fs/ext4/super.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/ext4/super.c
   4  *
   5  * Copyright (C) 1992, 1993, 1994, 1995
   6  * Remy Card ([email protected])
   7  * Laboratoire MASI - Institut Blaise Pascal
   8  * Universite Pierre et Marie Curie (Paris VI)
   9  *
  10  *  from
  11  *
  12  *  linux/fs/minix/inode.c
  13  *
  14  *  Copyright (C) 1991, 1992  Linus Torvalds
  15  *
  16  *  Big-endian to little-endian byte-swapping/bitmaps by
  17  *        David S. Miller ([email protected]), 1995
  18  */
  19
  20 #include <linux/module.h>
  21 #include <linux/string.h>
  22 #include <linux/fs.h>
  23 #include <linux/time.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/backing-dev.h>
  29 #include <linux/parser.h>
  30 #include <linux/buffer_head.h>
  31 #include <linux/exportfs.h>
  32 #include <linux/vfs.h>
  33 #include <linux/random.h>
  34 #include <linux/mount.h>
  35 #include <linux/namei.h>
  36 #include <linux/quotaops.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/dax.h>
  42 #include <linux/cleancache.h>
  43 #include <linux/uaccess.h>
  44 #include <linux/iversion.h>
  45 #include <linux/unicode.h>
  46 #include <linux/part_stat.h>
  47 #include <linux/kthread.h>
  48 #include <linux/freezer.h>
  49
  50 #include "ext4.h"
  51 #include "ext4_extents.h"       /* Needed for trace points definition */
  52 #include "ext4_jbd2.h"
  53 #include "xattr.h"
  54 #include "acl.h"
  55 #include "mballoc.h"
  56 #include "fsmap.h"
  57
  58 #define CREATE_TRACE_POINTS
  59 #include <trace/events/ext4.h>
  60
  61 static struct ext4_lazy_init *ext4_li_info;
  62 static DEFINE_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 void ext4_update_super(struct super_block *sb);
  69 static int ext4_commit_super(struct super_block *sb);
  70 static int ext4_mark_recovery_complete(struct super_block *sb,
  71                                         struct ext4_super_block *es);
  72 static int ext4_clear_journal_err(struct super_block *sb,
  73                                   struct ext4_super_block *es);
  74 static int ext4_sync_fs(struct super_block *sb, int wait);
  75 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  76 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  77 static int ext4_unfreeze(struct super_block *sb);
  78 static int ext4_freeze(struct super_block *sb);
  79 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  80                        const char *dev_name, void *data);
  81 static inline int ext2_feature_set_ok(struct super_block *sb);
  82 static inline int ext3_feature_set_ok(struct super_block *sb);
  83 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  84 static void ext4_destroy_lazyinit_thread(void);
  85 static void ext4_unregister_li_request(struct super_block *sb);
  86 static void ext4_clear_request_list(void);
  87 static struct inode *ext4_get_journal_inode(struct super_block *sb,
  88                                             unsigned int journal_inum);
  89
  90 /*
  91  * Lock ordering
  92  *
  93  * Note the difference between i_mmap_sem (EXT4_I(inode)->i_mmap_sem) and
  94  * i_mmap_rwsem (inode->i_mmap_rwsem)!
  95  *
  96  * page fault path:
  97  * mmap_lock -> sb_start_pagefault -> i_mmap_sem (r) -> transaction start ->
  98  *   page lock -> i_data_sem (rw)
  99  *
 100  * buffered write path:
 101  * sb_start_write -> i_mutex -> mmap_lock
 102  * sb_start_write -> i_mutex -> transaction start -> page lock ->
 103  *   i_data_sem (rw)
 104  *
 105  * truncate:
 106  * sb_start_write -> i_mutex -> i_mmap_sem (w) -> i_mmap_rwsem (w) -> page lock
 107  * sb_start_write -> i_mutex -> i_mmap_sem (w) -> transaction start ->
 108  *   i_data_sem (rw)
 109  *
 110  * direct IO:
 111  * sb_start_write -> i_mutex -> mmap_lock
 112  * sb_start_write -> i_mutex -> transaction start -> i_data_sem (rw)
 113  *
 114  * writepages:
 115  * transaction start -> page lock(s) -> i_data_sem (rw)
 116  */
 117
 118 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
 119 static struct file_system_type ext2_fs_type = {
 120         .owner          = THIS_MODULE,
 121         .name           = "ext2",
 122         .mount          = ext4_mount,
 123         .kill_sb        = kill_block_super,
 124         .fs_flags       = FS_REQUIRES_DEV,
 125 };
 126 MODULE_ALIAS_FS("ext2");
 127 MODULE_ALIAS("ext2");
 128 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
 129 #else
 130 #define IS_EXT2_SB(sb) (0)
 131 #endif
 132
 133
 134 static struct file_system_type ext3_fs_type = {
 135         .owner          = THIS_MODULE,
 136         .name           = "ext3",
 137         .mount          = ext4_mount,
 138         .kill_sb        = kill_block_super,
 139         .fs_flags       = FS_REQUIRES_DEV,
 140 };
 141 MODULE_ALIAS_FS("ext3");
 142 MODULE_ALIAS("ext3");
 143 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 144
 145
 146 static inline void __ext4_read_bh(struct buffer_head *bh, int op_flags,
 147                                   bh_end_io_t *end_io)
 148 {
 149         /*
 150          * buffer's verified bit is no longer valid after reading from
 151          * disk again due to write out error, clear it to make sure we
 152          * recheck the buffer contents.
 153          */
 154         clear_buffer_verified(bh);
 155
 156         bh->b_end_io = end_io ? end_io : end_buffer_read_sync;
 157         get_bh(bh);
 158         submit_bh(REQ_OP_READ, op_flags, bh);
 159 }
 160
 161 void ext4_read_bh_nowait(struct buffer_head *bh, int op_flags,
 162                          bh_end_io_t *end_io)
 163 {
 164         BUG_ON(!buffer_locked(bh));
 165
 166         if (ext4_buffer_uptodate(bh)) {
 167                 unlock_buffer(bh);
 168                 return;
 169         }
 170         __ext4_read_bh(bh, op_flags, end_io);
 171 }
 172
 173 int ext4_read_bh(struct buffer_head *bh, int op_flags, bh_end_io_t *end_io)
 174 {
 175         BUG_ON(!buffer_locked(bh));
 176
 177         if (ext4_buffer_uptodate(bh)) {
 178                 unlock_buffer(bh);
 179                 return 0;
 180         }
 181
 182         __ext4_read_bh(bh, op_flags, end_io);
 183
 184         wait_on_buffer(bh);
 185         if (buffer_uptodate(bh))
 186                 return 0;
 187         return -EIO;
 188 }
 189
 190 int ext4_read_bh_lock(struct buffer_head *bh, int op_flags, bool wait)
 191 {
 192         if (trylock_buffer(bh)) {
 193                 if (wait)
 194                         return ext4_read_bh(bh, op_flags, NULL);
 195                 ext4_read_bh_nowait(bh, op_flags, NULL);
 196                 return 0;
 197         }
 198         if (wait) {
 199                 wait_on_buffer(bh);
 200                 if (buffer_uptodate(bh))
 201                         return 0;
 202                 return -EIO;
 203         }
 204         return 0;
 205 }
 206
 207 /*
 208  * This works like __bread_gfp() except it uses ERR_PTR for error
 209  * returns.  Currently with sb_bread it's impossible to distinguish
 210  * between ENOMEM and EIO situations (since both result in a NULL
 211  * return.
 212  */
 213 static struct buffer_head *__ext4_sb_bread_gfp(struct super_block *sb,
 214                                                sector_t block, int op_flags,
 215                                                gfp_t gfp)
 216 {
 217         struct buffer_head *bh;
 218         int ret;
 219
 220         bh = sb_getblk_gfp(sb, block, gfp);
 221         if (bh == NULL)
 222                 return ERR_PTR(-ENOMEM);
 223         if (ext4_buffer_uptodate(bh))
 224                 return bh;
 225
 226         ret = ext4_read_bh_lock(bh, REQ_META | op_flags, true);
 227         if (ret) {
 228                 put_bh(bh);
 229                 return ERR_PTR(ret);
 230         }
 231         return bh;
 232 }
 233
 234 struct buffer_head *ext4_sb_bread(struct super_block *sb, sector_t block,
 235                                    int op_flags)
 236 {
 237         return __ext4_sb_bread_gfp(sb, block, op_flags, __GFP_MOVABLE);
 238 }
 239
 240 struct buffer_head *ext4_sb_bread_unmovable(struct super_block *sb,
 241                                             sector_t block)
 242 {
 243         return __ext4_sb_bread_gfp(sb, block, 0, 0);
 244 }
 245
 246 void ext4_sb_breadahead_unmovable(struct super_block *sb, sector_t block)
 247 {
 248         struct buffer_head *bh = sb_getblk_gfp(sb, block, 0);
 249
 250         if (likely(bh)) {
 251                 ext4_read_bh_lock(bh, REQ_RAHEAD, false);
 252                 brelse(bh);
 253         }
 254 }
 255
 256 static int ext4_verify_csum_type(struct super_block *sb,
 257                                  struct ext4_super_block *es)
 258 {
 259         if (!ext4_has_feature_metadata_csum(sb))
 260                 return 1;
 261
 262         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 263 }
 264
 265 static __le32 ext4_superblock_csum(struct super_block *sb,
 266                                    struct ext4_super_block *es)
 267 {
 268         struct ext4_sb_info *sbi = EXT4_SB(sb);
 269         int offset = offsetof(struct ext4_super_block, s_checksum);
 270         __u32 csum;
 271
 272         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 273
 274         return cpu_to_le32(csum);
 275 }
 276
 277 static int ext4_superblock_csum_verify(struct super_block *sb,
 278                                        struct ext4_super_block *es)
 279 {
 280         if (!ext4_has_metadata_csum(sb))
 281                 return 1;
 282
 283         return es->s_checksum == ext4_superblock_csum(sb, es);
 284 }
 285
 286 void ext4_superblock_csum_set(struct super_block *sb)
 287 {
 288         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 289
 290         if (!ext4_has_metadata_csum(sb))
 291                 return;
 292
 293         es->s_checksum = ext4_superblock_csum(sb, es);
 294 }
 295
 296 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 297                                struct ext4_group_desc *bg)
 298 {
 299         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 300                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 301                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 302 }
 303
 304 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 305                                struct ext4_group_desc *bg)
 306 {
 307         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 308                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 309                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 310 }
 311
 312 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 313                               struct ext4_group_desc *bg)
 314 {
 315         return le32_to_cpu(bg->bg_inode_table_lo) |
 316                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 317                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 318 }
 319
 320 __u32 ext4_free_group_clusters(struct super_block *sb,
 321                                struct ext4_group_desc *bg)
 322 {
 323         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 324                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 325                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 326 }
 327
 328 __u32 ext4_free_inodes_count(struct super_block *sb,
 329                               struct ext4_group_desc *bg)
 330 {
 331         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 332                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 333                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 334 }
 335
 336 __u32 ext4_used_dirs_count(struct super_block *sb,
 337                               struct ext4_group_desc *bg)
 338 {
 339         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 340                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 341                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 342 }
 343
 344 __u32 ext4_itable_unused_count(struct super_block *sb,
 345                               struct ext4_group_desc *bg)
 346 {
 347         return le16_to_cpu(bg->bg_itable_unused_lo) |
 348                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 349                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 350 }
 351
 352 void ext4_block_bitmap_set(struct super_block *sb,
 353                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 354 {
 355         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 356         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 357                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 358 }
 359
 360 void ext4_inode_bitmap_set(struct super_block *sb,
 361                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 362 {
 363         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 364         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 365                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 366 }
 367
 368 void ext4_inode_table_set(struct super_block *sb,
 369                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 370 {
 371         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 372         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 373                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 374 }
 375
 376 void ext4_free_group_clusters_set(struct super_block *sb,
 377                                   struct ext4_group_desc *bg, __u32 count)
 378 {
 379         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 380         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 381                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 382 }
 383
 384 void ext4_free_inodes_set(struct super_block *sb,
 385                           struct ext4_group_desc *bg, __u32 count)
 386 {
 387         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 388         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 389                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 390 }
 391
 392 void ext4_used_dirs_set(struct super_block *sb,
 393                           struct ext4_group_desc *bg, __u32 count)
 394 {
 395         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 396         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 397                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 398 }
 399
 400 void ext4_itable_unused_set(struct super_block *sb,
 401                           struct ext4_group_desc *bg, __u32 count)
 402 {
 403         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 404         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 405                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 406 }
 407
 408 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi, time64_t now)
 409 {
 410         now = clamp_val(now, 0, (1ull << 40) - 1);
 411
 412         *lo = cpu_to_le32(lower_32_bits(now));
 413         *hi = upper_32_bits(now);
 414 }
 415
 416 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi)
 417 {
 418         return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo);
 419 }
 420 #define ext4_update_tstamp(es, tstamp) \
 421         __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi, \
 422                              ktime_get_real_seconds())
 423 #define ext4_get_tstamp(es, tstamp) \
 424         __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
 425
 426 /*
 427  * The del_gendisk() function uninitializes the disk-specific data
 428  * structures, including the bdi structure, without telling anyone
 429  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 430  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 431  * This is a kludge to prevent these oops until we can put in a proper
 432  * hook in del_gendisk() to inform the VFS and file system layers.
 433  */
 434 static int block_device_ejected(struct super_block *sb)
 435 {
 436         struct inode *bd_inode = sb->s_bdev->bd_inode;
 437         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 438
 439         return bdi->dev == NULL;
 440 }
 441
 442 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 443 {
 444         struct super_block              *sb = journal->j_private;
 445         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 446         int                             error = is_journal_aborted(journal);
 447         struct ext4_journal_cb_entry    *jce;
 448
 449         BUG_ON(txn->t_state == T_FINISHED);
 450
 451         ext4_process_freed_data(sb, txn->t_tid);
 452
 453         spin_lock(&sbi->s_md_lock);
 454         while (!list_empty(&txn->t_private_list)) {
 455                 jce = list_entry(txn->t_private_list.next,
 456                                  struct ext4_journal_cb_entry, jce_list);
 457                 list_del_init(&jce->jce_list);
 458                 spin_unlock(&sbi->s_md_lock);
 459                 jce->jce_func(sb, jce, error);
 460                 spin_lock(&sbi->s_md_lock);
 461         }
 462         spin_unlock(&sbi->s_md_lock);
 463 }
 464
 465 /*
 466  * This writepage callback for write_cache_pages()
 467  * takes care of a few cases after page cleaning.
 468  *
 469  * write_cache_pages() already checks for dirty pages
 470  * and calls clear_page_dirty_for_io(), which we want,
 471  * to write protect the pages.
 472  *
 473  * However, we may have to redirty a page (see below.)
 474  */
 475 static int ext4_journalled_writepage_callback(struct page *page,
 476                                               struct writeback_control *wbc,
 477                                               void *data)
 478 {
 479         transaction_t *transaction = (transaction_t *) data;
 480         struct buffer_head *bh, *head;
 481         struct journal_head *jh;
 482
 483         bh = head = page_buffers(page);
 484         do {
 485                 /*
 486                  * We have to redirty a page in these cases:
 487                  * 1) If buffer is dirty, it means the page was dirty because it
 488                  * contains a buffer that needs checkpointing. So the dirty bit
 489                  * needs to be preserved so that checkpointing writes the buffer
 490                  * properly.
 491                  * 2) If buffer is not part of the committing transaction
 492                  * (we may have just accidentally come across this buffer because
 493                  * inode range tracking is not exact) or if the currently running
 494                  * transaction already contains this buffer as well, dirty bit
 495                  * needs to be preserved so that the buffer gets writeprotected
 496                  * properly on running transaction's commit.
 497                  */
 498                 jh = bh2jh(bh);
 499                 if (buffer_dirty(bh) ||
 500                     (jh && (jh->b_transaction != transaction ||
 501                             jh->b_next_transaction))) {
 502                         redirty_page_for_writepage(wbc, page);
 503                         goto out;
 504                 }
 505         } while ((bh = bh->b_this_page) != head);
 506
 507 out:
 508         return AOP_WRITEPAGE_ACTIVATE;
 509 }
 510
 511 static int ext4_journalled_submit_inode_data_buffers(struct jbd2_inode *jinode)
 512 {
 513         struct address_space *mapping = jinode->i_vfs_inode->i_mapping;
 514         struct writeback_control wbc = {
 515                 .sync_mode =  WB_SYNC_ALL,
 516                 .nr_to_write = LONG_MAX,
 517                 .range_start = jinode->i_dirty_start,
 518                 .range_end = jinode->i_dirty_end,
 519         };
 520
 521         return write_cache_pages(mapping, &wbc,
 522                                  ext4_journalled_writepage_callback,
 523                                  jinode->i_transaction);
 524 }
 525
 526 static int ext4_journal_submit_inode_data_buffers(struct jbd2_inode *jinode)
 527 {
 528         int ret;
 529
 530         if (ext4_should_journal_data(jinode->i_vfs_inode))
 531                 ret = ext4_journalled_submit_inode_data_buffers(jinode);
 532         else
 533                 ret = jbd2_journal_submit_inode_data_buffers(jinode);
 534
 535         return ret;
 536 }
 537
 538 static int ext4_journal_finish_inode_data_buffers(struct jbd2_inode *jinode)
 539 {
 540         int ret = 0;
 541
 542         if (!ext4_should_journal_data(jinode->i_vfs_inode))
 543                 ret = jbd2_journal_finish_inode_data_buffers(jinode);
 544
 545         return ret;
 546 }
 547
 548 static bool system_going_down(void)
 549 {
 550         return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF
 551                 || system_state == SYSTEM_RESTART;
 552 }
 553
 554 struct ext4_err_translation {
 555         int code;
 556         int errno;
 557 };
 558
 559 #define EXT4_ERR_TRANSLATE(err) { .code = EXT4_ERR_##err, .errno = err }
 560
 561 static struct ext4_err_translation err_translation[] = {
 562         EXT4_ERR_TRANSLATE(EIO),
 563         EXT4_ERR_TRANSLATE(ENOMEM),
 564         EXT4_ERR_TRANSLATE(EFSBADCRC),
 565         EXT4_ERR_TRANSLATE(EFSCORRUPTED),
 566         EXT4_ERR_TRANSLATE(ENOSPC),
 567         EXT4_ERR_TRANSLATE(ENOKEY),
 568         EXT4_ERR_TRANSLATE(EROFS),
 569         EXT4_ERR_TRANSLATE(EFBIG),
 570         EXT4_ERR_TRANSLATE(EEXIST),
 571         EXT4_ERR_TRANSLATE(ERANGE),
 572         EXT4_ERR_TRANSLATE(EOVERFLOW),
 573         EXT4_ERR_TRANSLATE(EBUSY),
 574         EXT4_ERR_TRANSLATE(ENOTDIR),
 575         EXT4_ERR_TRANSLATE(ENOTEMPTY),
 576         EXT4_ERR_TRANSLATE(ESHUTDOWN),
 577         EXT4_ERR_TRANSLATE(EFAULT),
 578 };
 579
 580 static int ext4_errno_to_code(int errno)
 581 {
 582         int i;
 583
 584         for (i = 0; i < ARRAY_SIZE(err_translation); i++)
 585                 if (err_translation[i].errno == errno)
 586                         return err_translation[i].code;
 587         return EXT4_ERR_UNKNOWN;
 588 }
 589
 590 static void save_error_info(struct super_block *sb, int error,
 591                             __u32 ino, __u64 block,
 592                             const char *func, unsigned int line)
 593 {
 594         struct ext4_sb_info *sbi = EXT4_SB(sb);
 595
 596         /* We default to EFSCORRUPTED error... */
 597         if (error == 0)
 598                 error = EFSCORRUPTED;
 599
 600         spin_lock(&sbi->s_error_lock);
 601         sbi->s_add_error_count++;
 602         sbi->s_last_error_code = error;
 603         sbi->s_last_error_line = line;
 604         sbi->s_last_error_ino = ino;
 605         sbi->s_last_error_block = block;
 606         sbi->s_last_error_func = func;
 607         sbi->s_last_error_time = ktime_get_real_seconds();
 608         if (!sbi->s_first_error_time) {
 609                 sbi->s_first_error_code = error;
 610                 sbi->s_first_error_line = line;
 611                 sbi->s_first_error_ino = ino;
 612                 sbi->s_first_error_block = block;
 613                 sbi->s_first_error_func = func;
 614                 sbi->s_first_error_time = sbi->s_last_error_time;
 615         }
 616         spin_unlock(&sbi->s_error_lock);
 617 }
 618
 619 /* Deal with the reporting of failure conditions on a filesystem such as
 620  * inconsistencies detected or read IO failures.
