633         } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
4634                    ext4_has_feature_journal_needs_recovery(sb)) {
4635                 ext4_msg(sb, KERN_ERR, "required journal recovery "
4636                        "suppressed and not mounted read-only");
4637                 goto failed_mount_wq;
4638         } else {
4639                 /* Nojournal mode, all journal mount options are illegal */
4640                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
4641                         ext4_msg(sb, KERN_ERR, "can't mount with "
4642                                  "journal_checksum, fs mounted w/o journal");
4643                         goto failed_mount_wq;
4644                 }
4645                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4646                         ext4_msg(sb, KERN_ERR, "can't mount with "
4647                                  "journal_async_commit, fs mounted w/o journal");
4648                         goto failed_mount_wq;
4649                 }
4650                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
4651                         ext4_msg(sb, KERN_ERR, "can't mount with "
4652                                  "commit=%lu, fs mounted w/o journal",
4653                                  sbi->s_commit_interval / HZ);
4654                         goto failed_mount_wq;
4655                 }
4656                 if (EXT4_MOUNT_DATA_FLAGS &
4657                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
4658                         ext4_msg(sb, KERN_ERR, "can't mount with "
4659                                  "data=, fs mounted w/o journal");
4660                         goto failed_mount_wq;
4661                 }
4662                 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM;
4663                 clear_opt(sb, JOURNAL_CHECKSUM);
4664                 clear_opt(sb, DATA_FLAGS);
4665                 sbi->s_journal = NULL;
4666                 needs_recovery = 0;
4667                 goto no_journal;
4668         }
4669
4670         if (ext4_has_feature_64bit(sb) &&
4671             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4672                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4673                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4674                 goto failed_mount_wq;
4675         }
4676
4677         if (!set_journal_csum_feature_set(sb)) {
4678                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4679                          "feature set");
4680                 goto failed_mount_wq;
4681         }
4682
4683         /* We have now updated the journal if required, so we can
4684          * validate the data journaling mode. */
4685         switch (test_opt(sb, DATA_FLAGS)) {
4686         case 0:
4687                 /* No mode set, assume a default based on the journal
4688                  * capabilities: ORDERED_DATA if the journal can
4689                  * cope, else JOURNAL_DATA
4690                  */
4691                 if (jbd2_journal_check_available_features
4692                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4693                         set_opt(sb, ORDERED_DATA);
4694                         sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
4695                 } else {
4696                         set_opt(sb, JOURNAL_DATA);
4697                         sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
4698                 }
4699                 break;
4700
4701         case EXT4_MOUNT_ORDERED_DATA:
4702         case EXT4_MOUNT_WRITEBACK_DATA:
4703                 if (!jbd2_journal_check_available_features
4704                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4705                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4706                                "requested data journaling mode");
4707                         goto failed_mount_wq;
4708                 }
4709         default:
4710                 break;
4711         }
4712
4713         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4714             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4715                 ext4_msg(sb, KERN_ERR, "can't mount with "
4716                         "journal_async_commit in data=ordered mode");
4717                 goto failed_mount_wq;
4718         }
4719
4720         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4721
4722         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4723
4724 no_journal:
4725         if (!test_opt(sb, NO_MBCACHE)) {
4726                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
4727                 if (!sbi->s_ea_block_cache) {
4728                         ext4_msg(sb, KERN_ERR,
4729                                  "Failed to create ea_block_cache");
4730                         goto failed_mount_wq;
4731                 }
4732
4733                 if (ext4_has_feature_ea_inode(sb)) {
4734                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
4735                         if (!sbi->s_ea_inode_cache) {
4736                                 ext4_msg(sb, KERN_ERR,
4737                                          "Failed to create ea_inode_cache");
4738                                 goto failed_mount_wq;
4739                         }
4740                 }
4741         }
4742
4743         if (ext4_has_feature_verity(sb) && blocksize != PAGE_SIZE) {
4744                 ext4_msg(sb, KERN_ERR, "Unsupported blocksize for fs-verity");
4745                 goto failed_mount_wq;
4746         }
4747
4748         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) &&
4749             !ext4_has_feature_encrypt(sb)) {
4750                 ext4_set_feature_encrypt(sb);
4751                 ext4_commit_super(sb, 1);
4752         }
4753
4754         /*
4755          * Get the # of file system overhead blocks from the
4756          * superblock if present.
4757          */
4758         if (es->s_overhead_clusters)
4759                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4760         else {
4761                 err = ext4_calculate_overhead(sb);
4762                 if (err)
4763                         goto failed_mount_wq;
4764         }
4765
4766         /*
4767          * The maximum number of concurrent works can be high and
4768          * concurrency isn't really necessary.  Limit it to 1.
4769          */
4770         EXT4_SB(sb)->rsv_conversion_wq =
4771                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4772         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4773                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4774                 ret = -ENOMEM;
4775                 goto failed_mount4;
4776         }
4777
4778         /*
4779          * The jbd2_journal_load will have done any necessary log recovery,
4780          * so we can safely mount the rest of the filesystem now.
4781          */
4782
4783         root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL);
4784         if (IS_ERR(root)) {
4785                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4786                 ret = PTR_ERR(root);
4787                 root = NULL;
4788                 goto failed_mount4;
4789         }
4790         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4791                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4792                 iput(root);
4793                 goto failed_mount4;
4794         }
4795
4796 #ifdef CONFIG_UNICODE
4797         if (sbi->s_encoding)
4798                 sb->s_d_op = &ext4_dentry_ops;
4799 #endif
4800
4801         sb->s_root = d_make_root(root);
4802         if (!sb->s_root) {
4803                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4804                 ret = -ENOMEM;
4805                 goto failed_mount4;
4806         }
4807
4808         ret = ext4_setup_super(sb, es, sb_rdonly(sb));
4809         if (ret == -EROFS) {
4810                 sb->s_flags |= SB_RDONLY;
4811                 ret = 0;
4812         } else if (ret)
4813                 goto failed_mount4a;
4814
4815         ext4_set_resv_clusters(sb);
4816
4817         if (test_opt(sb, BLOCK_VALIDITY)) {
4818                 err = ext4_setup_system_zone(sb);
4819                 if (err) {
4820                         ext4_msg(sb, KERN_ERR, "failed to initialize system "
4821                                  "zone (%d)", err);
4822                         goto failed_mount4a;
4823                 }
4824         }
4825
4826         ext4_ext_init(sb);
4827         err = ext4_mb_init(sb);
4828         if (err) {
4829                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4830                          err);
4831                 goto failed_mount5;
4832         }
4833
4834         block = ext4_count_free_clusters(sb);
4835         ext4_free_blocks_count_set(sbi->s_es,
4836                                    EXT4_C2B(sbi, block));
4837         ext4_superblock_csum_set(sb);
4838         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4839                                   GFP_KERNEL);
4840         if (!err) {
4841                 unsigned long freei = ext4_count_free_inodes(sb);
4842                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4843                 ext4_superblock_csum_set(sb);
4844                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4845                                           GFP_KERNEL);
4846         }
4847         if (!err)
4848                 err = percpu_counter_init(&sbi->s_dirs_counter,
4849                                           ext4_count_dirs(sb), GFP_KERNEL);
4850         if (!err)
4851                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4852                                           GFP_KERNEL);
4853         if (!err)
4854                 err = percpu_init_rwsem(&sbi->s_writepages_rwsem);
4855
4856         if (err) {
4857                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4858                 goto failed_mount6;
4859         }
4860
4861         if (ext4_has_feature_flex_bg(sb))
4862                 if (!ext4_fill_flex_info(sb)) {
4863                         ext4_msg(sb, KERN_ERR,
4864                                "unable to initialize "
4865                                "flex_bg meta info!");
4866                         goto failed_mount6;
4867                 }
4868
4869         err = ext4_register_li_request(sb, first_not_zeroed);
4870         if (err)
4871                 goto failed_mount6;
4872
4873         err = ext4_register_sysfs(sb);
4874         if (err)
4875                 goto failed_mount7;
4876
4877 #ifdef CONFIG_QUOTA
4878         /* Enable quota usage during mount. */
4879         if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
4880                 err = ext4_enable_quotas(sb);
4881                 if (err)
4882                         goto failed_mount8;
4883         }
4884 #endif  /* CONFIG_QUOTA */
4885
4886         /*
4887          * Save the original bdev mapping's wb_err value which could be
4888          * used to detect the metadata async write error.