 621  *
 622  * On ext2, we can store the error state of the filesystem in the
 623  * superblock.  That is not possible on ext4, because we may have other
 624  * write ordering constraints on the superblock which prevent us from
 625  * writing it out straight away; and given that the journal is about to
 626  * be aborted, we can't rely on the current, or future, transactions to
 627  * write out the superblock safely.
 628  *
 629  * We'll just use the jbd2_journal_abort() error code to record an error in
 630  * the journal instead.  On recovery, the journal will complain about
 631  * that error until we've noted it down and cleared it.
 632  *
 633  * If force_ro is set, we unconditionally force the filesystem into an
 634  * ABORT|READONLY state, unless the error response on the fs has been set to
 635  * panic in which case we take the easy way out and panic immediately. This is
 636  * used to deal with unrecoverable failures such as journal IO errors or ENOMEM
 637  * at a critical moment in log management.
 638  */
 639 static void ext4_handle_error(struct super_block *sb, bool force_ro, int error,
 640                               __u32 ino, __u64 block,
 641                               const char *func, unsigned int line)
 642 {
 643         journal_t *journal = EXT4_SB(sb)->s_journal;
 644         bool continue_fs = !force_ro && test_opt(sb, ERRORS_CONT);
 645
 646         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 647         if (test_opt(sb, WARN_ON_ERROR))
 648                 WARN_ON_ONCE(1);
 649
 650         if (!continue_fs && !sb_rdonly(sb)) {
 651                 ext4_set_mount_flag(sb, EXT4_MF_FS_ABORTED);
 652                 if (journal)
 653                         jbd2_journal_abort(journal, -EIO);
 654         }
 655
 656         if (!bdev_read_only(sb->s_bdev)) {
 657                 save_error_info(sb, error, ino, block, func, line);
 658                 /*
 659                  * In case the fs should keep running, we need to writeout
 660                  * superblock through the journal. Due to lock ordering
 661                  * constraints, it may not be safe to do it right here so we
 662                  * defer superblock flushing to a workqueue.
 663                  */
 664                 if (continue_fs)
 665                         schedule_work(&EXT4_SB(sb)->s_error_work);
 666                 else
 667                         ext4_commit_super(sb);
 668         }
 669
 670         if (sb_rdonly(sb) || continue_fs)
 671                 return;
 672
 673         /*
 674          * We force ERRORS_RO behavior when system is rebooting. Otherwise we
 675          * could panic during 'reboot -f' as the underlying device got already
 676          * disabled.
 677          */
 678         if (test_opt(sb, ERRORS_PANIC) && !system_going_down()) {
 679                 panic("EXT4-fs (device %s): panic forced after error\n",
 680                         sb->s_id);
 681         }
 682         ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 683         /*
 684          * Make sure updated value of ->s_mount_flags will be visible before
 685          * ->s_flags update
 686          */
 687         smp_wmb();
 688         sb->s_flags |= SB_RDONLY;
 689 }
 690
 691 static void flush_stashed_error_work(struct work_struct *work)
 692 {
 693         struct ext4_sb_info *sbi = container_of(work, struct ext4_sb_info,
 694                                                 s_error_work);
 695         journal_t *journal = sbi->s_journal;
 696         handle_t *handle;
 697
 698         /*
 699          * If the journal is still running, we have to write out superblock
 700          * through the journal to avoid collisions of other journalled sb
 701          * updates.
 702          *
 703          * We use directly jbd2 functions here to avoid recursing back into
 704          * ext4 error handling code during handling of previous errors.
 705          */
 706         if (!sb_rdonly(sbi->s_sb) && journal) {
 707                 handle = jbd2_journal_start(journal, 1);
 708                 if (IS_ERR(handle))
 709                         goto write_directly;
 710                 if (jbd2_journal_get_write_access(handle, sbi->s_sbh)) {
 711                         jbd2_journal_stop(handle);
 712                         goto write_directly;
 713                 }
 714                 ext4_update_super(sbi->s_sb);
 715                 if (jbd2_journal_dirty_metadata(handle, sbi->s_sbh)) {
 716                         jbd2_journal_stop(handle);
 717                         goto write_directly;
 718                 }
 719                 jbd2_journal_stop(handle);
 720                 return;
 721         }
 722 write_directly:
 723         /*
 724          * Write through journal failed. Write sb directly to get error info
 725          * out and hope for the best.
 726          */
 727         ext4_commit_super(sbi->s_sb);
 728 }
 729
 730 #define ext4_error_ratelimit(sb)                                        \
 731                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 732                              "EXT4-fs error")
 733
 734 void __ext4_error(struct super_block *sb, const char *function,
 735                   unsigned int line, bool force_ro, int error, __u64 block,
 736                   const char *fmt, ...)
 737 {
 738         struct va_format vaf;
 739         va_list args;
 740
 741         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 742                 return;
 743
 744         trace_ext4_error(sb, function, line);
 745         if (ext4_error_ratelimit(sb)) {
 746                 va_start(args, fmt);
 747                 vaf.fmt = fmt;
 748                 vaf.va = &args;
 749                 printk(KERN_CRIT
 750                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 751                        sb->s_id, function, line, current->comm, &vaf);
 752                 va_end(args);
 753         }
 754         ext4_handle_error(sb, force_ro, error, 0, block, function, line);
 755 }
 756
 757 void __ext4_error_inode(struct inode *inode, const char *function,
 758                         unsigned int line, ext4_fsblk_t block, int error,
 759                         const char *fmt, ...)
 760 {
 761         va_list args;
 762         struct va_format vaf;
 763
 764         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 765                 return;
 766
 767         trace_ext4_error(inode->i_sb, function, line);
 768         if (ext4_error_ratelimit(inode->i_sb)) {
 769                 va_start(args, fmt);
 770                 vaf.fmt = fmt;
 771                 vaf.va = &args;
 772                 if (block)
 773                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 774                                "inode #%lu: block %llu: comm %s: %pV\n",
 775                                inode->i_sb->s_id, function, line, inode->i_ino,
 776                                block, current->comm, &vaf);
 777                 else
 778                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 779                                "inode #%lu: comm %s: %pV\n",
 780                                inode->i_sb->s_id, function, line, inode->i_ino,
 781                                current->comm, &vaf);
 782                 va_end(args);
 783         }
 784         ext4_handle_error(inode->i_sb, false, error, inode->i_ino, block,
 785                           function, line);
 786 }
 787
 788 void __ext4_error_file(struct file *file, const char *function,
 789                        unsigned int line, ext4_fsblk_t block,
 790                        const char *fmt, ...)
 791 {
 792         va_list args;
 793         struct va_format vaf;
 794         struct inode *inode = file_inode(file);
 795         char pathname[80], *path;
 796
 797         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 798                 return;
 799
 800         trace_ext4_error(inode->i_sb, function, line);
 801         if (ext4_error_ratelimit(inode->i_sb)) {
 802                 path = file_path(file, pathname, sizeof(pathname));
 803                 if (IS_ERR(path))
 804                         path = "(unknown)";
 805                 va_start(args, fmt);
 806                 vaf.fmt = fmt;
 807                 vaf.va = &args;
 808                 if (block)
 809                         printk(KERN_CRIT
 810                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 811                                "block %llu: comm %s: path %s: %pV\n",
 812                                inode->i_sb->s_id, function, line, inode->i_ino,
 813                                block, current->comm, path, &vaf);
 814                 else
 815                         printk(KERN_CRIT
 816                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 817                                "comm %s: path %s: %pV\n",
 818                                inode->i_sb->s_id, function, line, inode->i_ino,
 819                                current->comm, path, &vaf);
 820                 va_end(args);
 821         }
 822         ext4_handle_error(inode->i_sb, false, EFSCORRUPTED, inode->i_ino, block,
 823                           function, line);
 824 }
 825
 826 const char *ext4_decode_error(struct super_block *sb, int errno,
 827                               char nbuf[16])
 828 {
 829         char *errstr = NULL;
 830
 831         switch (errno) {
 832         case -EFSCORRUPTED:
 833                 errstr = "Corrupt filesystem";
 834                 break;
 835         case -EFSBADCRC:
 836                 errstr = "Filesystem failed CRC";
 837                 break;
 838         case -EIO:
 839                 errstr = "IO failure";
 840                 break;
 841         case -ENOMEM:
 842                 errstr = "Out of memory";
 843                 break;
 844         case -EROFS:
 845                 if (!sb || (EXT4_SB(sb)->s_journal &&
 846                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 847                         errstr = "Journal has aborted";
 848                 else
 849                         errstr = "Readonly filesystem";
 850                 break;
 851         default:
 852                 /* If the caller passed in an extra buffer for unknown
 853                  * errors, textualise them now.  Else we just return
 854                  * NULL. */
 855                 if (nbuf) {
 856                         /* Check for truncated error codes... */
 857                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 858                                 errstr = nbuf;
 859                 }
 860                 break;
 861         }
 862
 863         return errstr;
 864 }
 865
 866 /* __ext4_std_error decodes expected errors from journaling functions
 867  * automatically and invokes the appropriate error response.  */
 868
 869 void __ext4_std_error(struct super_block *sb, const char *function,
 870                       unsigned int line, int errno)
 871 {
 872         char nbuf[16];
 873         const char *errstr;
 874
 875         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 876                 return;
 877
 878         /* Special case: if the error is EROFS, and we're not already
 879          * inside a transaction, then there's really no point in logging
 880          * an error. */
 881         if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb))
 882                 return;
 883
 884         if (ext4_error_ratelimit(sb)) {
 885                 errstr = ext4_decode_error(sb, errno, nbuf);
 886                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 887                        sb->s_id, function, line, errstr);
 888         }
 889
 890         ext4_handle_error(sb, false, -errno, 0, 0, function, line);
 891 }
 892
 893 void __ext4_msg(struct super_block *sb,
 894                 const char *prefix, const char *fmt, ...)
 895 {
 896         struct va_format vaf;
 897         va_list args;
 898
 899         atomic_inc(&EXT4_SB(sb)->s_msg_count);
 900         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 901                 return;
 902
 903         va_start(args, fmt);
 904         vaf.fmt = fmt;
 905         vaf.va = &args;
 906         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 907         va_end(args);
 908 }
 909
 910 static int ext4_warning_ratelimit(struct super_block *sb)
 911 {
 912         atomic_inc(&EXT4_SB(sb)->s_warning_count);
 913         return ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state),
 914                             "EXT4-fs warning");
 915 }
 916
 917 void __ext4_warning(struct super_block *sb, const char *function,
 918                     unsigned int line, const char *fmt, ...)
 919 {
 920         struct va_format vaf;
 921         va_list args;
 922
 923         if (!ext4_warning_ratelimit(sb))
 924                 return;
 925
 926         va_start(args, fmt);
 927         vaf.fmt = fmt;
 928         vaf.va = &args;
 929         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 930                sb->s_id, function, line, &vaf);
 931         va_end(args);
 932 }
 933
 934 void __ext4_warning_inode(const struct inode *inode, const char *function,
 935                           unsigned int line, const char *fmt, ...)
 936 {
 937         struct va_format vaf;
 938         va_list args;
 939
 940         if (!ext4_warning_ratelimit(inode->i_sb))
 941                 return;
 942
 943         va_start(args, fmt);
 944         vaf.fmt = fmt;
 945         vaf.va = &args;
 946         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 947                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 948                function, line, inode->i_ino, current->comm, &vaf);
 949         va_end(args);
 950 }
 951
 952 void __ext4_grp_locked_error(const char *function, unsigned int line,
 953                              struct super_block *sb, ext4_group_t grp,
 954                              unsigned long ino, ext4_fsblk_t block,
 955                              const char *fmt, ...)
 956 __releases(bitlock)
 957 __acquires(bitlock)
 958 {
 959         struct va_format vaf;
 960         va_list args;
 961
 962         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 963                 return;
 964
 965         trace_ext4_error(sb, function, line);
 966         if (ext4_error_ratelimit(sb)) {
 967                 va_start(args, fmt);
 968                 vaf.fmt = fmt;
 969                 vaf.va = &args;
 970                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 971                        sb->s_id, function, line, grp);
 972                 if (ino)
 973                         printk(KERN_CONT "inode %lu: ", ino);
 974                 if (block)
 975                         printk(KERN_CONT "block %llu:",
 976                                (unsigned long long) block);
 977                 printk(KERN_CONT "%pV\n", &vaf);
 978                 va_end(args);
 979         }
 980
 981         if (test_opt(sb, ERRORS_CONT)) {
 982                 if (test_opt(sb, WARN_ON_ERROR))
 983                         WARN_ON_ONCE(1);
 984                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 985                 if (!bdev_read_only(sb->s_bdev)) {
 986                         save_error_info(sb, EFSCORRUPTED, ino, block, function,
 987                                         line);
 988                         schedule_work(&EXT4_SB(sb)->s_error_work);
 989                 }
 990                 return;
 991         }
 992         ext4_unlock_group(sb, grp);
 993         ext4_handle_error(sb, false, EFSCORRUPTED, ino, block, function, line);
 994         /*
 995          * We only get here in the ERRORS_RO case; relocking the group
 996          * may be dangerous, but nothing bad will happen since the
 997          * filesystem will have already been marked read/only and the
 998          * journal has been aborted.  We return 1 as a hint to callers
 999          * who might what to use the return value from
1000          * ext4_grp_locked_error() to distinguish between the
1001          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
1002          * aggressively from the ext4 function in question, with a
1003          * more appropriate error code.
1004          */
1005         ext4_lock_group(sb, grp);
1006         return;
1007 }
1008
1009 void ext4_mark_group_bitmap_corrupted(struct super_block *sb,
1010                                      ext4_group_t group,
1011                                      unsigned int flags)
1012 {
1013         struct ext4_sb_info *sbi = EXT4_SB(sb);
1014         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1015         struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
1016         int ret;
1017
1018         if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) {
1019                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT,
1020                                             &grp->bb_state);
1021                 if (!ret)
1022                         percpu_counter_sub(&sbi->s_freeclusters_counter,
1023                                            grp->bb_free);
1024         }
1025
1026         if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) {
1027                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT,
1028                                             &grp->bb_state);
1029                 if (!ret && gdp) {
1030                         int count;
1031
1032                         count = ext4_free_inodes_count(sb, gdp);
1033                         percpu_counter_sub(&sbi->s_freeinodes_counter,
1034                                            count);
1035                 }
1036         }
1037 }
1038
1039 void ext4_update_dynamic_rev(struct super_block *sb)
1040 {
1041         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1042
1043         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
1044                 return;
1045
1046         ext4_warning(sb,
1047                      "updating to rev %d because of new feature flag, "
1048                      "running e2fsck is recommended",
1049                      EXT4_DYNAMIC_REV);
1050
1051         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
1052         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
1053         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
1054         /* leave es->s_feature_*compat flags alone */
1055         /* es->s_uuid will be set by e2fsck if empty */
1056
1057         /*
1058          * The rest of the superblock fields should be zero, and if not it
1059          * means they are likely already in use, so leave them alone.  We
1060          * can leave it up to e2fsck to clean up any inconsistencies there.