4889          */
4890         spin_lock_init(&sbi->s_bdev_wb_lock);
4891         errseq_check_and_advance(&sb->s_bdev->bd_inode->i_mapping->wb_err,
4892                                  &sbi->s_bdev_wb_err);
4893         sb->s_bdev->bd_super = sb;
4894         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4895         ext4_orphan_cleanup(sb, es);
4896         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4897         if (needs_recovery) {
4898                 ext4_msg(sb, KERN_INFO, "recovery complete");
4899                 err = ext4_mark_recovery_complete(sb, es);
4900                 if (err)
4901                         goto failed_mount8;
4902         }
4903         if (EXT4_SB(sb)->s_journal) {
4904                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4905                         descr = " journalled data mode";
4906                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4907                         descr = " ordered data mode";
4908                 else
4909                         descr = " writeback data mode";
4910         } else
4911                 descr = "out journal";
4912
4913         if (test_opt(sb, DISCARD)) {
4914                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4915                 if (!blk_queue_discard(q))
4916                         ext4_msg(sb, KERN_WARNING,
4917                                  "mounting with \"discard\" option, but "
4918                                  "the device does not support discard");
4919         }
4920
4921         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4922                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4923                          "Opts: %.*s%s%s", descr,
4924                          (int) sizeof(sbi->s_es->s_mount_opts),
4925                          sbi->s_es->s_mount_opts,
4926                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4927
4928         if (es->s_error_count)
4929                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4930
4931         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4932         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4933         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4934         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4935         atomic_set(&sbi->s_warning_count, 0);
4936         atomic_set(&sbi->s_msg_count, 0);
4937
4938         kfree(orig_data);
4939         return 0;
4940
4941 cantfind_ext4:
4942         if (!silent)
4943                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4944         goto failed_mount;
4945
4946 failed_mount8:
4947         ext4_unregister_sysfs(sb);
4948         kobject_put(&sbi->s_kobj);
4949 failed_mount7:
4950         ext4_unregister_li_request(sb);
4951 failed_mount6:
4952         ext4_mb_release(sb);
4953         rcu_read_lock();
4954         flex_groups = rcu_dereference(sbi->s_flex_groups);
4955         if (flex_groups) {
4956                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
4957                         kvfree(flex_groups[i]);
4958                 kvfree(flex_groups);
4959         }
4960         rcu_read_unlock();
4961         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4962         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4963         percpu_counter_destroy(&sbi->s_dirs_counter);
4964         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4965         percpu_free_rwsem(&sbi->s_writepages_rwsem);
4966 failed_mount5:
4967         ext4_ext_release(sb);
4968         ext4_release_system_zone(sb);
4969 failed_mount4a:
4970         dput(sb->s_root);
4971         sb->s_root = NULL;
4972 failed_mount4:
4973         ext4_msg(sb, KERN_ERR, "mount failed");
4974         if (EXT4_SB(sb)->rsv_conversion_wq)
4975                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4976 failed_mount_wq:
4977         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
4978         sbi->s_ea_inode_cache = NULL;
4979
4980         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
4981         sbi->s_ea_block_cache = NULL;
4982
4983         if (sbi->s_journal) {
4984                 jbd2_journal_destroy(sbi->s_journal);
4985                 sbi->s_journal = NULL;
4986         }
4987 failed_mount3a:
4988         ext4_es_unregister_shrinker(sbi);
4989 failed_mount3:
4990         del_timer_sync(&sbi->s_err_report);
4991         if (sbi->s_mmp_tsk)
4992                 kthread_stop(sbi->s_mmp_tsk);
4993 failed_mount2:
4994         rcu_read_lock();
4995         group_desc = rcu_dereference(sbi->s_group_desc);
4996         for (i = 0; i < db_count; i++)
4997                 brelse(group_desc[i]);
4998         kvfree(group_desc);
4999         rcu_read_unlock();
5000 failed_mount:
5001         if (sbi->s_chksum_driver)
5002                 crypto_free_shash(sbi->s_chksum_driver);
5003
5004 #ifdef CONFIG_UNICODE
5005         utf8_unload(sbi->s_encoding);
5006 #endif
5007
5008 #ifdef CONFIG_QUOTA
5009         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5010                 kfree(get_qf_name(sb, sbi, i));
5011 #endif
5012         fscrypt_free_dummy_context(&sbi->s_dummy_enc_ctx);
5013         ext4_blkdev_remove(sbi);
5014         brelse(bh);
5015 out_fail:
5016         sb->s_fs_info = NULL;
5017         kfree(sbi->s_blockgroup_lock);
5018 out_free_base:
5019         kfree(sbi);
5020         kfree(orig_data);
5021         fs_put_dax(dax_dev);
5022         return err ? err : ret;
5023 }
5024
5025 /*
5026  * Setup any per-fs journal parameters now.  We'll do this both on
5027  * initial mount, once the journal has been initialised but before we've
5028  * done any recovery; and again on any subsequent remount.
5029  */
5030 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
5031 {
5032         struct ext4_sb_info *sbi = EXT4_SB(sb);
5033
5034         journal->j_commit_interval = sbi->s_commit_interval;
5035         journal->j_min_batch_time = sbi->s_min_batch_time;
5036         journal->j_max_batch_time = sbi->s_max_batch_time;
5037
5038         write_lock(&journal->j_state_lock);
5039         if (test_opt(sb, BARRIER))
5040                 journal->j_flags |= JBD2_BARRIER;
5041         else
5042                 journal->j_flags &= ~JBD2_BARRIER;
5043         if (test_opt(sb, DATA_ERR_ABORT))
5044                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
5045         else
5046                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
5047         write_unlock(&journal->j_state_lock);
5048 }
5049
5050 static struct inode *ext4_get_journal_inode(struct super_block *sb,
5051                                              unsigned int journal_inum)
5052 {
5053         struct inode *journal_inode;
5054
5055         /*
5056          * Test for the existence of a valid inode on disk.  Bad things
5057          * happen if we iget() an unused inode, as the subsequent iput()
5058          * will try to delete it.