1061          */
1062 }
1063
1064 /*
1065  * Open the external journal device
1066  */
1067 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
1068 {
1069         struct block_device *bdev;
1070
1071         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
1072         if (IS_ERR(bdev))
1073                 goto fail;
1074         return bdev;
1075
1076 fail:
1077         ext4_msg(sb, KERN_ERR,
1078                  "failed to open journal device unknown-block(%u,%u) %ld",
1079                  MAJOR(dev), MINOR(dev), PTR_ERR(bdev));
1080         return NULL;
1081 }
1082
1083 /*
1084  * Release the journal device
1085  */
1086 static void ext4_blkdev_put(struct block_device *bdev)
1087 {
1088         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1089 }
1090
1091 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
1092 {
1093         struct block_device *bdev;
1094         bdev = sbi->s_journal_bdev;
1095         if (bdev) {
1096                 ext4_blkdev_put(bdev);
1097                 sbi->s_journal_bdev = NULL;
1098         }
1099 }
1100
1101 static inline struct inode *orphan_list_entry(struct list_head *l)
1102 {
1103         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
1104 }
1105
1106 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
1107 {
1108         struct list_head *l;
1109
1110         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
1111                  le32_to_cpu(sbi->s_es->s_last_orphan));
1112
1113         printk(KERN_ERR "sb_info orphan list:\n");
1114         list_for_each(l, &sbi->s_orphan) {
1115                 struct inode *inode = orphan_list_entry(l);
1116                 printk(KERN_ERR "  "
1117                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
1118                        inode->i_sb->s_id, inode->i_ino, inode,
1119                        inode->i_mode, inode->i_nlink,
1120                        NEXT_ORPHAN(inode));
1121         }
1122 }
1123
1124 #ifdef CONFIG_QUOTA
1125 static int ext4_quota_off(struct super_block *sb, int type);
1126
1127 static inline void ext4_quota_off_umount(struct super_block *sb)
1128 {
1129         int type;
1130
1131         /* Use our quota_off function to clear inode flags etc. */
1132         for (type = 0; type < EXT4_MAXQUOTAS; type++)
1133                 ext4_quota_off(sb, type);
1134 }
1135
1136 /*
1137  * This is a helper function which is used in the mount/remount
1138  * codepaths (which holds s_umount) to fetch the quota file name.
1139  */
1140 static inline char *get_qf_name(struct super_block *sb,
1141                                 struct ext4_sb_info *sbi,
1142                                 int type)
1143 {
1144         return rcu_dereference_protected(sbi->s_qf_names[type],
1145                                          lockdep_is_held(&sb->s_umount));
1146 }
1147 #else
1148 static inline void ext4_quota_off_umount(struct super_block *sb)
1149 {
1150 }
1151 #endif
1152
1153 static void ext4_put_super(struct super_block *sb)
1154 {
1155         struct ext4_sb_info *sbi = EXT4_SB(sb);
1156         struct ext4_super_block *es = sbi->s_es;
1157         struct buffer_head **group_desc;
1158         struct flex_groups **flex_groups;
1159         int aborted = 0;
1160         int i, err;
1161
1162         ext4_unregister_li_request(sb);
1163         ext4_quota_off_umount(sb);
1164
1165         flush_work(&sbi->s_error_work);
1166         destroy_workqueue(sbi->rsv_conversion_wq);
1167
1168         /*
1169          * Unregister sysfs before destroying jbd2 journal.
1170          * Since we could still access attr_journal_task attribute via sysfs
1171          * path which could have sbi->s_journal->j_task as NULL
1172          */
1173         ext4_unregister_sysfs(sb);
1174
1175         if (sbi->s_journal) {
1176                 aborted = is_journal_aborted(sbi->s_journal);
1177                 err = jbd2_journal_destroy(sbi->s_journal);
1178                 sbi->s_journal = NULL;
1179                 if ((err < 0) && !aborted) {
1180                         ext4_abort(sb, -err, "Couldn't clean up the journal");
1181                 }
1182         }
1183
1184         ext4_es_unregister_shrinker(sbi);
1185         del_timer_sync(&sbi->s_err_report);
1186         ext4_release_system_zone(sb);
1187         ext4_mb_release(sb);
1188         ext4_ext_release(sb);
1189
1190         if (!sb_rdonly(sb) && !aborted) {
1191                 ext4_clear_feature_journal_needs_recovery(sb);
1192                 es->s_state = cpu_to_le16(sbi->s_mount_state);
1193         }
1194         if (!sb_rdonly(sb))
1195                 ext4_commit_super(sb);
1196
1197         rcu_read_lock();
1198         group_desc = rcu_dereference(sbi->s_group_desc);
1199         for (i = 0; i < sbi->s_gdb_count; i++)
1200                 brelse(group_desc[i]);
1201         kvfree(group_desc);
1202         flex_groups = rcu_dereference(sbi->s_flex_groups);
1203         if (flex_groups) {
1204                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
1205                         kvfree(flex_groups[i]);
1206                 kvfree(flex_groups);
1207         }
1208         rcu_read_unlock();
1209         percpu_counter_destroy(&sbi->s_freeclusters_counter);
1210         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1211         percpu_counter_destroy(&sbi->s_dirs_counter);
1212         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
1213         percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit);
1214         percpu_free_rwsem(&sbi->s_writepages_rwsem);
1215 #ifdef CONFIG_QUOTA
1216         for (i = 0; i < EXT4_MAXQUOTAS; i++)
1217                 kfree(get_qf_name(sb, sbi, i));
1218 #endif
1219
1220         /* Debugging code just in case the in-memory inode orphan list
1221          * isn't empty.  The on-disk one can be non-empty if we've
1222          * detected an error and taken the fs readonly, but the
1223          * in-memory list had better be clean by this point. */
1224         if (!list_empty(&sbi->s_orphan))
1225                 dump_orphan_list(sb, sbi);
1226         ASSERT(list_empty(&sbi->s_orphan));
1227
1228         sync_blockdev(sb->s_bdev);
1229         invalidate_bdev(sb->s_bdev);
1230         if (sbi->s_journal_bdev && sbi->s_journal_bdev != sb->s_bdev) {
1231                 /*
1232                  * Invalidate the journal device's buffers.  We don't want them
1233                  * floating about in memory - the physical journal device may
1234                  * hotswapped, and it breaks the `ro-after' testing code.
1235                  */
1236                 sync_blockdev(sbi->s_journal_bdev);
1237                 invalidate_bdev(sbi->s_journal_bdev);
1238                 ext4_blkdev_remove(sbi);
1239         }
1240
1241         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
1242         sbi->s_ea_inode_cache = NULL;
1243
1244         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
1245         sbi->s_ea_block_cache = NULL;
1246
1247         if (sbi->s_mmp_tsk)
1248                 kthread_stop(sbi->s_mmp_tsk);
1249         brelse(sbi->s_sbh);
1250         sb->s_fs_info = NULL;
1251         /*
1252          * Now that we are completely done shutting down the
1253          * superblock, we need to actually destroy the kobject.
1254          */
1255         kobject_put(&sbi->s_kobj);
1256         wait_for_completion(&sbi->s_kobj_unregister);
1257         if (sbi->s_chksum_driver)
1258                 crypto_free_shash(sbi->s_chksum_driver);
1259         kfree(sbi->s_blockgroup_lock);
1260         fs_put_dax(sbi->s_daxdev);
1261         fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy);
1262 #ifdef CONFIG_UNICODE
1263         utf8_unload(sb->s_encoding);
1264 #endif
1265         kfree(sbi);
1266 }
1267
1268 static struct kmem_cache *ext4_inode_cachep;
1269
1270 /*
1271  * Called inside transaction, so use GFP_NOFS
1272  */
1273 static struct inode *ext4_alloc_inode(struct super_block *sb)
1274 {
1275         struct ext4_inode_info *ei;
1276
1277         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
1278         if (!ei)
1279                 return NULL;
1280
1281         inode_set_iversion(&ei->vfs_inode, 1);
1282         spin_lock_init(&ei->i_raw_lock);
1283         INIT_LIST_HEAD(&ei->i_prealloc_list);
1284         atomic_set(&ei->i_prealloc_active, 0);
1285         spin_lock_init(&ei->i_prealloc_lock);
1286         ext4_es_init_tree(&ei->i_es_tree);
1287         rwlock_init(&ei->i_es_lock);
1288         INIT_LIST_HEAD(&ei->i_es_list);
1289         ei->i_es_all_nr = 0;
1290         ei->i_es_shk_nr = 0;
1291         ei->i_es_shrink_lblk = 0;
1292         ei->i_reserved_data_blocks = 0;
1293         spin_lock_init(&(ei->i_block_reservation_lock));
1294         ext4_init_pending_tree(&ei->i_pending_tree);
1295 #ifdef CONFIG_QUOTA
1296         ei->i_reserved_quota = 0;
1297         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
1298 #endif
1299         ei->jinode = NULL;
1300         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
1301         spin_lock_init(&ei->i_completed_io_lock);
1302         ei->i_sync_tid = 0;
1303         ei->i_datasync_tid = 0;
1304         atomic_set(&ei->i_unwritten, 0);
1305         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
1306         ext4_fc_init_inode(&ei->vfs_inode);
1307         mutex_init(&ei->i_fc_lock);
1308         return &ei->vfs_inode;
1309 }
1310
1311 static int ext4_drop_inode(struct inode *inode)
1312 {
1313         int drop = generic_drop_inode(inode);
1314
1315         if (!drop)
1316                 drop = fscrypt_drop_inode(inode);
1317
1318         trace_ext4_drop_inode(inode, drop);
1319         return drop;
1320 }
1321
1322 static void ext4_free_in_core_inode(struct inode *inode)
1323 {
1324         fscrypt_free_inode(inode);
1325         if (!list_empty(&(EXT4_I(inode)->i_fc_list))) {
1326                 pr_warn("%s: inode %ld still in fc list",
1327                         __func__, inode->i_ino);
1328         }
1329         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1330 }
1331
1332 static void ext4_destroy_inode(struct inode *inode)
1333 {
1334         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1335                 ext4_msg(inode->i_sb, KERN_ERR,
1336                          "Inode %lu (%p): orphan list check failed!",
1337                          inode->i_ino, EXT4_I(inode));
1338                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1339                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1340                                 true);
1341                 dump_stack();
1342         }
1343 }
1344
1345 static void init_once(void *foo)
1346 {
1347         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1348
1349         INIT_LIST_HEAD(&ei->i_orphan);
1350         init_rwsem(&ei->xattr_sem);
1351         init_rwsem(&ei->i_data_sem);
1352         init_rwsem(&ei->i_mmap_sem);
1353         inode_init_once(&ei->vfs_inode);
1354         ext4_fc_init_inode(&ei->vfs_inode);
1355 }
1356
1357 static int __init init_inodecache(void)
1358 {
1359         ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache",
1360                                 sizeof(struct ext4_inode_info), 0,
1361                                 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
1362                                         SLAB_ACCOUNT),
1363                                 offsetof(struct ext4_inode_info, i_data),
1364                                 sizeof_field(struct ext4_inode_info, i_data),
1365                                 init_once);
1366         if (ext4_inode_cachep == NULL)
1367                 return -ENOMEM;
1368         return 0;
1369 }
1370
1371 static void destroy_inodecache(void)
1372 {
1373         /*
1374          * Make sure all delayed rcu free inodes are flushed before we
1375          * destroy cache.
1376          */
1377         rcu_barrier();
1378         kmem_cache_destroy(ext4_inode_cachep);
1379 }
1380
1381 void ext4_clear_inode(struct inode *inode)
1382 {
1383         ext4_fc_del(inode);
1384         invalidate_inode_buffers(inode);
1385         clear_inode(inode);
1386         ext4_discard_preallocations(inode, 0);
1387         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1388         dquot_drop(inode);
1389         if (EXT4_I(inode)->jinode) {
1390                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1391                                                EXT4_I(inode)->jinode);
1392                 jbd2_free_inode(EXT4_I(inode)->jinode);
1393                 EXT4_I(inode)->jinode = NULL;
1394         }
1395         fscrypt_put_encryption_info(inode);
1396         fsverity_cleanup_inode(inode);
1397 }
1398
1399 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1400                                         u64 ino, u32 generation)
1401 {
1402         struct inode *inode;
1403
1404         /*
1405          * Currently we don't know the generation for parent directory, so
1406          * a generation of 0 means "accept any"
1407          */
1408         inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE);
1409         if (IS_ERR(inode))
1410                 return ERR_CAST(inode);
1411         if (generation && inode->i_generation != generation) {
1412                 iput(inode);
1413                 return ERR_PTR(-ESTALE);
1414         }
1415
1416         return inode;
1417 }
1418
1419 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1420                                         int fh_len, int fh_type)
1421 {
1422         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1423                                     ext4_nfs_get_inode);
1424 }
1425
1426 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1427                                         int fh_len, int fh_type)
1428 {
1429         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1430                                     ext4_nfs_get_inode);
1431 }
1432
1433 static int ext4_nfs_commit_metadata(struct inode *inode)
1434 {
1435         struct writeback_control wbc = {
1436                 .sync_mode = WB_SYNC_ALL
1437         };
1438
1439         trace_ext4_nfs_commit_metadata(inode);
1440         return ext4_write_inode(inode, &wbc);
1441 }
1442
1443 /*
1444  * Try to release metadata pages (indirect blocks, directories) which are
1445  * mapped via the block device.  Since these pages could have journal heads
1446  * which would prevent try_to_free_buffers() from freeing them, we must use
1447  * jbd2 layer's try_to_free_buffers() function to release them.
1448  */
1449 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1450                                  gfp_t wait)
1451 {
1452         journal_t *journal = EXT4_SB(sb)->s_journal;
1453
1454         WARN_ON(PageChecked(page));
1455         if (!page_has_buffers(page))
1456                 return 0;
1457         if (journal)
1458                 return jbd2_journal_try_to_free_buffers(journal, page);
1459
1460         return try_to_free_buffers(page);
1461 }
1462
1463 #ifdef CONFIG_FS_ENCRYPTION
1464 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1465 {
1466         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1467                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1468 }
1469
1470 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1471                                                         void *fs_data)
1472 {
1473         handle_t *handle = fs_data;
1474         int res, res2, credits, retries = 0;
1475
1476         /*
1477          * Encrypting the root directory is not allowed because e2fsck expects
1478          * lost+found to exist and be unencrypted, and encrypting the root
1479          * directory would imply encrypting the lost+found directory as well as
1480          * the filename "lost+found" itself.
1481          */
1482         if (inode->i_ino == EXT4_ROOT_INO)
1483                 return -EPERM;
1484
1485         if (WARN_ON_ONCE(IS_DAX(inode) && i_size_read(inode)))
1486                 return -EINVAL;
1487
1488         if (ext4_test_inode_flag(inode, EXT4_INODE_DAX))
1489                 return -EOPNOTSUPP;
1490
1491         res = ext4_convert_inline_data(inode);
1492         if (res)
1493                 return res;
1494
1495         /*
1496          * If a journal handle was specified, then the encryption context is
1497          * being set on a new inode via inheritance and is part of a larger
1498          * transaction to create the inode.  Otherwise the encryption context is
1499          * being set on an existing inode in its own transaction.  Only in the
1500          * latter case should the "retry on ENOSPC" logic be used.
1501          */
1502
1503         if (handle) {
1504                 res = ext4_xattr_set_handle(handle, inode,
1505                                             EXT4_XATTR_INDEX_ENCRYPTION,
1506                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1507                                             ctx, len, 0);
1508                 if (!res) {
1509                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1510                         ext4_clear_inode_state(inode,
1511                                         EXT4_STATE_MAY_INLINE_DATA);
1512                         /*
1513                          * Update inode->i_flags - S_ENCRYPTED will be enabled,
1514                          * S_DAX may be disabled
1515                          */
1516                         ext4_set_inode_flags(inode, false);
1517                 }
1518                 return res;
1519         }
1520
1521         res = dquot_initialize(inode);
1522         if (res)
1523                 return res;
1524 retry:
1525         res = ext4_xattr_set_credits(inode, len, false /* is_create */,
1526                                      &credits);
1527         if (res)
1528                 return res;
1529
1530         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
1531         if (IS_ERR(handle))
1532                 return PTR_ERR(handle);
1533
1534         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1535                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1536                                     ctx, len, 0);
1537         if (!res) {
1538                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1539                 /*
1540                  * Update inode->i_flags - S_ENCRYPTED will be enabled,
1541                  * S_DAX may be disabled
1542                  */
1543                 ext4_set_inode_flags(inode, false);
1544                 res = ext4_mark_inode_dirty(handle, inode);
1545                 if (res)
1546                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1547         }
1548         res2 = ext4_journal_stop(handle);
1549
1550         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1551                 goto retry;
1552         if (!res)
1553                 res = res2;
1554         return res;
1555 }
1556
1557 static const union fscrypt_policy *ext4_get_dummy_policy(struct super_block *sb)
1558 {
1559         return EXT4_SB(sb)->s_dummy_enc_policy.policy;
1560 }
1561
1562 static bool ext4_has_stable_inodes(struct super_block *sb)
1563 {
1564         return ext4_has_feature_stable_inodes(sb);
1565 }
1566
1567 static void ext4_get_ino_and_lblk_bits(struct super_block *sb,
1568                                        int *ino_bits_ret, int *lblk_bits_ret)
1569 {
1570         *ino_bits_ret = 8 * sizeof(EXT4_SB(sb)->s_es->s_inodes_count);
1571         *lblk_bits_ret = 8 * sizeof(ext4_lblk_t);
1572 }
1573
1574 static const struct fscrypt_operations ext4_cryptops = {
1575         .key_prefix             = "ext4:",
1576         .get_context            = ext4_get_context,
1577         .set_context            = ext4_set_context,
1578         .get_dummy_policy       = ext4_get_dummy_policy,
1579         .empty_dir              = ext4_empty_dir,
1580         .max_namelen            = EXT4_NAME_LEN,
1581         .has_stable_inodes      = ext4_has_stable_inodes,
1582         .get_ino_and_lblk_bits  = ext4_get_ino_and_lblk_bits,
1583 };
1584 #endif
1585
1586 #ifdef CONFIG_QUOTA
1587 static const char * const quotatypes[] = INITQFNAMES;
1588 #define QTYPE2NAME(t) (quotatypes[t])
1589
1590 static int ext4_write_dquot(struct dquot *dquot);
1591 static int ext4_acquire_dquot(struct dquot *dquot);
1592 static int ext4_release_dquot(struct dquot *dquot);
1593 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1594 static int ext4_write_info(struct super_block *sb, int type);
1595 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1596                          const struct path *path);
1597 static int ext4_quota_on_mount(struct super_block *sb, int type);
1598 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1599                                size_t len, loff_t off);
1600 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1601                                 const char *data, size_t len, loff_t off);
1602 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1603                              unsigned int flags);
1604 static int ext4_enable_quotas(struct super_block *sb);
1605
1606 static struct dquot **ext4_get_dquots(struct inode *inode)
1607 {
1608         return EXT4_I(inode)->i_dquot;
1609 }
1610
1611 static const struct dquot_operations ext4_quota_operations = {
1612         .get_reserved_space     = ext4_get_reserved_space,
1613         .write_dquot            = ext4_write_dquot,
1614         .acquire_dquot          = ext4_acquire_dquot,
1615         .release_dquot          = ext4_release_dquot,
1616         .mark_dirty             = ext4_mark_dquot_dirty,
1617         .write_info             = ext4_write_info,
1618         .alloc_dquot            = dquot_alloc,
1619         .destroy_dquot          = dquot_destroy,
1620         .get_projid             = ext4_get_projid,
1621         .get_inode_usage        = ext4_get_inode_usage,
1622         .get_next_id            = dquot_get_next_id,
1623 };
1624
1625 static const struct quotactl_ops ext4_qctl_operations = {
1626         .quota_on       = ext4_quota_on,
1627         .quota_off      = ext4_quota_off,
1628         .quota_sync     = dquot_quota_sync,
1629         .get_state      = dquot_get_state,
1630         .set_info       = dquot_set_dqinfo,
1631         .get_dqblk      = dquot_get_dqblk,
1632         .set_dqblk      = dquot_set_dqblk,
1633         .get_nextdqblk  = dquot_get_next_dqblk,
1634 };
1635 #endif
1636
1637 static const struct super_operations ext4_sops = {
1638         .alloc_inode    = ext4_alloc_inode,
1639         .free_inode     = ext4_free_in_core_inode,
1640         .destroy_inode  = ext4_destroy_inode,
1641         .write_inode    = ext4_write_inode,
1642         .dirty_inode    = ext4_dirty_inode,
1643         .drop_inode     = ext4_drop_inode,
1644         .evict_inode    = ext4_evict_inode,
1645         .put_super      = ext4_put_super,
1646         .sync_fs        = ext4_sync_fs,
1647         .freeze_fs      = ext4_freeze,
1648         .unfreeze_fs    = ext4_unfreeze,
1649         .statfs         = ext4_statfs,
1650         .remount_fs     = ext4_remount,
1651         .show_options   = ext4_show_options,
1652 #ifdef CONFIG_QUOTA
1653         .quota_read     = ext4_quota_read,
1654         .quota_write    = ext4_quota_write,
1655         .get_dquots     = ext4_get_dquots,
1656 #endif
1657         .bdev_try_to_free_page = bdev_try_to_free_page,
1658 };
1659
1660 static const struct export_operations ext4_export_ops = {
1661         .fh_to_dentry = ext4_fh_to_dentry,
1662         .fh_to_parent = ext4_fh_to_parent,
1663         .get_parent = ext4_get_parent,
1664         .commit_metadata = ext4_nfs_commit_metadata,
1665 };
1666
1667 enum {
1668         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1669         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1670         Opt_nouid32, Opt_debug, Opt_removed,
1671         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1672         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1673         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1674         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1675         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1676         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1677         Opt_inlinecrypt,
1678         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1679         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1680         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1681         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version,
1682         Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never,
1683         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
1684         Opt_nowarn_on_error, Opt_mblk_io_submit,
1685         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1686         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1687         Opt_inode_readahead_blks, Opt_journal_ioprio,
1688         Opt_dioread_nolock, Opt_dioread_lock,
1689         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1690         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1691         Opt_prefetch_block_bitmaps,
1692 #ifdef CONFIG_EXT4_DEBUG
1693         Opt_fc_debug_max_replay, Opt_fc_debug_force
1694 #endif
1695 };
1696
1697 static const match_table_t tokens = {
1698         {Opt_bsd_df, "bsddf"},
1699         {Opt_minix_df, "minixdf"},
1700         {Opt_grpid, "grpid"},
1701         {Opt_grpid, "bsdgroups"},
1702         {Opt_nogrpid, "nogrpid"},
1703         {Opt_nogrpid, "sysvgroups"},
1704         {Opt_resgid, "resgid=%u"},
1705         {Opt_resuid, "resuid=%u"},
1706         {Opt_sb, "sb=%u"},
1707         {Opt_err_cont, "errors=continue"},
1708         {Opt_err_panic, "errors=panic"},
1709         {Opt_err_ro, "errors=remount-ro"},
1710         {Opt_nouid32, "nouid32"},
1711         {Opt_debug, "debug"},
1712         {Opt_removed, "oldalloc"},
1713         {Opt_removed, "orlov"},
1714         {Opt_user_xattr, "user_xattr"},
1715         {Opt_nouser_xattr, "nouser_xattr"},
1716         {Opt_acl, "acl"},
1717         {Opt_noacl, "noacl"},
1718         {Opt_noload, "norecovery"},
1719         {Opt_noload, "noload"},
1720         {Opt_removed, "nobh"},
1721         {Opt_removed, "bh"},
1722         {Opt_commit, "commit=%u"},
1723         {Opt_min_batch_time, "min_batch_time=%u"},
1724         {Opt_max_batch_time, "max_batch_time=%u"},
1725         {Opt_journal_dev, "journal_dev=%u"},
1726         {Opt_journal_path, "journal_path=%s"},
1727         {Opt_journal_checksum, "journal_checksum"},
1728         {Opt_nojournal_checksum, "nojournal_checksum"},
1729         {Opt_journal_async_commit, "journal_async_commit"},
1730         {Opt_abort, "abort"},
1731         {Opt_data_journal, "data=journal"},
1732         {Opt_data_ordered, "data=ordered"},
1733         {Opt_data_writeback, "data=writeback"},
1734         {Opt_data_err_abort, "data_err=abort"},
1735         {Opt_data_err_ignore, "data_err=ignore"},
1736         {Opt_offusrjquota, "usrjquota="},
1737         {Opt_usrjquota, "usrjquota=%s"},
1738         {Opt_offgrpjquota, "grpjquota="},
1739         {Opt_grpjquota, "grpjquota=%s"},
1740         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1741         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1742         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1743         {Opt_grpquota, "grpquota"},
1744         {Opt_noquota, "noquota"},
1745         {Opt_quota, "quota"},
1746         {Opt_usrquota, "usrquota"},
1747         {Opt_prjquota, "prjquota"},
1748         {Opt_barrier, "barrier=%u"},
1749         {Opt_barrier, "barrier"},
1750         {Opt_nobarrier, "nobarrier"},
1751         {Opt_i_version, "i_version"},
1752         {Opt_dax, "dax"},
1753         {Opt_dax_always, "dax=always"},
1754         {Opt_dax_inode, "dax=inode"},
1755         {Opt_dax_never, "dax=never"},
1756         {Opt_stripe, "stripe=%u"},
1757         {Opt_delalloc, "delalloc"},
1758         {Opt_warn_on_error, "warn_on_error"},
1759         {Opt_nowarn_on_error, "nowarn_on_error"},
1760         {Opt_lazytime, "lazytime"},
1761         {Opt_nolazytime, "nolazytime"},
1762         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1763         {Opt_nodelalloc, "nodelalloc"},
1764         {Opt_removed, "mblk_io_submit"},
1765         {Opt_removed, "nomblk_io_submit"},
1766         {Opt_block_validity, "block_validity"},
1767         {Opt_noblock_validity, "noblock_validity"},
1768         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1769         {Opt_journal_ioprio, "journal_ioprio=%u"},
1770         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1771         {Opt_auto_da_alloc, "auto_da_alloc"},
1772         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1773         {Opt_dioread_nolock, "dioread_nolock"},
1774         {Opt_dioread_lock, "nodioread_nolock"},
1775         {Opt_dioread_lock, "dioread_lock"},
1776         {Opt_discard, "discard"},
1777         {Opt_nodiscard, "nodiscard"},
1778         {Opt_init_itable, "init_itable=%u"},
1779         {Opt_init_itable, "init_itable"},
1780         {Opt_noinit_itable, "noinit_itable"},
1781 #ifdef CONFIG_EXT4_DEBUG
1782         {Opt_fc_debug_force, "fc_debug_force"},
1783         {Opt_fc_debug_max_replay, "fc_debug_max_replay=%u"},
1784 #endif
1785         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1786         {Opt_test_dummy_encryption, "test_dummy_encryption=%s"},
1787         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1788         {Opt_inlinecrypt, "inlinecrypt"},
1789         {Opt_nombcache, "nombcache"},
1790         {Opt_nombcache, "no_mbcache"},  /* for backward compatibility */
1791         {Opt_prefetch_block_bitmaps, "prefetch_block_bitmaps"},
1792         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1793         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1794         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1795         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1796         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1797         {Opt_err, NULL},
1798 };
1799
1800 static ext4_fsblk_t get_sb_block(void **data)
1801 {
1802         ext4_fsblk_t    sb_block;
1803         char            *options = (char *) *data;
1804
1805         if (!options || strncmp(options, "sb=", 3) != 0)
1806                 return 1;       /* Default location */
1807
1808         options += 3;
1809         /* TODO: use simple_strtoll with >32bit ext4 */
1810         sb_block = simple_strtoul(options, &options, 0);
1811         if (*options && *options != ',') {
1812                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1813                        (char *) *data);
1814                 return 1;
1815         }
1816         if (*options == ',')
1817                 options++;
1818         *data = (void *) options;
1819
1820         return sb_block;
1821 }
1822
1823 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1824 static const char deprecated_msg[] =
1825         "Mount option \"%s\" will be removed by %s\n"
1826         "Contact [email protected] if you think we should keep it.\n";
1827
1828 #ifdef CONFIG_QUOTA
1829 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1830 {
1831         struct ext4_sb_info *sbi = EXT4_SB(sb);
1832         char *qname, *old_qname = get_qf_name(sb, sbi, qtype);
1833         int ret = -1;
1834
1835         if (sb_any_quota_loaded(sb) && !old_qname) {
1836                 ext4_msg(sb, KERN_ERR,
1837                         "Cannot change journaled "
1838                         "quota options when quota turned on");
1839                 return -1;
1840         }
1841         if (ext4_has_feature_quota(sb)) {
1842                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1843                          "ignored when QUOTA feature is enabled");
1844                 return 1;
1845         }
1846         qname = match_strdup(args);
1847         if (!qname) {
1848                 ext4_msg(sb, KERN_ERR,
1849                         "Not enough memory for storing quotafile name");
1850                 return -1;
1851         }
1852         if (old_qname) {
1853                 if (strcmp(old_qname, qname) == 0)
1854                         ret = 1;
1855                 else
1856                         ext4_msg(sb, KERN_ERR,
1857                                  "%s quota file already specified",
1858                                  QTYPE2NAME(qtype));
1859                 goto errout;
1860         }
1861         if (strchr(qname, '/')) {
1862                 ext4_msg(sb, KERN_ERR,
1863                         "quotafile must be on filesystem root");
1864                 goto errout;
1865         }
1866         rcu_assign_pointer(sbi->s_qf_names[qtype], qname);
1867         set_opt(sb, QUOTA);
1868         return 1;
1869 errout:
1870         kfree(qname);
1871         return ret;
1872 }
1873
1874 static int clear_qf_name(struct super_block *sb, int qtype)
1875 {
1876
1877         struct ext4_sb_info *sbi = EXT4_SB(sb);
1878         char *old_qname = get_qf_name(sb, sbi, qtype);
1879
1880         if (sb_any_quota_loaded(sb) && old_qname) {
1881                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1882                         " when quota turned on");
1883                 return -1;
1884         }
1885         rcu_assign_pointer(sbi->s_qf_names[qtype], NULL);
1886         synchronize_rcu();
1887         kfree(old_qname);
1888         return 1;
1889 }
1890 #endif
1891
1892 #define MOPT_SET        0x0001
1893 #define MOPT_CLEAR      0x0002
1894 #define MOPT_NOSUPPORT  0x0004
1895 #define MOPT_EXPLICIT   0x0008
1896 #define MOPT_CLEAR_ERR  0x0010
1897 #define MOPT_GTE0       0x0020
1898 #ifdef CONFIG_QUOTA
1899 #define MOPT_Q          0
1900 #define MOPT_QFMT       0x0040
1901 #else
1902 #define MOPT_Q          MOPT_NOSUPPORT
1903 #define MOPT_QFMT       MOPT_NOSUPPORT
1904 #endif
1905 #define MOPT_DATAJ      0x0080
1906 #define MOPT_NO_EXT2    0x0100
1907 #define MOPT_NO_EXT3    0x0200
1908 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1909 #define MOPT_STRING     0x0400
1910 #define MOPT_SKIP       0x0800
1911 #define MOPT_2          0x1000
1912
1913 static const struct mount_opts {
1914         int     token;
1915         int     mount_opt;
1916         int     flags;
1917 } ext4_mount_opts[] = {
1918         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1919         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1920         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1921         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1922         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1923         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1924         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1925          MOPT_EXT4_ONLY | MOPT_SET},
1926         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1927          MOPT_EXT4_ONLY | MOPT_CLEAR},
1928         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1929         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1930         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1931          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1932         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1933          MOPT_EXT4_ONLY | MOPT_CLEAR},
1934         {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET},
1935         {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR},
1936         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1937          MOPT_EXT4_ONLY | MOPT_CLEAR},
1938         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1939          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1940         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1941                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1942          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1943         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1944         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1945         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1946         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1947         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1948          MOPT_NO_EXT2},
1949         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1950          MOPT_NO_EXT2},
1951         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1952         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1953         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1954         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1955         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1956         {Opt_commit, 0, MOPT_GTE0},
1957         {Opt_max_batch_time, 0, MOPT_GTE0},
1958         {Opt_min_batch_time, 0, MOPT_GTE0},
1959         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1960         {Opt_init_itable, 0, MOPT_GTE0},
1961         {Opt_dax, EXT4_MOUNT_DAX_ALWAYS, MOPT_SET | MOPT_SKIP},
1962         {Opt_dax_always, EXT4_MOUNT_DAX_ALWAYS,
1963                 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP},
1964         {Opt_dax_inode, EXT4_MOUNT2_DAX_INODE,
1965                 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP},
1966         {Opt_dax_never, EXT4_MOUNT2_DAX_NEVER,
1967                 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP},
1968         {Opt_stripe, 0, MOPT_GTE0},
1969         {Opt_resuid, 0, MOPT_GTE0},
1970         {Opt_resgid, 0, MOPT_GTE0},
1971         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1972         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1973         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1974         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1975         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1976         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1977          MOPT_NO_EXT2 | MOPT_DATAJ},
1978         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1979         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1980 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1981         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1982         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1983 #else
1984         {Opt_acl, 0, MOPT_NOSUPPORT},
1985         {Opt_noacl, 0, MOPT_NOSUPPORT},
1986 #endif
1987         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1988         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1989         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1990         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1991         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1992                                                         MOPT_SET | MOPT_Q},
1993         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1994                                                         MOPT_SET | MOPT_Q},
1995         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1996                                                         MOPT_SET | MOPT_Q},
1997         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1998                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1999                                                         MOPT_CLEAR | MOPT_Q},
2000         {Opt_usrjquota, 0, MOPT_Q | MOPT_STRING},
2001         {Opt_grpjquota, 0, MOPT_Q | MOPT_STRING},
2002         {Opt_offusrjquota, 0, MOPT_Q},
2003         {Opt_offgrpjquota, 0, MOPT_Q},
2004         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
2005         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
2006         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
2007         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
2008         {Opt_test_dummy_encryption, 0, MOPT_STRING},
2009         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
2010         {Opt_prefetch_block_bitmaps, EXT4_MOUNT_PREFETCH_BLOCK_BITMAPS,
2011          MOPT_SET},
2012 #ifdef CONFIG_EXT4_DEBUG
2013         {Opt_fc_debug_force, EXT4_MOUNT2_JOURNAL_FAST_COMMIT,
2014          MOPT_SET | MOPT_2 | MOPT_EXT4_ONLY},
2015         {Opt_fc_debug_max_replay, 0, MOPT_GTE0},
2016 #endif
2017         {Opt_err, 0, 0}
2018 };
2019
2020 #ifdef CONFIG_UNICODE
2021 static const struct ext4_sb_encodings {
2022         __u16 magic;
2023         char *name;
2024         char *version;
2025 } ext4_sb_encoding_map[] = {
2026         {EXT4_ENC_UTF8_12_1, "utf8", "12.1.0"},
2027 };
2028
2029 static int ext4_sb_read_encoding(const struct ext4_super_block *es,
2030                                  const struct ext4_sb_encodings **encoding,
2031                                  __u16 *flags)
2032 {
2033         __u16 magic = le16_to_cpu(es->s_encoding);
2034         int i;
2035
2036         for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++)
2037                 if (magic == ext4_sb_encoding_map[i].magic)
2038                         break;
2039
2040         if (i >= ARRAY_SIZE(ext4_sb_encoding_map))
2041                 return -EINVAL;
2042
2043         *encoding = &ext4_sb_encoding_map[i];
2044         *flags = le16_to_cpu(es->s_encoding_flags);
2045
2046         return 0;
2047 }
2048 #endif
2049
2050 static int ext4_set_test_dummy_encryption(struct super_block *sb,
2051                                           const char *opt,
2052                                           const substring_t *arg,
2053                                           bool is_remount)
2054 {
2055 #ifdef CONFIG_FS_ENCRYPTION
2056         struct ext4_sb_info *sbi = EXT4_SB(sb);
2057         int err;
2058
2059         /*
2060          * This mount option is just for testing, and it's not worthwhile to
2061          * implement the extra complexity (e.g. RCU protection) that would be
2062          * needed to allow it to be set or changed during remount.  We do allow
2063          * it to be specified during remount, but only if there is no change.