5059          */
5060         journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL);
5061         if (IS_ERR(journal_inode)) {
5062                 ext4_msg(sb, KERN_ERR, "no journal found");
5063                 return NULL;
5064         }
5065         if (!journal_inode->i_nlink) {
5066                 make_bad_inode(journal_inode);
5067                 iput(journal_inode);
5068                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
5069                 return NULL;
5070         }
5071
5072         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
5073                   journal_inode, journal_inode->i_size);
5074         if (!S_ISREG(journal_inode->i_mode)) {
5075                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
5076                 iput(journal_inode);
5077                 return NULL;
5078         }
5079         return journal_inode;
5080 }
5081
5082 static journal_t *ext4_get_journal(struct super_block *sb,
5083                                    unsigned int journal_inum)
5084 {
5085         struct inode *journal_inode;
5086         journal_t *journal;
5087
5088         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5089                 return NULL;
5090
5091         journal_inode = ext4_get_journal_inode(sb, journal_inum);
5092         if (!journal_inode)
5093                 return NULL;
5094
5095         journal = jbd2_journal_init_inode(journal_inode);
5096         if (!journal) {
5097                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
5098                 iput(journal_inode);
5099                 return NULL;
5100         }
5101         journal->j_private = sb;
5102         ext4_init_journal_params(sb, journal);
5103         return journal;
5104 }
5105
5106 static journal_t *ext4_get_dev_journal(struct super_block *sb,
5107                                        dev_t j_dev)
5108 {
5109         struct buffer_head *bh;
5110         journal_t *journal;
5111         ext4_fsblk_t start;
5112         ext4_fsblk_t len;
5113         int hblock, blocksize;
5114         ext4_fsblk_t sb_block;
5115         unsigned long offset;
5116         struct ext4_super_block *es;
5117         struct block_device *bdev;
5118
5119         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5120                 return NULL;
5121
5122         bdev = ext4_blkdev_get(j_dev, sb);
5123         if (bdev == NULL)
5124                 return NULL;
5125
5126         blocksize = sb->s_blocksize;
5127         hblock = bdev_logical_block_size(bdev);
5128         if (blocksize < hblock) {
5129                 ext4_msg(sb, KERN_ERR,
5130                         "blocksize too small for journal device");
5131                 goto out_bdev;
5132         }
5133
5134         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
5135         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
5136         set_blocksize(bdev, blocksize);
5137         if (!(bh = __bread(bdev, sb_block, blocksize))) {
5138                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
5139                        "external journal");
5140                 goto out_bdev;
5141         }
5142
5143         es = (struct ext4_super_block *) (bh->b_data + offset);
5144         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
5145             !(le32_to_cpu(es->s_feature_incompat) &
5146               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
5147                 ext4_msg(sb, KERN_ERR, "external journal has "
5148                                         "bad superblock");
5149                 brelse(bh);
5150                 goto out_bdev;
5151         }
5152
5153         if ((le32_to_cpu(es->s_feature_ro_compat) &
5154              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
5155             es->s_checksum != ext4_superblock_csum(sb, es)) {
5156                 ext4_msg(sb, KERN_ERR, "external journal has "
5157                                        "corrupt superblock");
5158                 brelse(bh);
5159                 goto out_bdev;
5160         }
5161
5162         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
5163                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
5164                 brelse(bh);
5165                 goto out_bdev;
5166         }
5167
5168         len = ext4_blocks_count(es);
5169         start = sb_block + 1;
5170         brelse(bh);     /* we're done with the superblock */
5171
5172         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
5173                                         start, len, blocksize);
5174         if (!journal) {
5175                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
5176                 goto out_bdev;
5177         }
5178         journal->j_private = sb;
5179         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
5180         wait_on_buffer(journal->j_sb_buffer);
5181         if (!buffer_uptodate(journal->j_sb_buffer)) {
5182                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
5183                 goto out_journal;
5184         }
5185         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
5186                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
5187                                         "user (unsupported) - %d",
5188                         be32_to_cpu(journal->j_superblock->s_nr_users));
5189                 goto out_journal;
5190         }
5191         EXT4_SB(sb)->s_journal_bdev = bdev;
5192         ext4_init_journal_params(sb, journal);
5193         return journal;
5194
5195 out_journal:
5196         jbd2_journal_destroy(journal);
5197 out_bdev:
5198         ext4_blkdev_put(bdev);
5199         return NULL;
5200 }
5201
5202 static int ext4_load_journal(struct super_block *sb,
5203                              struct ext4_super_block *es,
5204                              unsigned long journal_devnum)
5205 {
5206         journal_t *journal;
5207         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
5208         dev_t journal_dev;
5209         int err = 0;
5210         int really_read_only;
5211         int journal_dev_ro;
5212
5213         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5214                 return -EFSCORRUPTED;
5215
5216         if (journal_devnum &&
5217             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5218                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
5219                         "numbers have changed");
5220                 journal_dev = new_decode_dev(journal_devnum);
5221         } else
5222                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
5223
5224         if (journal_inum && journal_dev) {
5225                 ext4_msg(sb, KERN_ERR,
5226                          "filesystem has both journal inode and journal device!");
5227                 return -EINVAL;
5228         }
5229
5230         if (journal_inum) {
5231                 journal = ext4_get_journal(sb, journal_inum);
5232                 if (!journal)
5233                         return -EINVAL;
5234         } else {
5235                 journal = ext4_get_dev_journal(sb, journal_dev);
5236                 if (!journal)
5237                         return -EINVAL;
5238         }
5239
5240         journal_dev_ro = bdev_read_only(journal->j_dev);
5241         really_read_only = bdev_read_only(sb->s_bdev) | journal_dev_ro;
5242
5243         if (journal_dev_ro && !sb_rdonly(sb)) {
5244                 ext4_msg(sb, KERN_ERR,
5245                          "journal device read-only, try mounting with '-o ro'");
5246                 err = -EROFS;
5247                 goto err_out;
5248         }
5249
5250         /*
5251          * Are we loading a blank journal or performing recovery after a
5252          * crash?  For recovery, we need to check in advance whether we
5253          * can get read-write access to the device.
5254          */
5255         if (ext4_has_feature_journal_needs_recovery(sb)) {
5256                 if (sb_rdonly(sb)) {
5257                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
5258                                         "required on readonly filesystem");
5259                         if (really_read_only) {
5260                                 ext4_msg(sb, KERN_ERR, "write access "
5261                                         "unavailable, cannot proceed "
5262                                         "(try mounting with noload)");
5263                                 err = -EROFS;
5264                                 goto err_out;
5265                         }
5266                         ext4_msg(sb, KERN_INFO, "write access will "
5267                                "be enabled during recovery");
5268                 }
5269         }
5270
5271         if (!(journal->j_flags & JBD2_BARRIER))
5272                 ext4_msg(sb, KERN_INFO, "barriers disabled");
5273
5274         if (!ext4_has_feature_journal_needs_recovery(sb))
5275                 err = jbd2_journal_wipe(journal, !really_read_only);
5276         if (!err) {
5277                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
5278                 if (save)
5279                         memcpy(save, ((char *) es) +
5280                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
5281                 err = jbd2_journal_load(journal);
5282                 if (save)
5283                         memcpy(((char *) es) + EXT4_S_ERR_START,
5284                                save, EXT4_S_ERR_LEN);
5285                 kfree(save);
5286         }
5287
5288         if (err) {
5289                 ext4_msg(sb, KERN_ERR, "error loading journal");
5290                 goto err_out;
5291         }
5292
5293         EXT4_SB(sb)->s_journal = journal;
5294         err = ext4_clear_journal_err(sb, es);
5295         if (err) {
5296                 EXT4_SB(sb)->s_journal = NULL;
5297                 jbd2_journal_destroy(journal);
5298                 return err;
5299         }
5300
5301         if (!really_read_only && journal_devnum &&
5302             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5303                 es->s_journal_dev = cpu_to_le32(journal_devnum);
5304
5305                 /* Make sure we flush the recovery flag to disk. */
5306                 ext4_commit_super(sb, 1);
5307         }
5308
5309         return 0;
5310
5311 err_out:
5312         jbd2_journal_destroy(journal);
5313         return err;
5314 }
5315
5316 static int ext4_commit_super(struct super_block *sb, int sync)
5317 {
5318         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
5319         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
5320         int error = 0;
5321
5322         if (!sbh || block_device_ejected(sb))
5323                 return error;
5324
5325         /*
5326          * If the file system is mounted read-only, don't update the
5327          * superblock write time.  This avoids updating the superblock
5328          * write time when we are mounting the root file system
5329          * read/only but we need to replay the journal; at that point,
5330          * for people who are east of GMT and who make their clock
5331          * tick in localtime for Windows bug-for-bug compatibility,
5332          * the clock is set in the future, and this will cause e2fsck
5333          * to complain and force a full file system check.