2064          */
2065         if (is_remount && !sbi->s_dummy_enc_policy.policy) {
2066                 ext4_msg(sb, KERN_WARNING,
2067                          "Can't set test_dummy_encryption on remount");
2068                 return -1;
2069         }
2070         err = fscrypt_set_test_dummy_encryption(sb, arg->from,
2071                                                 &sbi->s_dummy_enc_policy);
2072         if (err) {
2073                 if (err == -EEXIST)
2074                         ext4_msg(sb, KERN_WARNING,
2075                                  "Can't change test_dummy_encryption on remount");
2076                 else if (err == -EINVAL)
2077                         ext4_msg(sb, KERN_WARNING,
2078                                  "Value of option \"%s\" is unrecognized", opt);
2079                 else
2080                         ext4_msg(sb, KERN_WARNING,
2081                                  "Error processing option \"%s\" [%d]",
2082                                  opt, err);
2083                 return -1;
2084         }
2085         ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled");
2086 #else
2087         ext4_msg(sb, KERN_WARNING,
2088                  "Test dummy encryption mount option ignored");
2089 #endif
2090         return 1;
2091 }
2092
2093 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
2094                             substring_t *args, unsigned long *journal_devnum,
2095                             unsigned int *journal_ioprio, int is_remount)
2096 {
2097         struct ext4_sb_info *sbi = EXT4_SB(sb);
2098         const struct mount_opts *m;
2099         kuid_t uid;
2100         kgid_t gid;
2101         int arg = 0;
2102
2103 #ifdef CONFIG_QUOTA
2104         if (token == Opt_usrjquota)
2105                 return set_qf_name(sb, USRQUOTA, &args[0]);
2106         else if (token == Opt_grpjquota)
2107                 return set_qf_name(sb, GRPQUOTA, &args[0]);
2108         else if (token == Opt_offusrjquota)
2109                 return clear_qf_name(sb, USRQUOTA);
2110         else if (token == Opt_offgrpjquota)
2111                 return clear_qf_name(sb, GRPQUOTA);
2112 #endif
2113         switch (token) {
2114         case Opt_noacl:
2115         case Opt_nouser_xattr:
2116                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
2117                 break;
2118         case Opt_sb:
2119                 return 1;       /* handled by get_sb_block() */
2120         case Opt_removed:
2121                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
2122                 return 1;
2123         case Opt_abort:
2124                 ext4_set_mount_flag(sb, EXT4_MF_FS_ABORTED);
2125                 return 1;
2126         case Opt_i_version:
2127                 sb->s_flags |= SB_I_VERSION;
2128                 return 1;
2129         case Opt_lazytime:
2130                 sb->s_flags |= SB_LAZYTIME;
2131                 return 1;
2132         case Opt_nolazytime:
2133                 sb->s_flags &= ~SB_LAZYTIME;
2134                 return 1;
2135         case Opt_inlinecrypt:
2136 #ifdef CONFIG_FS_ENCRYPTION_INLINE_CRYPT
2137                 sb->s_flags |= SB_INLINECRYPT;
2138 #else
2139                 ext4_msg(sb, KERN_ERR, "inline encryption not supported");
2140 #endif
2141                 return 1;
2142         }
2143
2144         for (m = ext4_mount_opts; m->token != Opt_err; m++)
2145                 if (token == m->token)
2146                         break;
2147
2148         if (m->token == Opt_err) {
2149                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
2150                          "or missing value", opt);
2151                 return -1;
2152         }
2153
2154         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
2155                 ext4_msg(sb, KERN_ERR,
2156                          "Mount option \"%s\" incompatible with ext2", opt);
2157                 return -1;
2158         }
2159         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
2160                 ext4_msg(sb, KERN_ERR,
2161                          "Mount option \"%s\" incompatible with ext3", opt);
2162                 return -1;
2163         }
2164
2165         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
2166                 return -1;
2167         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
2168                 return -1;
2169         if (m->flags & MOPT_EXPLICIT) {
2170                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
2171                         set_opt2(sb, EXPLICIT_DELALLOC);
2172                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
2173                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
2174                 } else
2175                         return -1;
2176         }
2177         if (m->flags & MOPT_CLEAR_ERR)
2178                 clear_opt(sb, ERRORS_MASK);
2179         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
2180                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
2181                          "options when quota turned on");
2182                 return -1;
2183         }
2184
2185         if (m->flags & MOPT_NOSUPPORT) {
2186                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
2187         } else if (token == Opt_commit) {
2188                 if (arg == 0)
2189                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
2190                 else if (arg > INT_MAX / HZ) {
2191                         ext4_msg(sb, KERN_ERR,
2192                                  "Invalid commit interval %d, "
2193                                  "must be smaller than %d",
2194                                  arg, INT_MAX / HZ);
2195                         return -1;
2196                 }
2197                 sbi->s_commit_interval = HZ * arg;
2198         } else if (token == Opt_debug_want_extra_isize) {
2199                 if ((arg & 1) ||
2200                     (arg < 4) ||
2201                     (arg > (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE))) {
2202                         ext4_msg(sb, KERN_ERR,
2203                                  "Invalid want_extra_isize %d", arg);
2204                         return -1;
2205                 }
2206                 sbi->s_want_extra_isize = arg;
2207         } else if (token == Opt_max_batch_time) {
2208                 sbi->s_max_batch_time = arg;
2209         } else if (token == Opt_min_batch_time) {
2210                 sbi->s_min_batch_time = arg;
2211         } else if (token == Opt_inode_readahead_blks) {
2212                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
2213                         ext4_msg(sb, KERN_ERR,
2214                                  "EXT4-fs: inode_readahead_blks must be "
2215                                  "0 or a power of 2 smaller than 2^31");
2216                         return -1;
2217                 }
2218                 sbi->s_inode_readahead_blks = arg;
2219         } else if (token == Opt_init_itable) {
2220                 set_opt(sb, INIT_INODE_TABLE);
2221                 if (!args->from)
2222                         arg = EXT4_DEF_LI_WAIT_MULT;
2223                 sbi->s_li_wait_mult = arg;
2224         } else if (token == Opt_max_dir_size_kb) {
2225                 sbi->s_max_dir_size_kb = arg;
2226 #ifdef CONFIG_EXT4_DEBUG
2227         } else if (token == Opt_fc_debug_max_replay) {
2228                 sbi->s_fc_debug_max_replay = arg;
2229 #endif
2230         } else if (token == Opt_stripe) {
2231                 sbi->s_stripe = arg;
2232         } else if (token == Opt_resuid) {
2233                 uid = make_kuid(current_user_ns(), arg);
2234                 if (!uid_valid(uid)) {
2235                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
2236                         return -1;
2237                 }
2238                 sbi->s_resuid = uid;
2239         } else if (token == Opt_resgid) {
2240                 gid = make_kgid(current_user_ns(), arg);
2241                 if (!gid_valid(gid)) {
2242                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
2243                         return -1;
2244                 }
2245                 sbi->s_resgid = gid;
2246         } else if (token == Opt_journal_dev) {
2247                 if (is_remount) {
2248                         ext4_msg(sb, KERN_ERR,
2249                                  "Cannot specify journal on remount");
2250                         return -1;
2251                 }
2252                 *journal_devnum = arg;
2253         } else if (token == Opt_journal_path) {
2254                 char *journal_path;
2255                 struct inode *journal_inode;
2256                 struct path path;
2257                 int error;
2258
2259                 if (is_remount) {
2260                         ext4_msg(sb, KERN_ERR,
2261                                  "Cannot specify journal on remount");
2262                         return -1;
2263                 }
2264                 journal_path = match_strdup(&args[0]);
2265                 if (!journal_path) {
2266                         ext4_msg(sb, KERN_ERR, "error: could not dup "
2267                                 "journal device string");
2268                         return -1;
2269                 }
2270
2271                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
2272                 if (error) {
2273                         ext4_msg(sb, KERN_ERR, "error: could not find "
2274                                 "journal device path: error %d", error);
2275                         kfree(journal_path);
2276                         return -1;
2277                 }
2278
2279                 journal_inode = d_inode(path.dentry);
2280                 if (!S_ISBLK(journal_inode->i_mode)) {
2281                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
2282                                 "is not a block device", journal_path);
2283                         path_put(&path);
2284                         kfree(journal_path);
2285                         return -1;
2286                 }
2287
2288                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
2289                 path_put(&path);
2290                 kfree(journal_path);
2291         } else if (token == Opt_journal_ioprio) {
2292                 if (arg > 7) {
2293                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
2294                                  " (must be 0-7)");
2295                         return -1;
2296                 }
2297                 *journal_ioprio =
2298                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
2299         } else if (token == Opt_test_dummy_encryption) {
2300                 return ext4_set_test_dummy_encryption(sb, opt, &args[0],
2301                                                       is_remount);
2302         } else if (m->flags & MOPT_DATAJ) {
2303                 if (is_remount) {
2304                         if (!sbi->s_journal)
2305                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
2306                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
2307                                 ext4_msg(sb, KERN_ERR,
2308                                          "Cannot change data mode on remount");
2309                                 return -1;
2310                         }
2311                 } else {
2312                         clear_opt(sb, DATA_FLAGS);
2313                         sbi->s_mount_opt |= m->mount_opt;
2314                 }
2315 #ifdef CONFIG_QUOTA
2316         } else if (m->flags & MOPT_QFMT) {
2317                 if (sb_any_quota_loaded(sb) &&
2318                     sbi->s_jquota_fmt != m->mount_opt) {
2319                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
2320                                  "quota options when quota turned on");
2321                         return -1;
2322                 }
2323                 if (ext4_has_feature_quota(sb)) {
2324                         ext4_msg(sb, KERN_INFO,
2325                                  "Quota format mount options ignored "
2326                                  "when QUOTA feature is enabled");
2327                         return 1;
2328                 }
2329                 sbi->s_jquota_fmt = m->mount_opt;
2330 #endif
2331         } else if (token == Opt_dax || token == Opt_dax_always ||
2332                    token == Opt_dax_inode || token == Opt_dax_never) {
2333 #ifdef CONFIG_FS_DAX
2334                 switch (token) {
2335                 case Opt_dax:
2336                 case Opt_dax_always:
2337                         if (is_remount &&
2338                             (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2339                              (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) {
2340                         fail_dax_change_remount:
2341                                 ext4_msg(sb, KERN_ERR, "can't change "
2342                                          "dax mount option while remounting");
2343                                 return -1;
2344                         }
2345                         if (is_remount &&
2346                             (test_opt(sb, DATA_FLAGS) ==
2347                              EXT4_MOUNT_JOURNAL_DATA)) {
2348                                     ext4_msg(sb, KERN_ERR, "can't mount with "
2349                                              "both data=journal and dax");
2350                                     return -1;
2351                         }
2352                         ext4_msg(sb, KERN_WARNING,
2353                                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
2354                         sbi->s_mount_opt |= EXT4_MOUNT_DAX_ALWAYS;
2355                         sbi->s_mount_opt2 &= ~EXT4_MOUNT2_DAX_NEVER;
2356                         break;
2357                 case Opt_dax_never:
2358                         if (is_remount &&
2359                             (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2360                              (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS)))
2361                                 goto fail_dax_change_remount;
2362                         sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_NEVER;
2363                         sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS;
2364                         break;
2365                 case Opt_dax_inode:
2366                         if (is_remount &&
2367                             ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2368                              (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2369                              !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE)))
2370                                 goto fail_dax_change_remount;
2371                         sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS;
2372                         sbi->s_mount_opt2 &= ~EXT4_MOUNT2_DAX_NEVER;
2373                         /* Strictly for printing options */
2374                         sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_INODE;
2375                         break;
2376                 }
2377 #else
2378                 ext4_msg(sb, KERN_INFO, "dax option not supported");
2379                 sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_NEVER;
2380                 sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS;
2381                 return -1;
2382 #endif
2383         } else if (token == Opt_data_err_abort) {
2384                 sbi->s_mount_opt |= m->mount_opt;
2385         } else if (token == Opt_data_err_ignore) {
2386                 sbi->s_mount_opt &= ~m->mount_opt;
2387         } else {
2388                 if (!args->from)
2389                         arg = 1;
2390                 if (m->flags & MOPT_CLEAR)
2391                         arg = !arg;
2392                 else if (unlikely(!(m->flags & MOPT_SET))) {
2393                         ext4_msg(sb, KERN_WARNING,
2394                                  "buggy handling of option %s", opt);
2395                         WARN_ON(1);
2396                         return -1;
2397                 }
2398                 if (m->flags & MOPT_2) {
2399                         if (arg != 0)
2400                                 sbi->s_mount_opt2 |= m->mount_opt;
2401                         else
2402                                 sbi->s_mount_opt2 &= ~m->mount_opt;
2403                 } else {
2404                         if (arg != 0)
2405                                 sbi->s_mount_opt |= m->mount_opt;
2406                         else
2407                                 sbi->s_mount_opt &= ~m->mount_opt;
2408                 }
2409         }
2410         return 1;
2411 }
2412
2413 static int parse_options(char *options, struct super_block *sb,
2414                          unsigned long *journal_devnum,
2415                          unsigned int *journal_ioprio,
2416                          int is_remount)
2417 {
2418         struct ext4_sb_info __maybe_unused *sbi = EXT4_SB(sb);
2419         char *p, __maybe_unused *usr_qf_name, __maybe_unused *grp_qf_name;
2420         substring_t args[MAX_OPT_ARGS];
2421         int token;
2422
2423         if (!options)
2424                 return 1;
2425
2426         while ((p = strsep(&options, ",")) != NULL) {
2427                 if (!*p)
2428                         continue;
2429                 /*
2430                  * Initialize args struct so we know whether arg was
2431                  * found; some options take optional arguments.
2432                  */
2433                 args[0].to = args[0].from = NULL;
2434                 token = match_token(p, tokens, args);
2435                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
2436                                      journal_ioprio, is_remount) < 0)
2437                         return 0;
2438         }
2439 #ifdef CONFIG_QUOTA
2440         /*
2441          * We do the test below only for project quotas. 'usrquota' and
2442          * 'grpquota' mount options are allowed even without quota feature
2443          * to support legacy quotas in quota files.
2444          */
2445         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
2446                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
2447                          "Cannot enable project quota enforcement.");
2448                 return 0;
2449         }
2450         usr_qf_name = get_qf_name(sb, sbi, USRQUOTA);
2451         grp_qf_name = get_qf_name(sb, sbi, GRPQUOTA);
2452         if (usr_qf_name || grp_qf_name) {
2453                 if (test_opt(sb, USRQUOTA) && usr_qf_name)
2454                         clear_opt(sb, USRQUOTA);
2455
2456                 if (test_opt(sb, GRPQUOTA) && grp_qf_name)
2457                         clear_opt(sb, GRPQUOTA);
2458
2459                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
2460                         ext4_msg(sb, KERN_ERR, "old and new quota "
2461                                         "format mixing");
2462                         return 0;
2463                 }
2464
2465                 if (!sbi->s_jquota_fmt) {
2466                         ext4_msg(sb, KERN_ERR, "journaled quota format "
2467                                         "not specified");
2468                         return 0;
2469                 }
2470         }
2471 #endif
2472         if (test_opt(sb, DIOREAD_NOLOCK)) {
2473                 int blocksize =
2474                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
2475                 if (blocksize < PAGE_SIZE)
2476                         ext4_msg(sb, KERN_WARNING, "Warning: mounting with an "
2477                                  "experimental mount option 'dioread_nolock' "
2478                                  "for blocksize < PAGE_SIZE");
2479         }
2480         return 1;
2481 }
2482
2483 static inline void ext4_show_quota_options(struct seq_file *seq,
2484                                            struct super_block *sb)
2485 {
2486 #if defined(CONFIG_QUOTA)
2487         struct ext4_sb_info *sbi = EXT4_SB(sb);
2488         char *usr_qf_name, *grp_qf_name;
2489
2490         if (sbi->s_jquota_fmt) {
2491                 char *fmtname = "";
2492
2493                 switch (sbi->s_jquota_fmt) {
2494                 case QFMT_VFS_OLD:
2495                         fmtname = "vfsold";
2496                         break;
2497                 case QFMT_VFS_V0:
2498                         fmtname = "vfsv0";
2499                         break;
2500                 case QFMT_VFS_V1:
2501                         fmtname = "vfsv1";
2502                         break;
2503                 }
2504                 seq_printf(seq, ",jqfmt=%s", fmtname);
2505         }
2506
2507         rcu_read_lock();
2508         usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]);
2509         grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]);
2510         if (usr_qf_name)
2511                 seq_show_option(seq, "usrjquota", usr_qf_name);
2512         if (grp_qf_name)
2513                 seq_show_option(seq, "grpjquota", grp_qf_name);
2514         rcu_read_unlock();
2515 #endif
2516 }
2517
2518 static const char *token2str(int token)
2519 {
2520         const struct match_token *t;
2521
2522         for (t = tokens; t->token != Opt_err; t++)
2523                 if (t->token == token && !strchr(t->pattern, '='))
2524                         break;
2525         return t->pattern;
2526 }
2527
2528 /*
2529  * Show an option if
2530  *  - it's set to a non-default value OR
2531  *  - if the per-sb default is different from the global default
2532  */
2533 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2534                               int nodefs)
2535 {
2536         struct ext4_sb_info *sbi = EXT4_SB(sb);
2537         struct ext4_super_block *es = sbi->s_es;
2538         int def_errors, def_mount_opt = sbi->s_def_mount_opt;
2539         const struct mount_opts *m;
2540         char sep = nodefs ? '\n' : ',';
2541
2542 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2543 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2544
2545         if (sbi->s_sb_block != 1)
2546                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2547
2548         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2549                 int want_set = m->flags & MOPT_SET;
2550                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2551                     (m->flags & MOPT_CLEAR_ERR) || m->flags & MOPT_SKIP)
2552                         continue;
2553                 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2554                         continue; /* skip if same as the default */
2555                 if ((want_set &&
2556                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2557                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2558                         continue; /* select Opt_noFoo vs Opt_Foo */
2559                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2560         }
2561
2562         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2563             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2564                 SEQ_OPTS_PRINT("resuid=%u",
2565                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2566         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2567             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2568                 SEQ_OPTS_PRINT("resgid=%u",
2569                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2570         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2571         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2572                 SEQ_OPTS_PUTS("errors=remount-ro");
2573         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2574                 SEQ_OPTS_PUTS("errors=continue");
2575         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2576                 SEQ_OPTS_PUTS("errors=panic");
2577         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2578                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2579         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2580                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2581         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2582                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2583         if (sb->s_flags & SB_I_VERSION)
2584                 SEQ_OPTS_PUTS("i_version");
2585         if (nodefs || sbi->s_stripe)
2586                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2587         if (nodefs || EXT4_MOUNT_DATA_FLAGS &
2588                         (sbi->s_mount_opt ^ def_mount_opt)) {
2589                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2590                         SEQ_OPTS_PUTS("data=journal");
2591                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2592                         SEQ_OPTS_PUTS("data=ordered");
2593                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2594                         SEQ_OPTS_PUTS("data=writeback");
2595         }
2596         if (nodefs ||
2597             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2598                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2599                                sbi->s_inode_readahead_blks);
2600
2601         if (test_opt(sb, INIT_INODE_TABLE) && (nodefs ||
2602                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2603                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2604         if (nodefs || sbi->s_max_dir_size_kb)
2605                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2606         if (test_opt(sb, DATA_ERR_ABORT))
2607                 SEQ_OPTS_PUTS("data_err=abort");
2608
2609         fscrypt_show_test_dummy_encryption(seq, sep, sb);
2610
2611         if (sb->s_flags & SB_INLINECRYPT)
2612                 SEQ_OPTS_PUTS("inlinecrypt");
2613
2614         if (test_opt(sb, DAX_ALWAYS)) {
2615                 if (IS_EXT2_SB(sb))
2616                         SEQ_OPTS_PUTS("dax");
2617                 else
2618                         SEQ_OPTS_PUTS("dax=always");
2619         } else if (test_opt2(sb, DAX_NEVER)) {
2620                 SEQ_OPTS_PUTS("dax=never");
2621         } else if (test_opt2(sb, DAX_INODE)) {
2622                 SEQ_OPTS_PUTS("dax=inode");
2623         }
2624         ext4_show_quota_options(seq, sb);
2625         return 0;
2626 }
2627
2628 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2629 {
2630         return _ext4_show_options(seq, root->d_sb, 0);
2631 }
2632
2633 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2634 {
2635         struct super_block *sb = seq->private;
2636         int rc;
2637
2638         seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
2639         rc = _ext4_show_options(seq, sb, 1);
2640         seq_puts(seq, "\n");
2641         return rc;
2642 }
2643
2644 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2645                             int read_only)
2646 {
2647         struct ext4_sb_info *sbi = EXT4_SB(sb);
2648         int err = 0;
2649
2650         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2651                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2652                          "forcing read-only mode");
2653                 err = -EROFS;
2654                 goto done;
2655         }
2656         if (read_only)
2657                 goto done;
2658         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2659                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2660                          "running e2fsck is recommended");
2661         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2662                 ext4_msg(sb, KERN_WARNING,
2663                          "warning: mounting fs with errors, "
2664                          "running e2fsck is recommended");
2665         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2666                  le16_to_cpu(es->s_mnt_count) >=
2667                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2668                 ext4_msg(sb, KERN_WARNING,
2669                          "warning: maximal mount count reached, "
2670                          "running e2fsck is recommended");
2671         else if (le32_to_cpu(es->s_checkinterval) &&
2672                  (ext4_get_tstamp(es, s_lastcheck) +
2673                   le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds()))
2674                 ext4_msg(sb, KERN_WARNING,
2675                          "warning: checktime reached, "
2676                          "running e2fsck is recommended");
2677         if (!sbi->s_journal)
2678                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2679         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2680                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2681         le16_add_cpu(&es->s_mnt_count, 1);
2682         ext4_update_tstamp(es, s_mtime);
2683         if (sbi->s_journal)
2684                 ext4_set_feature_journal_needs_recovery(sb);
2685
2686         err = ext4_commit_super(sb);
2687 done:
2688         if (test_opt(sb, DEBUG))
2689                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2690                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2691                         sb->s_blocksize,
2692                         sbi->s_groups_count,
2693                         EXT4_BLOCKS_PER_GROUP(sb),
2694                         EXT4_INODES_PER_GROUP(sb),
2695                         sbi->s_mount_opt, sbi->s_mount_opt2);
2696
2697         cleancache_init_fs(sb);
2698         return err;
2699 }
2700
2701 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2702 {
2703         struct ext4_sb_info *sbi = EXT4_SB(sb);
2704         struct flex_groups **old_groups, **new_groups;
2705         int size, i, j;
2706
2707         if (!sbi->s_log_groups_per_flex)
2708                 return 0;
2709
2710         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2711         if (size <= sbi->s_flex_groups_allocated)
2712                 return 0;
2713
2714         new_groups = kvzalloc(roundup_pow_of_two(size *
2715                               sizeof(*sbi->s_flex_groups)), GFP_KERNEL);
2716         if (!new_groups) {
2717                 ext4_msg(sb, KERN_ERR,
2718                          "not enough memory for %d flex group pointers", size);
2719                 return -ENOMEM;
2720         }
2721         for (i = sbi->s_flex_groups_allocated; i < size; i++) {
2722                 new_groups[i] = kvzalloc(roundup_pow_of_two(
2723                                          sizeof(struct flex_groups)),
2724                                          GFP_KERNEL);
2725                 if (!new_groups[i]) {
2726                         for (j = sbi->s_flex_groups_allocated; j < i; j++)
2727                                 kvfree(new_groups[j]);
2728                         kvfree(new_groups);
2729                         ext4_msg(sb, KERN_ERR,
2730                                  "not enough memory for %d flex groups", size);
2731                         return -ENOMEM;
2732                 }
2733         }
2734         rcu_read_lock();
2735         old_groups = rcu_dereference(sbi->s_flex_groups);
2736         if (old_groups)
2737                 memcpy(new_groups, old_groups,
2738                        (sbi->s_flex_groups_allocated *
2739                         sizeof(struct flex_groups *)));
2740         rcu_read_unlock();
2741         rcu_assign_pointer(sbi->s_flex_groups, new_groups);
2742         sbi->s_flex_groups_allocated = size;
2743         if (old_groups)
2744                 ext4_kvfree_array_rcu(old_groups);
2745         return 0;
2746 }
2747
2748 static int ext4_fill_flex_info(struct super_block *sb)
2749 {
2750         struct ext4_sb_info *sbi = EXT4_SB(sb);
2751         struct ext4_group_desc *gdp = NULL;
2752         struct flex_groups *fg;
2753         ext4_group_t flex_group;
2754         int i, err;
2755
2756         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2757         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2758                 sbi->s_log_groups_per_flex = 0;
2759                 return 1;
2760         }
2761
2762         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2763         if (err)
2764                 goto failed;
2765
2766         for (i = 0; i < sbi->s_groups_count; i++) {
2767                 gdp = ext4_get_group_desc(sb, i, NULL);
2768
2769                 flex_group = ext4_flex_group(sbi, i);
2770                 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
2771                 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes);
2772                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2773                              &fg->free_clusters);
2774                 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs);
2775         }
2776
2777         return 1;
2778 failed:
2779         return 0;
2780 }
2781
2782 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2783                                    struct ext4_group_desc *gdp)
2784 {
2785         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2786         __u16 crc = 0;
2787         __le32 le_group = cpu_to_le32(block_group);
2788         struct ext4_sb_info *sbi = EXT4_SB(sb);
2789
2790         if (ext4_has_metadata_csum(sbi->s_sb)) {
2791                 /* Use new metadata_csum algorithm */
2792                 __u32 csum32;
2793                 __u16 dummy_csum = 0;
2794
2795                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2796                                      sizeof(le_group));
2797                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2798                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2799                                      sizeof(dummy_csum));
2800                 offset += sizeof(dummy_csum);
2801                 if (offset < sbi->s_desc_size)
2802                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2803                                              sbi->s_desc_size - offset);
2804
2805                 crc = csum32 & 0xFFFF;
2806                 goto out;
2807         }
2808
2809         /* old crc16 code */
2810         if (!ext4_has_feature_gdt_csum(sb))
2811                 return 0;
2812
2813         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2814         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2815         crc = crc16(crc, (__u8 *)gdp, offset);
2816         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2817         /* for checksum of struct ext4_group_desc do the rest...*/
2818         if (ext4_has_feature_64bit(sb) &&
2819             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2820                 crc = crc16(crc, (__u8 *)gdp + offset,
2821                             le16_to_cpu(sbi->s_es->s_desc_size) -
2822                                 offset);
2823
2824 out:
2825         return cpu_to_le16(crc);
2826 }
2827
2828 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2829                                 struct ext4_group_desc *gdp)
2830 {
2831         if (ext4_has_group_desc_csum(sb) &&
2832             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2833                 return 0;
2834
2835         return 1;
2836 }
2837
2838 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2839                               struct ext4_group_desc *gdp)
2840 {
2841         if (!ext4_has_group_desc_csum(sb))
2842                 return;
2843         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2844 }
2845
2846 /* Called at mount-time, super-block is locked */
2847 static int ext4_check_descriptors(struct super_block *sb,
2848                                   ext4_fsblk_t sb_block,
2849                                   ext4_group_t *first_not_zeroed)
2850 {
2851         struct ext4_sb_info *sbi = EXT4_SB(sb);
2852         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2853         ext4_fsblk_t last_block;
2854         ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0);
2855         ext4_fsblk_t block_bitmap;
2856         ext4_fsblk_t inode_bitmap;
2857         ext4_fsblk_t inode_table;
2858         int flexbg_flag = 0;
2859         ext4_group_t i, grp = sbi->s_groups_count;
2860
2861         if (ext4_has_feature_flex_bg(sb))
2862                 flexbg_flag = 1;
2863
2864         ext4_debug("Checking group descriptors");
2865
2866         for (i = 0; i < sbi->s_groups_count; i++) {
2867                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2868
2869                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2870                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2871                 else
2872                         last_block = first_block +
2873                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2874
2875                 if ((grp == sbi->s_groups_count) &&
2876                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2877                         grp = i;
2878
2879                 block_bitmap = ext4_block_bitmap(sb, gdp);
2880                 if (block_bitmap == sb_block) {
2881                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2882                                  "Block bitmap for group %u overlaps "
2883                                  "superblock", i);
2884                         if (!sb_rdonly(sb))
2885                                 return 0;
2886                 }
2887                 if (block_bitmap >= sb_block + 1 &&
2888                     block_bitmap <= last_bg_block) {
2889                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2890                                  "Block bitmap for group %u overlaps "
2891                                  "block group descriptors", i);
2892                         if (!sb_rdonly(sb))
2893                                 return 0;
2894                 }
2895                 if (block_bitmap < first_block || block_bitmap > last_block) {
2896                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2897                                "Block bitmap for group %u not in group "
2898                                "(block %llu)!", i, block_bitmap);
2899                         return 0;
2900                 }
2901                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2902                 if (inode_bitmap == sb_block) {
2903                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2904                                  "Inode bitmap for group %u overlaps "
2905                                  "superblock", i);
2906                         if (!sb_rdonly(sb))
2907                                 return 0;
2908                 }
2909                 if (inode_bitmap >= sb_block + 1 &&
2910                     inode_bitmap <= last_bg_block) {
2911                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2912                                  "Inode bitmap for group %u overlaps "
2913                                  "block group descriptors", i);
2914                         if (!sb_rdonly(sb))
2915                                 return 0;
2916                 }
2917                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2918                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2919                                "Inode bitmap for group %u not in group "
2920                                "(block %llu)!", i, inode_bitmap);
2921                         return 0;
2922                 }
2923                 inode_table = ext4_inode_table(sb, gdp);
2924                 if (inode_table == sb_block) {
2925                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2926                                  "Inode table for group %u overlaps "
2927                                  "superblock", i);
2928                         if (!sb_rdonly(sb))
2929                                 return 0;
2930                 }
2931                 if (inode_table >= sb_block + 1 &&
2932                     inode_table <= last_bg_block) {
2933                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2934                                  "Inode table for group %u overlaps "
2935                                  "block group descriptors", i);
2936                         if (!sb_rdonly(sb))
2937                                 return 0;
2938                 }
2939                 if (inode_table < first_block ||
2940                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2941                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2942                                "Inode table for group %u not in group "
2943                                "(block %llu)!", i, inode_table);
2944                         return 0;
2945                 }
2946                 ext4_lock_group(sb, i);
2947                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2948                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2949                                  "Checksum for group %u failed (%u!=%u)",
2950                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2951                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2952                         if (!sb_rdonly(sb)) {
2953                                 ext4_unlock_group(sb, i);
2954                                 return 0;
2955                         }
2956                 }
2957                 ext4_unlock_group(sb, i);
2958                 if (!flexbg_flag)
2959                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2960         }
2961         if (NULL != first_not_zeroed)
2962                 *first_not_zeroed = grp;
2963         return 1;
2964 }
2965
2966 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2967  * the superblock) which were deleted from all directories, but held open by
2968  * a process at the time of a crash.  We walk the list and try to delete these
2969  * inodes at recovery time (only with a read-write filesystem).
2970  *
2971  * In order to keep the orphan inode chain consistent during traversal (in
2972  * case of crash during recovery), we link each inode into the superblock
2973  * orphan list_head and handle it the same way as an inode deletion during
2974  * normal operation (which journals the operations for us).
2975  *
2976  * We only do an iget() and an iput() on each inode, which is very safe if we
2977  * accidentally point at an in-use or already deleted inode.  The worst that
2978  * can happen in this case is that we get a "bit already cleared" message from
2979  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2980  * e2fsck was run on this filesystem, and it must have already done the orphan
2981  * inode cleanup for us, so we can safely abort without any further action.
2982  */
2983 static void ext4_orphan_cleanup(struct super_block *sb,
2984                                 struct ext4_super_block *es)
2985 {
2986         unsigned int s_flags = sb->s_flags;
2987         int ret, nr_orphans = 0, nr_truncates = 0;
2988 #ifdef CONFIG_QUOTA
2989         int quota_update = 0;
2990         int i;
2991 #endif
2992         if (!es->s_last_orphan) {
2993                 jbd_debug(4, "no orphan inodes to clean up\n");
2994                 return;
2995         }
2996
2997         if (bdev_read_only(sb->s_bdev)) {
2998                 ext4_msg(sb, KERN_ERR, "write access "
2999                         "unavailable, skipping orphan cleanup");
3000                 return;
3001         }
3002
3003         /* Check if feature set would not allow a r/w mount */
3004         if (!ext4_feature_set_ok(sb, 0)) {
3005                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
3006                          "unknown ROCOMPAT features");
3007                 return;
3008         }
3009
3010         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
3011                 /* don't clear list on RO mount w/ errors */
3012                 if (es->s_last_orphan && !(s_flags & SB_RDONLY)) {
3013                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
3014                                   "clearing orphan list.\n");
3015                         es->s_last_orphan = 0;
3016                 }
3017                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
3018                 return;
3019         }
3020
3021         if (s_flags & SB_RDONLY) {
3022                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
3023                 sb->s_flags &= ~SB_RDONLY;
3024         }
3025 #ifdef CONFIG_QUOTA
3026         /* Needed for iput() to work correctly and not trash data */
3027         sb->s_flags |= SB_ACTIVE;
3028
3029         /*
3030          * Turn on quotas which were not enabled for read-only mounts if
3031          * filesystem has quota feature, so that they are updated correctly.
3032          */
3033         if (ext4_has_feature_quota(sb) && (s_flags & SB_RDONLY)) {
3034                 int ret = ext4_enable_quotas(sb);
3035
3036                 if (!ret)
3037                         quota_update = 1;
3038                 else
3039                         ext4_msg(sb, KERN_ERR,
3040                                 "Cannot turn on quotas: error %d", ret);
3041         }
3042
3043         /* Turn on journaled quotas used for old sytle */
3044         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
3045                 if (EXT4_SB(sb)->s_qf_names[i]) {
3046                         int ret = ext4_quota_on_mount(sb, i);
3047
3048                         if (!ret)
3049                                 quota_update = 1;
3050                         else
3051                                 ext4_msg(sb, KERN_ERR,
3052                                         "Cannot turn on journaled "
3053                                         "quota: type %d: error %d", i, ret);
3054                 }
3055         }
3056 #endif
3057
3058         while (es->s_last_orphan) {
3059                 struct inode *inode;
3060
3061                 /*
3062                  * We may have encountered an error during cleanup; if
3063                  * so, skip the rest.
3064                  */
3065                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
3066                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
3067                         es->s_last_orphan = 0;
3068                         break;
3069                 }
3070
3071                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
3072                 if (IS_ERR(inode)) {
3073                         es->s_last_orphan = 0;
3074                         break;
3075                 }
3076
3077                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
3078                 dquot_initialize(inode);
3079                 if (inode->i_nlink) {
3080                         if (test_opt(sb, DEBUG))
3081                                 ext4_msg(sb, KERN_DEBUG,
3082                                         "%s: truncating inode %lu to %lld bytes",
3083                                         __func__, inode->i_ino, inode->i_size);
3084                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
3085                                   inode->i_ino, inode->i_size);
3086                         inode_lock(inode);
3087                         truncate_inode_pages(inode->i_mapping, inode->i_size);
3088                         ret = ext4_truncate(inode);
3089                         if (ret)
3090                                 ext4_std_error(inode->i_sb, ret);
3091                         inode_unlock(inode);
3092                         nr_truncates++;
3093                 } else {
3094                         if (test_opt(sb, DEBUG))
3095                                 ext4_msg(sb, KERN_DEBUG,
3096                                         "%s: deleting unreferenced inode %lu",
3097                                         __func__, inode->i_ino);
3098                         jbd_debug(2, "deleting unreferenced inode %lu\n",
3099                                   inode->i_ino);
3100                         nr_orphans++;
3101                 }
3102                 iput(inode);  /* The delete magic happens here! */
3103         }
3104
3105 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
3106
3107         if (nr_orphans)
3108                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
3109                        PLURAL(nr_orphans));
3110         if (nr_truncates)
3111                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
3112                        PLURAL(nr_truncates));
3113 #ifdef CONFIG_QUOTA
3114         /* Turn off quotas if they were enabled for orphan cleanup */
3115         if (quota_update) {
3116                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
3117                         if (sb_dqopt(sb)->files[i])
3118                                 dquot_quota_off(sb, i);
3119                 }
3120         }
3121 #endif
3122         sb->s_flags = s_flags; /* Restore SB_RDONLY status */
3123 }
3124
3125 /*
3126  * Maximal extent format file size.
3127  * Resulting logical blkno at s_maxbytes must fit in our on-disk
3128  * extent format containers, within a sector_t, and within i_blocks
3129  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
3130  * so that won't be a limiting factor.