5334          */
5335         if (!(sb->s_flags & SB_RDONLY))
5336                 ext4_update_tstamp(es, s_wtime);
5337         if (sb->s_bdev->bd_part)
5338                 es->s_kbytes_written =
5339                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
5340                             ((part_stat_read(sb->s_bdev->bd_part,
5341                                              sectors[STAT_WRITE]) -
5342                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
5343         else
5344                 es->s_kbytes_written =
5345                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
5346         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
5347                 ext4_free_blocks_count_set(es,
5348                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
5349                                 &EXT4_SB(sb)->s_freeclusters_counter)));
5350         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
5351                 es->s_free_inodes_count =
5352                         cpu_to_le32(percpu_counter_sum_positive(
5353                                 &EXT4_SB(sb)->s_freeinodes_counter));
5354         BUFFER_TRACE(sbh, "marking dirty");
5355         ext4_superblock_csum_set(sb);
5356         if (sync)
5357                 lock_buffer(sbh);
5358         if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
5359                 /*
5360                  * Oh, dear.  A previous attempt to write the
5361                  * superblock failed.  This could happen because the
5362                  * USB device was yanked out.  Or it could happen to
5363                  * be a transient write error and maybe the block will
5364                  * be remapped.  Nothing we can do but to retry the
5365                  * write and hope for the best.
5366                  */
5367                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
5368                        "superblock detected");
5369                 clear_buffer_write_io_error(sbh);
5370                 set_buffer_uptodate(sbh);
5371         }
5372         mark_buffer_dirty(sbh);
5373         if (sync) {
5374                 unlock_buffer(sbh);
5375                 error = __sync_dirty_buffer(sbh,
5376                         REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
5377                 if (buffer_write_io_error(sbh)) {
5378                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
5379                                "superblock");
5380                         clear_buffer_write_io_error(sbh);
5381                         set_buffer_uptodate(sbh);
5382                 }
5383         }
5384         return error;
5385 }
5386
5387 /*
5388  * Have we just finished recovery?  If so, and if we are mounting (or
5389  * remounting) the filesystem readonly, then we will end up with a
5390  * consistent fs on disk.  Record that fact.
5391  */
5392 static int ext4_mark_recovery_complete(struct super_block *sb,
5393                                        struct ext4_super_block *es)
5394 {
5395         int err;
5396         journal_t *journal = EXT4_SB(sb)->s_journal;
5397
5398         if (!ext4_has_feature_journal(sb)) {
5399                 if (journal != NULL) {
5400                         ext4_error(sb, "Journal got removed while the fs was "
5401                                    "mounted!");
5402                         return -EFSCORRUPTED;
5403                 }
5404                 return 0;
5405         }
5406         jbd2_journal_lock_updates(journal);
5407         err = jbd2_journal_flush(journal);
5408         if (err < 0)
5409                 goto out;
5410
5411         if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) {
5412                 ext4_clear_feature_journal_needs_recovery(sb);
5413                 ext4_commit_super(sb, 1);
5414         }
5415 out:
5416         jbd2_journal_unlock_updates(journal);
5417         return err;
5418 }
5419
5420 /*
5421  * If we are mounting (or read-write remounting) a filesystem whose journal
5422  * has recorded an error from a previous lifetime, move that error to the
5423  * main filesystem now.
5424  */
5425 static int ext4_clear_journal_err(struct super_block *sb,
5426                                    struct ext4_super_block *es)
5427 {
5428         journal_t *journal;
5429         int j_errno;
5430         const char *errstr;
5431
5432         if (!ext4_has_feature_journal(sb)) {
5433                 ext4_error(sb, "Journal got removed while the fs was mounted!");
5434                 return -EFSCORRUPTED;
5435         }
5436
5437         journal = EXT4_SB(sb)->s_journal;
5438
5439         /*
5440          * Now check for any error status which may have been recorded in the
5441          * journal by a prior ext4_error() or ext4_abort()
5442          */
5443
5444         j_errno = jbd2_journal_errno(journal);
5445         if (j_errno) {
5446                 char nbuf[16];
5447
5448                 errstr = ext4_decode_error(sb, j_errno, nbuf);
5449                 ext4_warning(sb, "Filesystem error recorded "
5450                              "from previous mount: %s", errstr);
5451                 ext4_warning(sb, "Marking fs in need of filesystem check.");
5452
5453                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
5454                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
5455                 ext4_commit_super(sb, 1);
5456
5457                 jbd2_journal_clear_err(journal);
5458                 jbd2_journal_update_sb_errno(journal);
5459         }
5460         return 0;
5461 }
5462
5463 /*
5464  * Force the running and committing transactions to commit,
5465  * and wait on the commit.
5466  */
5467 int ext4_force_commit(struct super_block *sb)
5468 {
5469         journal_t *journal;
5470
5471         if (sb_rdonly(sb))
5472                 return 0;
5473
5474         journal = EXT4_SB(sb)->s_journal;
5475         return ext4_journal_force_commit(journal);
5476 }
5477
5478 static int ext4_sync_fs(struct super_block *sb, int wait)
5479 {
5480         int ret = 0;
5481         tid_t target;
5482         bool needs_barrier = false;
5483         struct ext4_sb_info *sbi = EXT4_SB(sb);
5484
5485         if (unlikely(ext4_forced_shutdown(sbi)))
5486                 return 0;
5487
5488         trace_ext4_sync_fs(sb, wait);
5489         flush_workqueue(sbi->rsv_conversion_wq);
5490         /*
5491          * Writeback quota in non-journalled quota case - journalled quota has
5492          * no dirty dquots
5493          */
5494         dquot_writeback_dquots(sb, -1);
5495         /*
5496          * Data writeback is possible w/o journal transaction, so barrier must
5497          * being sent at the end of the function. But we can skip it if
5498          * transaction_commit will do it for us.
5499          */
5500         if (sbi->s_journal) {
5501                 target = jbd2_get_latest_transaction(sbi->s_journal);
5502                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
5503                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
5504                         needs_barrier = true;
5505
5506                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
5507                         if (wait)
5508                                 ret = jbd2_log_wait_commit(sbi->s_journal,
5509                                                            target);
5510                 }
5511         } else if (wait && test_opt(sb, BARRIER))
5512                 needs_barrier = true;
5513         if (needs_barrier) {
5514                 int err;
5515                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL);
5516                 if (!ret)
5517                         ret = err;
5518         }
5519
5520         return ret;
5521 }
5522
5523 /*
5524  * LVM calls this function before a (read-only) snapshot is created.  This
5525  * gives us a chance to flush the journal completely and mark the fs clean.
5526  *
5527  * Note that only this function cannot bring a filesystem to be in a clean
5528  * state independently. It relies on upper layer to stop all data & metadata
5529  * modifications.
5530  */
5531 static int ext4_freeze(struct super_block *sb)
5532 {
5533         int error = 0;
5534         journal_t *journal;
5535
5536         if (sb_rdonly(sb))
5537                 return 0;
5538
5539         journal = EXT4_SB(sb)->s_journal;
5540
5541         if (journal) {
5542                 /* Now we set up the journal barrier. */
5543                 jbd2_journal_lock_updates(journal);
5544
5545                 /*
5546                  * Don't clear the needs_recovery flag if we failed to
5547                  * flush the journal.
5548                  */
5549                 error = jbd2_journal_flush(journal);
5550                 if (error < 0)
5551                         goto out;
5552
5553                 /* Journal blocked and flushed, clear needs_recovery flag. */
5554                 ext4_clear_feature_journal_needs_recovery(sb);
5555         }
5556
5557         error = ext4_commit_super(sb, 1);
5558 out:
5559         if (journal)
5560                 /* we rely on upper layer to stop further updates */
5561                 jbd2_journal_unlock_updates(journal);
5562         return error;
5563 }
5564
5565 /*
5566  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
5567  * flag here, even though the filesystem is not technically dirty yet.
5568  */
5569 static int ext4_unfreeze(struct super_block *sb)
5570 {
5571         if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
5572                 return 0;
5573
5574         if (EXT4_SB(sb)->s_journal) {
5575                 /* Reset the needs_recovery flag before the fs is unlocked. */
5576                 ext4_set_feature_journal_needs_recovery(sb);
5577         }
5578
5579         ext4_commit_super(sb, 1);
5580         return 0;
5581 }
5582
5583 /*
5584  * Structure to save mount options for ext4_remount's benefit
5585  */
5586 struct ext4_mount_options {
5587         unsigned long s_mount_opt;
5588         unsigned long s_mount_opt2;
5589         kuid_t s_resuid;
5590         kgid_t s_resgid;
5591         unsigned long s_commit_interval;
5592         u32 s_min_batch_time, s_max_batch_time;
5593 #ifdef CONFIG_QUOTA
5594         int s_jquota_fmt;
5595         char *s_qf_names[EXT4_MAXQUOTAS];
5596 #endif
5597 };
5598
5599 static int ext4_remount(struct super_block *sb, int *flags, char *data)
5600 {
5601         struct ext4_super_block *es;
5602         struct ext4_sb_info *sbi = EXT4_SB(sb);
5603         unsigned long old_sb_flags, vfs_flags;
5604         struct ext4_mount_options old_opts;
5605         int enable_quota = 0;
5606         ext4_group_t g;
5607         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
5608         int err = 0;
5609 #ifdef CONFIG_QUOTA
5610         int i, j;
5611         char *to_free[EXT4_MAXQUOTAS];
5612 #endif
5613         char *orig_data = kstrdup(data, GFP_KERNEL);
5614
5615         if (data && !orig_data)
5616                 return -ENOMEM;
5617
5618         /* Store the original options */
5619         old_sb_flags = sb->s_flags;
5620         old_opts.s_mount_opt = sbi->s_mount_opt;
5621         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
5622         old_opts.s_resuid = sbi->s_resuid;
5623         old_opts.s_resgid = sbi->s_resgid;
5624         old_opts.s_commit_interval = sbi->s_commit_interval;
5625         old_opts.s_min_batch_time = sbi->s_min_batch_time;
5626         old_opts.s_max_batch_time = sbi->s_max_batch_time;
5627 #ifdef CONFIG_QUOTA
5628         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
5629         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5630                 if (sbi->s_qf_names[i]) {
5631                         char *qf_name = get_qf_name(sb, sbi, i);
5632
5633                         old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL);
5634                         if (!old_opts.s_qf_names[i]) {
5635                                 for (j = 0; j < i; j++)
5636                                         kfree(old_opts.s_qf_names[j]);
5637                                 kfree(orig_data);
5638                                 return -ENOMEM;
5639                         }
5640                 } else
5641                         old_opts.s_qf_names[i] = NULL;
5642 #endif
5643         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
5644                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
5645
5646         /*
5647          * Some options can be enabled by ext4 and/or by VFS mount flag
5648          * either way we need to make sure it matches in both *flags and
5649          * s_flags. Copy those selected flags from *flags to s_flags
5650          */
5651         vfs_flags = SB_LAZYTIME | SB_I_VERSION;
5652         sb->s_flags = (sb->s_flags & ~vfs_flags) | (*flags & vfs_flags);
5653
5654         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
5655                 err = -EINVAL;
5656                 goto restore_opts;
5657         }
5658
5659         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
5660             test_opt(sb, JOURNAL_CHECKSUM)) {
5661                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
5662                          "during remount not supported; ignoring");
5663                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
5664         }
5665
5666         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
5667                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
5668                         ext4_msg(sb, KERN_ERR, "can't mount with "
5669                                  "both data=journal and delalloc");
5670                         err = -EINVAL;
5671                         goto restore_opts;
5672                 }
5673                 if (test_opt(sb, DIOREAD_NOLOCK)) {
5674                         ext4_msg(sb, KERN_ERR, "can't mount with "
5675                                  "both data=journal and dioread_nolock");
5676                         err = -EINVAL;
5677                         goto restore_opts;
5678                 }
5679         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
5680                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5681                         ext4_msg(sb, KERN_ERR, "can't mount with "
5682                                 "journal_async_commit in data=ordered mode");
5683                         err = -EINVAL;
5684                         goto restore_opts;
5685                 }
5686         }
5687
5688         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
5689                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
5690                 err = -EINVAL;
5691                 goto restore_opts;
5692         }
5693
5694         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
5695                 ext4_abort(sb, EXT4_ERR_ESHUTDOWN, "Abort forced by user");
5696
5697         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
5698                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
5699
5700         es = sbi->s_es;
5701
5702         if (sbi->s_journal) {
5703                 ext4_init_journal_params(sb, sbi->s_journal);
5704                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
5705         }
5706
5707         if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) {
5708                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
5709                         err = -EROFS;
5710                         goto restore_opts;
5711                 }
5712
5713                 if (*flags & SB_RDONLY) {
5714                         err = sync_filesystem(sb);
5715                         if (err < 0)
5716                                 goto restore_opts;
5717                         err = dquot_suspend(sb, -1);
5718                         if (err < 0)
5719                                 goto restore_opts;
5720
5721                         /*
5722                          * First of all, the unconditional stuff we have to do
5723                          * to disable replay of the journal when we next remount
5724                          */
5725                         sb->s_flags |= SB_RDONLY;
5726
5727                         /*
5728                          * OK, test if we are remounting a valid rw partition
5729                          * readonly, and if so set the rdonly flag and then
5730                          * mark the partition as valid again.
5731                          */
5732                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5733                             (sbi->s_mount_state & EXT4_VALID_FS))
5734                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5735
5736                         if (sbi->s_journal) {
5737                                 /*
5738                                  * We let remount-ro finish even if marking fs
5739                                  * as clean failed...
5740                                  */
5741                                 ext4_mark_recovery_complete(sb, es);
5742                         }
5743                         if (sbi->s_mmp_tsk)
5744                                 kthread_stop(sbi->s_mmp_tsk);
5745                 } else {
5746                         /* Make sure we can mount this feature set readwrite */
5747                         if (ext4_has_feature_readonly(sb) ||
5748                             !ext4_feature_set_ok(sb, 0)) {
5749                                 err = -EROFS;
5750                                 goto restore_opts;
5751                         }
5752                         /*
5753                          * Make sure the group descriptor checksums
5754                          * are sane.  If they aren't, refuse to remount r/w.
5755                          */
5756                         for (g = 0; g < sbi->s_groups_count; g++) {
5757                                 struct ext4_group_desc *gdp =
5758                                         ext4_get_group_desc(sb, g, NULL);
5759
5760                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5761                                         ext4_msg(sb, KERN_ERR,
5762                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5763                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
5764                                                le16_to_cpu(gdp->bg_checksum));
5765                                         err = -EFSBADCRC;
5766                                         goto restore_opts;
5767                                 }
5768                         }
5769
5770                         /*
5771                          * If we have an unprocessed orphan list hanging
5772                          * around from a previously readonly bdev mount,
5773                          * require a full umount/remount for now.
5774                          */
5775                         if (es->s_last_orphan) {
5776                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5777                                        "remount RDWR because of unprocessed "
5778                                        "orphan inode list.  Please "
5779                                        "umount/remount instead");
5780                                 err = -EINVAL;
5781                                 goto restore_opts;
5782                         }
5783
5784                         /*
5785                          * Mounting a RDONLY partition read-write, so reread
5786                          * and store the current valid flag.  (It may have
5787                          * been changed by e2fsck since we originally mounted
5788                          * the partition.)
5789                          */
5790                         if (sbi->s_journal) {
5791                                 err = ext4_clear_journal_err(sb, es);
5792                                 if (err)
5793                                         goto restore_opts;
5794                         }
5795                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5796
5797                         err = ext4_setup_super(sb, es, 0);
5798                         if (err)
5799                                 goto restore_opts;
5800
5801                         sb->s_flags &= ~SB_RDONLY;
5802                         if (ext4_has_feature_mmp(sb))
5803                                 if (ext4_multi_mount_protect(sb,
5804                                                 le64_to_cpu(es->s_mmp_block))) {
5805                                         err = -EROFS;
5806                                         goto restore_opts;
5807                                 }
5808                         enable_quota = 1;
5809                 }
5810         }
5811
5812         /*
5813          * Reinitialize lazy itable initialization thread based on
5814          * current settings
5815          */
5816         if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
5817                 ext4_unregister_li_request(sb);
5818         else {
5819                 ext4_group_t first_not_zeroed;
5820                 first_not_zeroed = ext4_has_uninit_itable(sb);
5821                 ext4_register_li_request(sb, first_not_zeroed);
5822         }
5823
5824         /*
5825          * Handle creation of system zone data early because it can fail.