3131  *
3132  * However there is other limiting factor. We do store extents in the form
3133  * of starting block and length, hence the resulting length of the extent
3134  * covering maximum file size must fit into on-disk format containers as
3135  * well. Given that length is always by 1 unit bigger than max unit (because
3136  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
3137  *
3138  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
3139  */
3140 static loff_t ext4_max_size(int blkbits, int has_huge_files)
3141 {
3142         loff_t res;
3143         loff_t upper_limit = MAX_LFS_FILESIZE;
3144
3145         BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64));
3146
3147         if (!has_huge_files) {
3148                 upper_limit = (1LL << 32) - 1;
3149
3150                 /* total blocks in file system block size */
3151                 upper_limit >>= (blkbits - 9);
3152                 upper_limit <<= blkbits;
3153         }
3154
3155         /*
3156          * 32-bit extent-start container, ee_block. We lower the maxbytes
3157          * by one fs block, so ee_len can cover the extent of maximum file
3158          * size
3159          */
3160         res = (1LL << 32) - 1;
3161         res <<= blkbits;
3162
3163         /* Sanity check against vm- & vfs- imposed limits */
3164         if (res > upper_limit)
3165                 res = upper_limit;
3166
3167         return res;
3168 }
3169
3170 /*
3171  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
3172  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
3173  * We need to be 1 filesystem block less than the 2^48 sector limit.
3174  */
3175 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
3176 {
3177         loff_t res = EXT4_NDIR_BLOCKS;
3178         int meta_blocks;
3179         loff_t upper_limit;
3180         /* This is calculated to be the largest file size for a dense, block
3181          * mapped file such that the file's total number of 512-byte sectors,
3182          * including data and all indirect blocks, does not exceed (2^48 - 1).
3183          *
3184          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
3185          * number of 512-byte sectors of the file.
3186          */
3187
3188         if (!has_huge_files) {
3189                 /*
3190                  * !has_huge_files or implies that the inode i_block field
3191                  * represents total file blocks in 2^32 512-byte sectors ==
3192                  * size of vfs inode i_blocks * 8
3193                  */
3194                 upper_limit = (1LL << 32) - 1;
3195
3196                 /* total blocks in file system block size */
3197                 upper_limit >>= (bits - 9);
3198
3199         } else {
3200                 /*
3201                  * We use 48 bit ext4_inode i_blocks
3202                  * With EXT4_HUGE_FILE_FL set the i_blocks
3203                  * represent total number of blocks in
3204                  * file system block size
3205                  */
3206                 upper_limit = (1LL << 48) - 1;
3207
3208         }
3209
3210         /* indirect blocks */
3211         meta_blocks = 1;
3212         /* double indirect blocks */
3213         meta_blocks += 1 + (1LL << (bits-2));
3214         /* tripple indirect blocks */
3215         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
3216
3217         upper_limit -= meta_blocks;
3218         upper_limit <<= bits;
3219
3220         res += 1LL << (bits-2);
3221         res += 1LL << (2*(bits-2));
3222         res += 1LL << (3*(bits-2));
3223         res <<= bits;
3224         if (res > upper_limit)
3225                 res = upper_limit;
3226
3227         if (res > MAX_LFS_FILESIZE)
3228                 res = MAX_LFS_FILESIZE;
3229
3230         return res;
3231 }
3232
3233 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
3234                                    ext4_fsblk_t logical_sb_block, int nr)
3235 {
3236         struct ext4_sb_info *sbi = EXT4_SB(sb);
3237         ext4_group_t bg, first_meta_bg;
3238         int has_super = 0;
3239
3240         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
3241
3242         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
3243                 return logical_sb_block + nr + 1;
3244         bg = sbi->s_desc_per_block * nr;
3245         if (ext4_bg_has_super(sb, bg))
3246                 has_super = 1;
3247
3248         /*
3249          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
3250          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
3251          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
3252          * compensate.
3253          */
3254         if (sb->s_blocksize == 1024 && nr == 0 &&
3255             le32_to_cpu(sbi->s_es->s_first_data_block) == 0)
3256                 has_super++;
3257
3258         return (has_super + ext4_group_first_block_no(sb, bg));
3259 }
3260
3261 /**
3262  * ext4_get_stripe_size: Get the stripe size.
3263  * @sbi: In memory super block info
3264  *
3265  * If we have specified it via mount option, then
3266  * use the mount option value. If the value specified at mount time is
3267  * greater than the blocks per group use the super block value.
3268  * If the super block value is greater than blocks per group return 0.
3269  * Allocator needs it be less than blocks per group.
3270  *
3271  */
3272 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
3273 {
3274         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
3275         unsigned long stripe_width =
3276                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
3277         int ret;
3278
3279         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
3280                 ret = sbi->s_stripe;
3281         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
3282                 ret = stripe_width;
3283         else if (stride && stride <= sbi->s_blocks_per_group)
3284                 ret = stride;
3285         else
3286                 ret = 0;
3287
3288         /*
3289          * If the stripe width is 1, this makes no sense and
3290          * we set it to 0 to turn off stripe handling code.
3291          */
3292         if (ret <= 1)
3293                 ret = 0;
3294
3295         return ret;
3296 }
3297
3298 /*
3299  * Check whether this filesystem can be mounted based on
3300  * the features present and the RDONLY/RDWR mount requested.
3301  * Returns 1 if this filesystem can be mounted as requested,
3302  * 0 if it cannot be.
3303  */
3304 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
3305 {
3306         if (ext4_has_unknown_ext4_incompat_features(sb)) {
3307                 ext4_msg(sb, KERN_ERR,
3308                         "Couldn't mount because of "
3309                         "unsupported optional features (%x)",
3310                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
3311                         ~EXT4_FEATURE_INCOMPAT_SUPP));
3312                 return 0;
3313         }
3314
3315 #ifndef CONFIG_UNICODE
3316         if (ext4_has_feature_casefold(sb)) {
3317                 ext4_msg(sb, KERN_ERR,
3318                          "Filesystem with casefold feature cannot be "
3319                          "mounted without CONFIG_UNICODE");
3320                 return 0;
3321         }
3322 #endif
3323
3324         if (readonly)
3325                 return 1;
3326
3327         if (ext4_has_feature_readonly(sb)) {
3328                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
3329                 sb->s_flags |= SB_RDONLY;
3330                 return 1;
3331         }
3332
3333         /* Check that feature set is OK for a read-write mount */
3334         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
3335                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
3336                          "unsupported optional features (%x)",
3337                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
3338                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
3339                 return 0;
3340         }
3341         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
3342                 ext4_msg(sb, KERN_ERR,
3343                          "Can't support bigalloc feature without "
3344                          "extents feature\n");
3345                 return 0;
3346         }
3347
3348 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2)
3349         if (!readonly && (ext4_has_feature_quota(sb) ||
3350                           ext4_has_feature_project(sb))) {
3351                 ext4_msg(sb, KERN_ERR,
3352                          "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2");
3353                 return 0;
3354         }
3355 #endif  /* CONFIG_QUOTA */
3356         return 1;
3357 }
3358
3359 /*
3360  * This function is called once a day if we have errors logged
3361  * on the file system
3362  */
3363 static void print_daily_error_info(struct timer_list *t)
3364 {
3365         struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report);
3366         struct super_block *sb = sbi->s_sb;
3367         struct ext4_super_block *es = sbi->s_es;
3368
3369         if (es->s_error_count)
3370                 /* fsck newer than v1.41.13 is needed to clean this condition. */
3371                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
3372                          le32_to_cpu(es->s_error_count));
3373         if (es->s_first_error_time) {
3374                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d",
3375                        sb->s_id,
3376                        ext4_get_tstamp(es, s_first_error_time),
3377                        (int) sizeof(es->s_first_error_func),
3378                        es->s_first_error_func,
3379                        le32_to_cpu(es->s_first_error_line));
3380                 if (es->s_first_error_ino)
3381                         printk(KERN_CONT ": inode %u",
3382                                le32_to_cpu(es->s_first_error_ino));
3383                 if (es->s_first_error_block)
3384                         printk(KERN_CONT ": block %llu", (unsigned long long)
3385                                le64_to_cpu(es->s_first_error_block));
3386                 printk(KERN_CONT "\n");
3387         }
3388         if (es->s_last_error_time) {
3389                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d",
3390                        sb->s_id,
3391                        ext4_get_tstamp(es, s_last_error_time),
3392                        (int) sizeof(es->s_last_error_func),
3393                        es->s_last_error_func,
3394                        le32_to_cpu(es->s_last_error_line));
3395                 if (es->s_last_error_ino)
3396                         printk(KERN_CONT ": inode %u",
3397                                le32_to_cpu(es->s_last_error_ino));
3398                 if (es->s_last_error_block)
3399                         printk(KERN_CONT ": block %llu", (unsigned long long)
3400                                le64_to_cpu(es->s_last_error_block));
3401                 printk(KERN_CONT "\n");
3402         }
3403         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
3404 }
3405
3406 /* Find next suitable group and run ext4_init_inode_table */
3407 static int ext4_run_li_request(struct ext4_li_request *elr)
3408 {
3409         struct ext4_group_desc *gdp = NULL;
3410         struct super_block *sb = elr->lr_super;
3411         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3412         ext4_group_t group = elr->lr_next_group;
3413         unsigned long timeout = 0;
3414         unsigned int prefetch_ios = 0;
3415         int ret = 0;
3416
3417         if (elr->lr_mode == EXT4_LI_MODE_PREFETCH_BBITMAP) {
3418                 elr->lr_next_group = ext4_mb_prefetch(sb, group,
3419                                 EXT4_SB(sb)->s_mb_prefetch, &prefetch_ios);
3420                 if (prefetch_ios)
3421                         ext4_mb_prefetch_fini(sb, elr->lr_next_group,
3422                                               prefetch_ios);
3423                 trace_ext4_prefetch_bitmaps(sb, group, elr->lr_next_group,
3424                                             prefetch_ios);
3425                 if (group >= elr->lr_next_group) {
3426                         ret = 1;
3427                         if (elr->lr_first_not_zeroed != ngroups &&
3428                             !sb_rdonly(sb) && test_opt(sb, INIT_INODE_TABLE)) {
3429                                 elr->lr_next_group = elr->lr_first_not_zeroed;
3430                                 elr->lr_mode = EXT4_LI_MODE_ITABLE;
3431                                 ret = 0;
3432                         }
3433                 }
3434                 return ret;
3435         }
3436
3437         for (; group < ngroups; group++) {
3438                 gdp = ext4_get_group_desc(sb, group, NULL);
3439                 if (!gdp) {
3440                         ret = 1;
3441                         break;
3442                 }
3443
3444                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3445                         break;
3446         }
3447
3448         if (group >= ngroups)
3449                 ret = 1;
3450
3451         if (!ret) {
3452                 timeout = jiffies;
3453                 ret = ext4_init_inode_table(sb, group,
3454                                             elr->lr_timeout ? 0 : 1);
3455                 trace_ext4_lazy_itable_init(sb, group);
3456                 if (elr->lr_timeout == 0) {
3457                         timeout = (jiffies - timeout) *
3458                                 EXT4_SB(elr->lr_super)->s_li_wait_mult;
3459                         elr->lr_timeout = timeout;
3460                 }
3461                 elr->lr_next_sched = jiffies + elr->lr_timeout;
3462                 elr->lr_next_group = group + 1;
3463         }
3464         return ret;
3465 }
3466
3467 /*
3468  * Remove lr_request from the list_request and free the
3469  * request structure. Should be called with li_list_mtx held
3470  */
3471 static void ext4_remove_li_request(struct ext4_li_request *elr)
3472 {
3473         if (!elr)
3474                 return;
3475
3476         list_del(&elr->lr_request);
3477         EXT4_SB(elr->lr_super)->s_li_request = NULL;
3478         kfree(elr);
3479 }
3480
3481 static void ext4_unregister_li_request(struct super_block *sb)
3482 {
3483         mutex_lock(&ext4_li_mtx);
3484         if (!ext4_li_info) {
3485                 mutex_unlock(&ext4_li_mtx);
3486                 return;
3487         }
3488
3489         mutex_lock(&ext4_li_info->li_list_mtx);
3490         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
3491         mutex_unlock(&ext4_li_info->li_list_mtx);
3492         mutex_unlock(&ext4_li_mtx);
3493 }
3494
3495 static struct task_struct *ext4_lazyinit_task;
3496
3497 /*
3498  * This is the function where ext4lazyinit thread lives. It walks
3499  * through the request list searching for next scheduled filesystem.
3500  * When such a fs is found, run the lazy initialization request
3501  * (ext4_rn_li_request) and keep track of the time spend in this
3502  * function. Based on that time we compute next schedule time of
3503  * the request. When walking through the list is complete, compute
3504  * next waking time and put itself into sleep.
3505  */
3506 static int ext4_lazyinit_thread(void *arg)
3507 {
3508         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
3509         struct list_head *pos, *n;
3510         struct ext4_li_request *elr;
3511         unsigned long next_wakeup, cur;
3512
3513         BUG_ON(NULL == eli);
3514
3515 cont_thread:
3516         while (true) {
3517                 next_wakeup = MAX_JIFFY_OFFSET;
3518
3519                 mutex_lock(&eli->li_list_mtx);
3520                 if (list_empty(&eli->li_request_list)) {
3521                         mutex_unlock(&eli->li_list_mtx);
3522                         goto exit_thread;
3523                 }
3524                 list_for_each_safe(pos, n, &eli->li_request_list) {
3525                         int err = 0;
3526                         int progress = 0;
3527                         elr = list_entry(pos, struct ext4_li_request,
3528                                          lr_request);
3529
3530                         if (time_before(jiffies, elr->lr_next_sched)) {
3531                                 if (time_before(elr->lr_next_sched, next_wakeup))
3532                                         next_wakeup = elr->lr_next_sched;
3533                                 continue;
3534                         }
3535                         if (down_read_trylock(&elr->lr_super->s_umount)) {
3536                                 if (sb_start_write_trylock(elr->lr_super)) {
3537                                         progress = 1;
3538                                         /*
3539                                          * We hold sb->s_umount, sb can not
3540                                          * be removed from the list, it is
3541                                          * now safe to drop li_list_mtx
3542                                          */
3543                                         mutex_unlock(&eli->li_list_mtx);
3544                                         err = ext4_run_li_request(elr);
3545                                         sb_end_write(elr->lr_super);
3546                                         mutex_lock(&eli->li_list_mtx);
3547                                         n = pos->next;
3548                                 }
3549                                 up_read((&elr->lr_super->s_umount));
3550                         }
3551                         /* error, remove the lazy_init job */
3552                         if (err) {
3553                                 ext4_remove_li_request(elr);
3554                                 continue;
3555                         }
3556                         if (!progress) {
3557                                 elr->lr_next_sched = jiffies +
3558                                         (prandom_u32()
3559                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3560                         }
3561                         if (time_before(elr->lr_next_sched, next_wakeup))
3562                                 next_wakeup = elr->lr_next_sched;
3563                 }
3564                 mutex_unlock(&eli->li_list_mtx);
3565
3566                 try_to_freeze();
3567
3568                 cur = jiffies;
3569                 if ((time_after_eq(cur, next_wakeup)) ||
3570                     (MAX_JIFFY_OFFSET == next_wakeup)) {
3571                         cond_resched();
3572                         continue;
3573                 }
3574
3575                 schedule_timeout_interruptible(next_wakeup - cur);
3576
3577                 if (kthread_should_stop()) {
3578                         ext4_clear_request_list();
3579                         goto exit_thread;
3580                 }
3581         }
3582
3583 exit_thread:
3584         /*
3585          * It looks like the request list is empty, but we need
3586          * to check it under the li_list_mtx lock, to prevent any
3587          * additions into it, and of course we should lock ext4_li_mtx
3588          * to atomically free the list and ext4_li_info, because at
3589          * this point another ext4 filesystem could be registering
3590          * new one.
3591          */
3592         mutex_lock(&ext4_li_mtx);
3593         mutex_lock(&eli->li_list_mtx);
3594         if (!list_empty(&eli->li_request_list)) {
3595                 mutex_unlock(&eli->li_list_mtx);
3596                 mutex_unlock(&ext4_li_mtx);
3597                 goto cont_thread;
3598         }
3599         mutex_unlock(&eli->li_list_mtx);
3600         kfree(ext4_li_info);
3601         ext4_li_info = NULL;
3602         mutex_unlock(&ext4_li_mtx);
3603
3604         return 0;
3605 }
3606
3607 static void ext4_clear_request_list(void)
3608 {
3609         struct list_head *pos, *n;
3610         struct ext4_li_request *elr;
3611
3612         mutex_lock(&ext4_li_info->li_list_mtx);
3613         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3614                 elr = list_entry(pos, struct ext4_li_request,
3615                                  lr_request);
3616                 ext4_remove_li_request(elr);
3617         }
3618         mutex_unlock(&ext4_li_info->li_list_mtx);
3619 }
3620
3621 static int ext4_run_lazyinit_thread(void)
3622 {
3623         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3624                                          ext4_li_info, "ext4lazyinit");
3625         if (IS_ERR(ext4_lazyinit_task)) {
3626                 int err = PTR_ERR(ext4_lazyinit_task);
3627                 ext4_clear_request_list();
3628                 kfree(ext4_li_info);
3629                 ext4_li_info = NULL;
3630                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3631                                  "initialization thread\n",
3632                                  err);
3633                 return err;
3634         }
3635         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3636         return 0;
3637 }
3638
3639 /*
3640  * Check whether it make sense to run itable init. thread or not.
3641  * If there is at least one uninitialized inode table, return
3642  * corresponding group number, else the loop goes through all
3643  * groups and return total number of groups.
3644  */
3645 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3646 {
3647         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3648         struct ext4_group_desc *gdp = NULL;
3649
3650         if (!ext4_has_group_desc_csum(sb))
3651                 return ngroups;
3652
3653         for (group = 0; group < ngroups; group++) {
3654                 gdp = ext4_get_group_desc(sb, group, NULL);
3655                 if (!gdp)
3656                         continue;
3657
3658                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3659                         break;
3660         }
3661
3662         return group;
3663 }
3664
3665 static int ext4_li_info_new(void)
3666 {
3667         struct ext4_lazy_init *eli = NULL;
3668
3669         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3670         if (!eli)
3671                 return -ENOMEM;
3672
3673         INIT_LIST_HEAD(&eli->li_request_list);
3674         mutex_init(&eli->li_list_mtx);
3675
3676         eli->li_state |= EXT4_LAZYINIT_QUIT;
3677
3678         ext4_li_info = eli;
3679
3680         return 0;
3681 }
3682
3683 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3684                                             ext4_group_t start)
3685 {
3686         struct ext4_li_request *elr;
3687
3688         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3689         if (!elr)
3690                 return NULL;
3691
3692         elr->lr_super = sb;
3693         elr->lr_first_not_zeroed = start;
3694         if (test_opt(sb, PREFETCH_BLOCK_BITMAPS))
3695                 elr->lr_mode = EXT4_LI_MODE_PREFETCH_BBITMAP;
3696         else {
3697                 elr->lr_mode = EXT4_LI_MODE_ITABLE;
3698                 elr->lr_next_group = start;
3699         }
3700
3701         /*
3702          * Randomize first schedule time of the request to
3703          * spread the inode table initialization requests
3704          * better.