5826          * Releasing of existing data is done when we are sure remount will
5827          * succeed.
5828          */
5829         if (test_opt(sb, BLOCK_VALIDITY) && !sbi->s_system_blks) {
5830                 err = ext4_setup_system_zone(sb);
5831                 if (err)
5832                         goto restore_opts;
5833         }
5834
5835         if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) {
5836                 err = ext4_commit_super(sb, 1);
5837                 if (err)
5838                         goto restore_opts;
5839         }
5840
5841 #ifdef CONFIG_QUOTA
5842         /* Release old quota file names */
5843         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5844                 kfree(old_opts.s_qf_names[i]);
5845         if (enable_quota) {
5846                 if (sb_any_quota_suspended(sb))
5847                         dquot_resume(sb, -1);
5848                 else if (ext4_has_feature_quota(sb)) {
5849                         err = ext4_enable_quotas(sb);
5850                         if (err)
5851                                 goto restore_opts;
5852                 }
5853         }
5854 #endif
5855         if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks)
5856                 ext4_release_system_zone(sb);
5857
5858         /*
5859          * Some options can be enabled by ext4 and/or by VFS mount flag
5860          * either way we need to make sure it matches in both *flags and
5861          * s_flags. Copy those selected flags from s_flags to *flags
5862          */
5863         *flags = (*flags & ~vfs_flags) | (sb->s_flags & vfs_flags);
5864
5865         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5866         kfree(orig_data);
5867         return 0;
5868
5869 restore_opts:
5870         sb->s_flags = old_sb_flags;
5871         sbi->s_mount_opt = old_opts.s_mount_opt;
5872         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5873         sbi->s_resuid = old_opts.s_resuid;
5874         sbi->s_resgid = old_opts.s_resgid;
5875         sbi->s_commit_interval = old_opts.s_commit_interval;
5876         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5877         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5878         if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks)
5879                 ext4_release_system_zone(sb);
5880 #ifdef CONFIG_QUOTA
5881         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5882         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5883                 to_free[i] = get_qf_name(sb, sbi, i);
5884                 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]);
5885         }
5886         synchronize_rcu();
5887         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5888                 kfree(to_free[i]);
5889 #endif
5890         kfree(orig_data);
5891         return err;
5892 }
5893
5894 #ifdef CONFIG_QUOTA
5895 static int ext4_statfs_project(struct super_block *sb,
5896                                kprojid_t projid, struct kstatfs *buf)
5897 {
5898         struct kqid qid;
5899         struct dquot *dquot;
5900         u64 limit;
5901         u64 curblock;
5902
5903         qid = make_kqid_projid(projid);
5904         dquot = dqget(sb, qid);
5905         if (IS_ERR(dquot))
5906                 return PTR_ERR(dquot);
5907         spin_lock(&dquot->dq_dqb_lock);
5908
5909         limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit,
5910                              dquot->dq_dqb.dqb_bhardlimit);
5911         limit >>= sb->s_blocksize_bits;
5912
5913         if (limit && buf->f_blocks > limit) {
5914                 curblock = (dquot->dq_dqb.dqb_curspace +
5915                             dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits;
5916                 buf->f_blocks = limit;
5917                 buf->f_bfree = buf->f_bavail =
5918                         (buf->f_blocks > curblock) ?
5919                          (buf->f_blocks - curblock) : 0;
5920         }
5921
5922         limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit,
5923                              dquot->dq_dqb.dqb_ihardlimit);
5924         if (limit && buf->f_files > limit) {
5925                 buf->f_files = limit;
5926                 buf->f_ffree =
5927                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
5928                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
5929         }
5930
5931         spin_unlock(&dquot->dq_dqb_lock);
5932         dqput(dquot);
5933         return 0;
5934 }
5935 #endif
5936
5937 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5938 {
5939         struct super_block *sb = dentry->d_sb;
5940         struct ext4_sb_info *sbi = EXT4_SB(sb);
5941         struct ext4_super_block *es = sbi->s_es;
5942         ext4_fsblk_t overhead = 0, resv_blocks;
5943         u64 fsid;
5944         s64 bfree;
5945         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5946
5947         if (!test_opt(sb, MINIX_DF))
5948                 overhead = sbi->s_overhead;
5949
5950         buf->f_type = EXT4_SUPER_MAGIC;
5951         buf->f_bsize = sb->s_blocksize;
5952         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5953         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5954                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5955         /* prevent underflow in case that few free space is available */
5956         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5957         buf->f_bavail = buf->f_bfree -
5958                         (ext4_r_blocks_count(es) + resv_blocks);
5959         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5960                 buf->f_bavail = 0;
5961         buf->f_files = le32_to_cpu(es->s_inodes_count);
5962         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5963         buf->f_namelen = EXT4_NAME_LEN;
5964         fsid = le64_to_cpup((void *)es->s_uuid) ^
5965                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5966         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5967         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5968
5969 #ifdef CONFIG_QUOTA
5970         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
5971             sb_has_quota_limits_enabled(sb, PRJQUOTA))
5972                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
5973 #endif
5974         return 0;
5975 }
5976
5977
5978 #ifdef CONFIG_QUOTA
5979
5980 /*
5981  * Helper functions so that transaction is started before we acquire dqio_sem
5982  * to keep correct lock ordering of transaction > dqio_sem
5983  */
5984 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5985 {
5986         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5987 }
5988
5989 static int ext4_write_dquot(struct dquot *dquot)
5990 {
5991         int ret, err;
5992         handle_t *handle;
5993         struct inode *inode;
5994
5995         inode = dquot_to_inode(dquot);
5996         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5997                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5998         if (IS_ERR(handle))
5999                 return PTR_ERR(handle);
6000         ret = dquot_commit(dquot);
6001         err = ext4_journal_stop(handle);
6002         if (!ret)
6003                 ret = err;
6004         return ret;
6005 }
6006
6007 static int ext4_acquire_dquot(struct dquot *dquot)
6008 {
6009         int ret, err;
6010         handle_t *handle;
6011
6012         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
6013                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
6014         if (IS_ERR(handle))
6015                 return PTR_ERR(handle);
6016         ret = dquot_acquire(dquot);
6017         err = ext4_journal_stop(handle);
6018         if (!ret)
6019                 ret = err;
6020         return ret;
6021 }
6022
6023 static int ext4_release_dquot(struct dquot *dquot)
6024 {
6025         int ret, err;
6026         handle_t *handle;
6027
6028         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
6029                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
6030         if (IS_ERR(handle)) {
6031                 /* Release dquot anyway to avoid endless cycle in dqput() */
6032                 dquot_release(dquot);
6033                 return PTR_ERR(handle);
6034         }
6035         ret = dquot_release(dquot);
6036         err = ext4_journal_stop(handle);
6037         if (!ret)
6038                 ret = err;
6039         return ret;
6040 }
6041
6042 static int ext4_mark_dquot_dirty(struct dquot *dquot)
6043 {
6044         struct super_block *sb = dquot->dq_sb;
6045         struct ext4_sb_info *sbi = EXT4_SB(sb);
6046
6047         /* Are we journaling quotas? */
6048         if (ext4_has_feature_quota(sb) ||
6049             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
6050                 dquot_mark_dquot_dirty(dquot);
6051                 return ext4_write_dquot(dquot);
6052         } else {
6053                 return dquot_mark_dquot_dirty(dquot);
6054         }
6055 }
6056
6057 static int ext4_write_info(struct super_block *sb, int type)
6058 {
6059         int ret, err;
6060         handle_t *handle;
6061
6062         /* Data block + inode block */
6063         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
6064         if (IS_ERR(handle))
6065                 return PTR_ERR(handle);
6066         ret = dquot_commit_info(sb, type);
6067         err = ext4_journal_stop(handle);
6068         if (!ret)
6069                 ret = err;
6070         return ret;
6071 }
6072
6073 /*
6074  * Turn on quotas during mount time - we need to find
6075  * the quota file and such...
6076  */
6077 static int ext4_quota_on_mount(struct super_block *sb, int type)
6078 {
6079         return dquot_quota_on_mount(sb, get_qf_name(sb, EXT4_SB(sb), type),
6080                                         EXT4_SB(sb)->s_jquota_fmt, type);
6081 }
6082
6083 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
6084 {
6085         struct ext4_inode_info *ei = EXT4_I(inode);
6086
6087         /* The first argument of lockdep_set_subclass has to be
6088          * *exactly* the same as the argument to init_rwsem() --- in
6089          * this case, in init_once() --- or lockdep gets unhappy
6090          * because the name of the lock is set using the
6091          * stringification of the argument to init_rwsem().
6092          */
6093         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
6094         lockdep_set_subclass(&ei->i_data_sem, subclass);
6095 }
6096
6097 /*
6098  * Standard function to be called on quota_on
6099  */
6100 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
6101                          const struct path *path)
6102 {
6103         int err;
6104
6105         if (!test_opt(sb, QUOTA))
6106                 return -EINVAL;
6107
6108         /* Quotafile not on the same filesystem? */
6109         if (path->dentry->d_sb != sb)
6110                 return -EXDEV;
6111         /* Journaling quota? */
6112         if (EXT4_SB(sb)->s_qf_names[type]) {
6113                 /* Quotafile not in fs root? */
6114                 if (path->dentry->d_parent != sb->s_root)
6115                         ext4_msg(sb, KERN_WARNING,
6116                                 "Quota file not on filesystem root. "
6117                                 "Journaled quota will not work");
6118                 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
6119         } else {
6120                 /*
6121                  * Clear the flag just in case mount options changed since
6122                  * last time.
6123                  */
6124                 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
6125         }
6126
6127         /*
6128          * When we journal data on quota file, we have to flush journal to see
6129          * all updates to the file when we bypass pagecache...
6130          */
6131         if (EXT4_SB(sb)->s_journal &&
6132             ext4_should_journal_data(d_inode(path->dentry))) {
6133                 /*
6134                  * We don't need to lock updates but journal_flush() could
6135                  * otherwise be livelocked...
6136                  */
6137                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
6138                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
6139                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
6140                 if (err)
6141                         return err;
6142         }
6143
6144         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
6145         err = dquot_quota_on(sb, type, format_id, path);
6146         if (err) {
6147                 lockdep_set_quota_inode(path->dentry->d_inode,
6148                                              I_DATA_SEM_NORMAL);
6149         } else {
6150                 struct inode *inode = d_inode(path->dentry);
6151                 handle_t *handle;
6152
6153                 /*
6154                  * Set inode flags to prevent userspace from messing with quota
6155                  * files. If this fails, we return success anyway since quotas
6156                  * are already enabled and this is not a hard failure.
6157                  */
6158                 inode_lock(inode);
6159                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
6160                 if (IS_ERR(handle))
6161                         goto unlock_inode;
6162                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
6163                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
6164                                 S_NOATIME | S_IMMUTABLE);
6165                 err = ext4_mark_inode_dirty(handle, inode);
6166                 ext4_journal_stop(handle);
6167         unlock_inode:
6168                 inode_unlock(inode);
6169         }
6170         return err;
6171 }
6172
6173 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
6174                              unsigned int flags)
6175 {
6176         int err;
6177         struct inode *qf_inode;
6178         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
6179                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
6180                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
6181                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
6182         };
6183
6184         BUG_ON(!ext4_has_feature_quota(sb));
6185
6186         if (!qf_inums[type])
6187                 return -EPERM;
6188
6189         qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL);
6190         if (IS_ERR(qf_inode)) {
6191                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
6192                 return PTR_ERR(qf_inode);
6193         }
6194
6195         /* Don't account quota for quota files to avoid recursion */
6196         qf_inode->i_flags |= S_NOQUOTA;
6197         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
6198         err = dquot_load_quota_inode(qf_inode, type, format_id, flags);
6199         if (err)
6200                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
6201         iput(qf_inode);
6202
6203         return err;
6204 }
6205
6206 /* Enable usage tracking for all quota types. */
6207 static int ext4_enable_quotas(struct super_block *sb)
6208 {
6209         int type, err = 0;
6210         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
6211                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
6212                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
6213                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
6214         };
6215         bool quota_mopt[EXT4_MAXQUOTAS] = {
6216                 test_opt(sb, USRQUOTA),
6217                 test_opt(sb, GRPQUOTA),
6218                 test_opt(sb, PRJQUOTA),
6219         };
6220
6221         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
6222         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
6223                 if (qf_inums[type]) {
6224                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
6225                                 DQUOT_USAGE_ENABLED |
6226                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
6227                         if (err) {
6228                                 ext4_warning(sb,
6229                                         "Failed to enable quota tracking "
6230                                         "(type=%d, err=%d). Please run "
6231                                         "e2fsck to fix.", type, err);
6232                                 for (type--; type >= 0; type--)
6233                                         dquot_quota_off(sb, type);
6234
6235                                 return err;
6236                         }
6237                 }
6238         }
6239         return 0;
6240 }
6241
6242 static int ext4_quota_off(struct super_block *sb, int type)
6243 {
6244         struct inode *inode = sb_dqopt(sb)->files[type];
6245         handle_t *handle;
6246         int err;
6247
6248         /* Force all delayed allocation blocks to be allocated.
6249          * Caller already holds s_umount sem */
6250         if (test_opt(sb, DELALLOC))
6251                 sync_filesystem(sb);
6252
6253         if (!inode || !igrab(inode))
6254                 goto out;
6255
6256         err = dquot_quota_off(sb, type);
6257         if (err || ext4_has_feature_quota(sb))
6258                 goto out_put;
6259
6260         inode_lock(inode);
6261         /*
6262          * Update modification times of quota files when userspace can
6263          * start looking at them. If we fail, we return success anyway since
6264          * this is not a hard failure and quotas are already disabled.
6265          */
6266         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
6267         if (IS_ERR(handle)) {
6268                 err = PTR_ERR(handle);
6269                 goto out_unlock;
6270         }
6271         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
6272         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
6273         inode->i_mtime = inode->i_ctime = current_time(inode);
6274         err = ext4_mark_inode_dirty(handle, inode);
6275         ext4_journal_stop(handle);
6276 out_unlock:
6277         inode_unlock(inode);
6278 out_put:
6279         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
6280         iput(inode);
6281         return err;
6282 out:
6283         return dquot_quota_off(sb, type);
6284 }
6285
6286 /* Read data from quotafile - avoid pagecache and such because we cannot afford
6287  * acquiring the locks... As quota files are never truncated and quota code
6288  * itself serializes the operations (and no one else should touch the files)
6289  * we don't have to be afraid of races */
6290 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
6291                                size_t len, loff_t off)
6292 {
6293         struct inode *inode = sb_dqopt(sb)->files[type];
6294         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
6295         int offset = off & (sb->s_blocksize - 1);
6296         int tocopy;
6297         size_t toread;
6298         struct buffer_head *bh;
6299         loff_t i_size = i_size_read(inode);
6300
6301         if (off > i_size)
6302                 return 0;
6303         if (off+len > i_size)
6304                 len = i_size-off;
6305         toread = len;
6306         while (toread > 0) {
6307                 tocopy = sb->s_blocksize - offset < toread ?
6308                                 sb->s_blocksize - offset : toread;
6309                 bh = ext4_bread(NULL, inode, blk, 0);
6310                 if (IS_ERR(bh))
6311                         return PTR_ERR(bh);
6312                 if (!bh)        /* A hole? */
6313                         memset(data, 0, tocopy);
6314                 else
6315                         memcpy(data, bh->b_data+offset, tocopy);
6316                 brelse(bh);
6317                 offset = 0;
6318                 toread -= tocopy;
6319                 data += tocopy;
6320                 blk++;
6321         }
6322         return len;
6323 }
6324
6325 /* Write to quotafile (we know the transaction is already started and has
6326  * enough credits) */
6327 static ssize_t ext4_quota_write(struct super_block *sb, int type,
6328                                 const char *data, size_t len, loff_t off)
6329 {
6330         struct inode *inode = sb_dqopt(sb)->files[type];
6331         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
6332         int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1);
6333         int retries = 0;
6334         struct buffer_head *bh;
6335         handle_t *handle = journal_current_handle();
6336
6337         if (EXT4_SB(sb)->s_journal && !handle) {
6338                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
6339                         " cancelled because transaction is not started",
6340                         (unsigned long long)off, (unsigned long long)len);
6341                 return -EIO;
6342         }
6343         /*
6344          * Since we account only one data block in transaction credits,
6345          * then it is impossible to cross a block boundary.
6346          */
6347         if (sb->s_blocksize - offset < len) {
6348                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
6349                         " cancelled because not block aligned",
6350                         (unsigned long long)off, (unsigned long long)len);
6351                 return -EIO;
6352         }
6353
6354         do {
6355                 bh = ext4_bread(handle, inode, blk,
6356                                 EXT4_GET_BLOCKS_CREATE |
6357                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
6358         } while (PTR_ERR(bh) == -ENOSPC &&
6359                  ext4_should_retry_alloc(inode->i_sb, &retries));
6360         if (IS_ERR(bh))
6361                 return PTR_ERR(bh);
6362         if (!bh)
6363                 goto out;
6364         BUFFER_TRACE(bh, "get write access");
6365         err = ext4_journal_get_write_access(handle, bh);
6366         if (err) {
6367                 brelse(bh);
6368                 return err;
6369         }
6370         lock_buffer(bh);
6371         memcpy(bh->b_data+offset, data, len);
6372         flush_dcache_page(bh->b_page);
6373         unlock_buffer(bh);
6374         err = ext4_handle_dirty_metadata(handle, NULL, bh);
6375         brelse(bh);
6376 out:
6377         if (inode->i_size < off + len) {
6378                 i_size_write(inode, off + len);
6379                 EXT4_I(inode)->i_disksize = inode->i_size;
6380                 err2 = ext4_mark_inode_dirty(handle, inode);
6381                 if (unlikely(err2 && !err))
6382                         err = err2;
6383         }
6384         return err ? err : len;
6385 }
6386 #endif
6387
6388 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
6389                        const char *dev_name, void *data)
6390 {
6391         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
6392 }
6393
6394 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
6395 static inline void register_as_ext2(void)
6396 {
6397         int err = register_filesystem(&ext2_fs_type);
6398         if (err)
6399                 printk(KERN_WARNING
6400                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
6401 }
6402
6403 static inline void unregister_as_ext2(void)
6404 {
6405         unregister_filesystem(&ext2_fs_type);
6406 }
6407
6408 static inline int ext2_feature_set_ok(struct super_block *sb)
6409 {
6410         if (ext4_has_unknown_ext2_incompat_features(sb))
6411                 return 0;
6412         if (sb_rdonly(sb))
6413                 return 1;
6414         if (ext4_has_unknown_ext2_ro_compat_features(sb))
6415                 return 0;
6416         return 1;
6417 }
6418 #else
6419 static inline void register_as_ext2(void) { }
6420 static inline void unregister_as_ext2(void) { }
6421 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
6422 #endif
6423
6424 static inline void register_as_ext3(void)
6425 {
6426         int err = register_filesystem(&ext3_fs_type);
6427         if (err)
6428                 printk(KERN_WARNING
6429                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
6430 }
6431
6432 static inline void unregister_as_ext3(void)
6433 {
6434         unregister_filesystem(&ext3_fs_type);
6435 }
6436
6437 static inline int ext3_feature_set_ok(struct super_block *sb)
6438 {
6439         if (ext4_has_unknown_ext3_incompat_features(sb))
6440                 return 0;
6441         if (!ext4_has_feature_journal(sb))
6442                 return 0;
6443         if (sb_rdonly(sb))
6444                 return 1;
6445         if (ext4_has_unknown_ext3_ro_compat_features(sb))
6446                 return 0;
6447         return 1;
6448 }
6449
6450 static struct file_system_type ext4_fs_type = {
6451         .owner          = THIS_MODULE,
6452         .name           = "ext4",
6453         .mount          = ext4_mount,
6454         .kill_sb        = kill_block_super,
6455         .fs_flags       = FS_REQUIRES_DEV,
6456 };
6457 MODULE_ALIAS_FS("ext4");
6458
6459 /* Shared across all ext4 file systems */
6460 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
6461
6462 static int __init ext4_init_fs(void)
6463 {
6464         int i, err;
6465
6466         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
6467         ext4_li_info = NULL;
6468         mutex_init(&ext4_li_mtx);
6469
6470         /* Build-time check for flags consistency */
6471         ext4_check_flag_values();
6472
6473         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
6474                 init_waitqueue_head(&ext4__ioend_wq[i]);
6475
6476         err = ext4_init_es();
6477         if (err)
6478                 return err;
6479
6480         err = ext4_init_pending();
6481         if (err)
6482                 goto out7;
6483
6484         err = ext4_init_post_read_processing();
6485         if (err)
6486                 goto out6;
6487
6488         err = ext4_init_pageio();
6489         if (err)
6490                 goto out5;
6491
6492         err = ext4_init_system_zone();
6493         if (err)
6494                 goto out4;
6495
6496         err = ext4_init_sysfs();
6497         if (err)
6498                 goto out3;
6499
6500         err = ext4_init_mballoc();
6501         if (err)
6502                 goto out2;
6503         err = init_inodecache();
6504         if (err)
6505                 goto out1;
6506         register_as_ext3();
6507         register_as_ext2();
6508         err = register_filesystem(&ext4_fs_type);
6509         if (err)
6510                 goto out;
6511
6512         return 0;
6513 out:
6514         unregister_as_ext2();
6515         unregister_as_ext3();
6516         destroy_inodecache();
6517 out1:
6518         ext4_exit_mballoc();
6519 out2:
6520         ext4_exit_sysfs();
6521 out3:
6522         ext4_exit_system_zone();
6523 out4:
6524         ext4_exit_pageio();
6525 out5:
6526         ext4_exit_post_read_processing();
6527 out6:
6528         ext4_exit_pending();
6529 out7:
6530         ext4_exit_es();
6531
6532         return err;
6533 }
6534
6535 static void __exit ext4_exit_fs(void)
6536 {
6537         ext4_destroy_lazyinit_thread();
6538         unregister_as_ext2();
6539         unregister_as_ext3();
6540         unregister_filesystem(&ext4_fs_type);
6541         destroy_inodecache();
6542         ext4_exit_mballoc();
6543         ext4_exit_sysfs();
6544         ext4_exit_system_zone();
6545         ext4_exit_pageio();
6546         ext4_exit_post_read_processing();
6547         ext4_exit_es();
6548         ext4_exit_pending();
6549 }
6550
6551 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
6552 MODULE_DESCRIPTION("Fourth Extended Filesystem");
6553 MODULE_LICENSE("GPL");
6554 MODULE_SOFTDEP("pre: crc32c");
6555 module_init(ext4_init_fs)
6556 module_exit(ext4_exit_fs)