3705          */
3706         elr->lr_next_sched = jiffies + (prandom_u32() %
3707                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3708         return elr;
3709 }
3710
3711 int ext4_register_li_request(struct super_block *sb,
3712                              ext4_group_t first_not_zeroed)
3713 {
3714         struct ext4_sb_info *sbi = EXT4_SB(sb);
3715         struct ext4_li_request *elr = NULL;
3716         ext4_group_t ngroups = sbi->s_groups_count;
3717         int ret = 0;
3718
3719         mutex_lock(&ext4_li_mtx);
3720         if (sbi->s_li_request != NULL) {
3721                 /*
3722                  * Reset timeout so it can be computed again, because
3723                  * s_li_wait_mult might have changed.
3724                  */
3725                 sbi->s_li_request->lr_timeout = 0;
3726                 goto out;
3727         }
3728
3729         if (!test_opt(sb, PREFETCH_BLOCK_BITMAPS) &&
3730             (first_not_zeroed == ngroups || sb_rdonly(sb) ||
3731              !test_opt(sb, INIT_INODE_TABLE)))
3732                 goto out;
3733
3734         elr = ext4_li_request_new(sb, first_not_zeroed);
3735         if (!elr) {
3736                 ret = -ENOMEM;
3737                 goto out;
3738         }
3739
3740         if (NULL == ext4_li_info) {
3741                 ret = ext4_li_info_new();
3742                 if (ret)
3743                         goto out;
3744         }
3745
3746         mutex_lock(&ext4_li_info->li_list_mtx);
3747         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3748         mutex_unlock(&ext4_li_info->li_list_mtx);
3749
3750         sbi->s_li_request = elr;
3751         /*
3752          * set elr to NULL here since it has been inserted to
3753          * the request_list and the removal and free of it is
3754          * handled by ext4_clear_request_list from now on.
3755          */
3756         elr = NULL;
3757
3758         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3759                 ret = ext4_run_lazyinit_thread();
3760                 if (ret)
3761                         goto out;
3762         }
3763 out:
3764         mutex_unlock(&ext4_li_mtx);
3765         if (ret)
3766                 kfree(elr);
3767         return ret;
3768 }
3769
3770 /*
3771  * We do not need to lock anything since this is called on
3772  * module unload.
3773  */
3774 static void ext4_destroy_lazyinit_thread(void)
3775 {
3776         /*
3777          * If thread exited earlier
3778          * there's nothing to be done.
3779          */
3780         if (!ext4_li_info || !ext4_lazyinit_task)
3781                 return;
3782
3783         kthread_stop(ext4_lazyinit_task);
3784 }
3785
3786 static int set_journal_csum_feature_set(struct super_block *sb)
3787 {
3788         int ret = 1;
3789         int compat, incompat;
3790         struct ext4_sb_info *sbi = EXT4_SB(sb);
3791
3792         if (ext4_has_metadata_csum(sb)) {
3793                 /* journal checksum v3 */
3794                 compat = 0;
3795                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3796         } else {
3797                 /* journal checksum v1 */
3798                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3799                 incompat = 0;
3800         }
3801
3802         jbd2_journal_clear_features(sbi->s_journal,
3803                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3804                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3805                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3806         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3807                 ret = jbd2_journal_set_features(sbi->s_journal,
3808                                 compat, 0,
3809                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3810                                 incompat);
3811         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3812                 ret = jbd2_journal_set_features(sbi->s_journal,
3813                                 compat, 0,
3814                                 incompat);
3815                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3816                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3817         } else {
3818                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3819                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3820         }
3821
3822         return ret;
3823 }
3824
3825 /*
3826  * Note: calculating the overhead so we can be compatible with
3827  * historical BSD practice is quite difficult in the face of
3828  * clusters/bigalloc.  This is because multiple metadata blocks from
3829  * different block group can end up in the same allocation cluster.
3830  * Calculating the exact overhead in the face of clustered allocation
3831  * requires either O(all block bitmaps) in memory or O(number of block
3832  * groups**2) in time.  We will still calculate the superblock for
3833  * older file systems --- and if we come across with a bigalloc file
3834  * system with zero in s_overhead_clusters the estimate will be close to
3835  * correct especially for very large cluster sizes --- but for newer
3836  * file systems, it's better to calculate this figure once at mkfs
3837  * time, and store it in the superblock.  If the superblock value is
3838  * present (even for non-bigalloc file systems), we will use it.
3839  */
3840 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3841                           char *buf)
3842 {
3843         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3844         struct ext4_group_desc  *gdp;
3845         ext4_fsblk_t            first_block, last_block, b;
3846         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3847         int                     s, j, count = 0;
3848
3849         if (!ext4_has_feature_bigalloc(sb))
3850                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3851                         sbi->s_itb_per_group + 2);
3852
3853         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3854                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3855         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3856         for (i = 0; i < ngroups; i++) {
3857                 gdp = ext4_get_group_desc(sb, i, NULL);
3858                 b = ext4_block_bitmap(sb, gdp);
3859                 if (b >= first_block && b <= last_block) {
3860                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3861                         count++;
3862                 }
3863                 b = ext4_inode_bitmap(sb, gdp);
3864                 if (b >= first_block && b <= last_block) {
3865                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3866                         count++;
3867                 }
3868                 b = ext4_inode_table(sb, gdp);
3869                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3870                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3871                                 int c = EXT4_B2C(sbi, b - first_block);
3872                                 ext4_set_bit(c, buf);
3873                                 count++;
3874                         }
3875                 if (i != grp)
3876                         continue;
3877                 s = 0;
3878                 if (ext4_bg_has_super(sb, grp)) {
3879                         ext4_set_bit(s++, buf);
3880                         count++;
3881                 }
3882                 j = ext4_bg_num_gdb(sb, grp);
3883                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3884                         ext4_error(sb, "Invalid number of block group "
3885                                    "descriptor blocks: %d", j);
3886                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3887                 }
3888                 count += j;
3889                 for (; j > 0; j--)
3890                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3891         }
3892         if (!count)
3893                 return 0;
3894         return EXT4_CLUSTERS_PER_GROUP(sb) -
3895                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3896 }
3897
3898 /*
3899  * Compute the overhead and stash it in sbi->s_overhead
3900  */
3901 int ext4_calculate_overhead(struct super_block *sb)
3902 {
3903         struct ext4_sb_info *sbi = EXT4_SB(sb);
3904         struct ext4_super_block *es = sbi->s_es;
3905         struct inode *j_inode;
3906         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3907         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3908         ext4_fsblk_t overhead = 0;
3909         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3910
3911         if (!buf)
3912                 return -ENOMEM;
3913
3914         /*
3915          * Compute the overhead (FS structures).  This is constant
3916          * for a given filesystem unless the number of block groups
3917          * changes so we cache the previous value until it does.
3918          */
3919
3920         /*
3921          * All of the blocks before first_data_block are overhead
3922          */
3923         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3924
3925         /*
3926          * Add the overhead found in each block group
3927          */
3928         for (i = 0; i < ngroups; i++) {
3929                 int blks;
3930
3931                 blks = count_overhead(sb, i, buf);
3932                 overhead += blks;
3933                 if (blks)
3934                         memset(buf, 0, PAGE_SIZE);
3935                 cond_resched();
3936         }
3937
3938         /*
3939          * Add the internal journal blocks whether the journal has been
3940          * loaded or not
3941          */
3942         if (sbi->s_journal && !sbi->s_journal_bdev)
3943                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_total_len);
3944         else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) {
3945                 /* j_inum for internal journal is non-zero */
3946                 j_inode = ext4_get_journal_inode(sb, j_inum);
3947                 if (j_inode) {
3948                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3949                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
3950                         iput(j_inode);
3951                 } else {
3952                         ext4_msg(sb, KERN_ERR, "can't get journal size");
3953                 }
3954         }
3955         sbi->s_overhead = overhead;
3956         smp_wmb();
3957         free_page((unsigned long) buf);
3958         return 0;
3959 }
3960
3961 static void ext4_set_resv_clusters(struct super_block *sb)
3962 {
3963         ext4_fsblk_t resv_clusters;
3964         struct ext4_sb_info *sbi = EXT4_SB(sb);
3965
3966         /*
3967          * There's no need to reserve anything when we aren't using extents.
3968          * The space estimates are exact, there are no unwritten extents,
3969          * hole punching doesn't need new metadata... This is needed especially
3970          * to keep ext2/3 backward compatibility.
3971          */
3972         if (!ext4_has_feature_extents(sb))
3973                 return;
3974         /*
3975          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3976          * This should cover the situations where we can not afford to run
3977          * out of space like for example punch hole, or converting
3978          * unwritten extents in delalloc path. In most cases such
3979          * allocation would require 1, or 2 blocks, higher numbers are
3980          * very rare.
3981          */
3982         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3983                          sbi->s_cluster_bits);
3984
3985         do_div(resv_clusters, 50);
3986         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3987
3988         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3989 }
3990
3991 static const char *ext4_quota_mode(struct super_block *sb)
3992 {
3993 #ifdef CONFIG_QUOTA
3994         if (!ext4_quota_capable(sb))
3995                 return "none";
3996
3997         if (EXT4_SB(sb)->s_journal && ext4_is_quota_journalled(sb))
3998                 return "journalled";
3999         else
4000                 return "writeback";
4001 #else
4002         return "disabled";
4003 #endif
4004 }
4005
4006 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
4007 {
4008         struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev);
4009         char *orig_data = kstrdup(data, GFP_KERNEL);
4010         struct buffer_head *bh, **group_desc;
4011         struct ext4_super_block *es = NULL;
4012         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
4013         struct flex_groups **flex_groups;
4014         ext4_fsblk_t block;
4015         ext4_fsblk_t sb_block = get_sb_block(&data);
4016         ext4_fsblk_t logical_sb_block;
4017         unsigned long offset = 0;
4018         unsigned long journal_devnum = 0;
4019         unsigned long def_mount_opts;
4020         struct inode *root;
4021         const char *descr;
4022         int ret = -ENOMEM;
4023         int blocksize, clustersize;
4024         unsigned int db_count;
4025         unsigned int i;
4026         int needs_recovery, has_huge_files;
4027         __u64 blocks_count;
4028         int err = 0;
4029         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4030         ext4_group_t first_not_zeroed;
4031
4032         if ((data && !orig_data) || !sbi)
4033                 goto out_free_base;
4034
4035         sbi->s_daxdev = dax_dev;
4036         sbi->s_blockgroup_lock =
4037                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
4038         if (!sbi->s_blockgroup_lock)
4039                 goto out_free_base;
4040
4041         sb->s_fs_info = sbi;
4042         sbi->s_sb = sb;
4043         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
4044         sbi->s_sb_block = sb_block;
4045         sbi->s_sectors_written_start =
4046                 part_stat_read(sb->s_bdev, sectors[STAT_WRITE]);
4047
4048         /* Cleanup superblock name */
4049         strreplace(sb->s_id, '/', '!');
4050
4051         /* -EINVAL is default */
4052         ret = -EINVAL;
4053         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
4054         if (!blocksize) {
4055                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
4056                 goto out_fail;
4057         }
4058
4059         /*
4060          * The ext4 superblock will not be buffer aligned for other than 1kB
4061          * block sizes.  We need to calculate the offset from buffer start.
4062          */
4063         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
4064                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
4065                 offset = do_div(logical_sb_block, blocksize);
4066         } else {
4067                 logical_sb_block = sb_block;
4068         }
4069
4070         bh = ext4_sb_bread_unmovable(sb, logical_sb_block);
4071         if (IS_ERR(bh)) {
4072                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
4073                 ret = PTR_ERR(bh);
4074                 goto out_fail;
4075         }
4076         /*
4077          * Note: s_es must be initialized as soon as possible because
4078          *       some ext4 macro-instructions depend on its value
4079          */
4080         es = (struct ext4_super_block *) (bh->b_data + offset);
4081         sbi->s_es = es;
4082         sb->s_magic = le16_to_cpu(es->s_magic);
4083         if (sb->s_magic != EXT4_SUPER_MAGIC)
4084                 goto cantfind_ext4;
4085         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
4086
4087         /* Warn if metadata_csum and gdt_csum are both set. */
4088         if (ext4_has_feature_metadata_csum(sb) &&
4089             ext4_has_feature_gdt_csum(sb))
4090                 ext4_warning(sb, "metadata_csum and uninit_bg are "
4091                              "redundant flags; please run fsck.");
4092
4093         /* Check for a known checksum algorithm */
4094         if (!ext4_verify_csum_type(sb, es)) {
4095                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
4096                          "unknown checksum algorithm.");
4097                 silent = 1;
4098                 goto cantfind_ext4;
4099         }
4100
4101         /* Load the checksum driver */
4102         sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
4103         if (IS_ERR(sbi->s_chksum_driver)) {
4104                 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
4105                 ret = PTR_ERR(sbi->s_chksum_driver);
4106                 sbi->s_chksum_driver = NULL;
4107                 goto failed_mount;
4108         }
4109
4110         /* Check superblock checksum */
4111         if (!ext4_superblock_csum_verify(sb, es)) {
4112                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
4113                          "invalid superblock checksum.  Run e2fsck?");
4114                 silent = 1;
4115                 ret = -EFSBADCRC;
4116                 goto cantfind_ext4;
4117         }
4118
4119         /* Precompute checksum seed for all metadata */
4120         if (ext4_has_feature_csum_seed(sb))
4121                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
4122         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
4123                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
4124                                                sizeof(es->s_uuid));
4125
4126         /* Set defaults before we parse the mount options */
4127         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
4128         set_opt(sb, INIT_INODE_TABLE);
4129         if (def_mount_opts & EXT4_DEFM_DEBUG)
4130                 set_opt(sb, DEBUG);
4131         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
4132                 set_opt(sb, GRPID);
4133         if (def_mount_opts & EXT4_DEFM_UID16)
4134                 set_opt(sb, NO_UID32);
4135         /* xattr user namespace & acls are now defaulted on */
4136         set_opt(sb, XATTR_USER);
4137 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4138         set_opt(sb, POSIX_ACL);
4139 #endif
4140         if (ext4_has_feature_fast_commit(sb))
4141                 set_opt2(sb, JOURNAL_FAST_COMMIT);
4142         /* don't forget to enable journal_csum when metadata_csum is enabled. */
4143         if (ext4_has_metadata_csum(sb))
4144                 set_opt(sb, JOURNAL_CHECKSUM);
4145
4146         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
4147                 set_opt(sb, JOURNAL_DATA);
4148         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
4149                 set_opt(sb, ORDERED_DATA);
4150         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
4151                 set_opt(sb, WRITEBACK_DATA);
4152
4153         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
4154                 set_opt(sb, ERRORS_PANIC);
4155         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
4156                 set_opt(sb, ERRORS_CONT);
4157         else
4158                 set_opt(sb, ERRORS_RO);
4159         /* block_validity enabled by default; disable with noblock_validity */
4160         set_opt(sb, BLOCK_VALIDITY);
4161         if (def_mount_opts & EXT4_DEFM_DISCARD)
4162                 set_opt(sb, DISCARD);
4163
4164         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
4165         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
4166         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
4167         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
4168         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
4169
4170         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
4171                 set_opt(sb, BARRIER);
4172
4173         /*
4174          * enable delayed allocation by default
4175          * Use -o nodelalloc to turn it off
4176          */
4177         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
4178             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
4179                 set_opt(sb, DELALLOC);
4180
4181         /*
4182          * set default s_li_wait_mult for lazyinit, for the case there is
4183          * no mount option specified.
4184          */
4185         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
4186
4187         if (le32_to_cpu(es->s_log_block_size) >
4188             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4189                 ext4_msg(sb, KERN_ERR,
4190                          "Invalid log block size: %u",
4191                          le32_to_cpu(es->s_log_block_size));
4192                 goto failed_mount;
4193         }
4194         if (le32_to_cpu(es->s_log_cluster_size) >
4195             (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4196                 ext4_msg(sb, KERN_ERR,
4197                          "Invalid log cluster size: %u",
4198                          le32_to_cpu(es->s_log_cluster_size));
4199                 goto failed_mount;
4200         }
4201
4202         blocksize = EXT4_MIN_BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
4203
4204         if (blocksize == PAGE_SIZE)
4205                 set_opt(sb, DIOREAD_NOLOCK);
4206
4207         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
4208                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
4209                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
4210         } else {
4211                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
4212                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
4213                 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) {
4214                         ext4_msg(sb, KERN_ERR, "invalid first ino: %u",
4215                                  sbi->s_first_ino);
4216                         goto failed_mount;
4217                 }
4218                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
4219                     (!is_power_of_2(sbi->s_inode_size)) ||
4220                     (sbi->s_inode_size > blocksize)) {
4221                         ext4_msg(sb, KERN_ERR,
4222                                "unsupported inode size: %d",
4223                                sbi->s_inode_size);
4224                         ext4_msg(sb, KERN_ERR, "blocksize: %d", blocksize);
4225                         goto failed_mount;
4226                 }
4227                 /*
4228                  * i_atime_extra is the last extra field available for
4229                  * [acm]times in struct ext4_inode. Checking for that
4230                  * field should suffice to ensure we have extra space
4231                  * for all three.
4232                  */
4233                 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) +
4234                         sizeof(((struct ext4_inode *)0)->i_atime_extra)) {
4235                         sb->s_time_gran = 1;
4236                         sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX;
4237                 } else {
4238                         sb->s_time_gran = NSEC_PER_SEC;
4239                         sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX;
4240                 }
4241                 sb->s_time_min = EXT4_TIMESTAMP_MIN;
4242         }
4243         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
4244                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4245                         EXT4_GOOD_OLD_INODE_SIZE;
4246                 if (ext4_has_feature_extra_isize(sb)) {
4247                         unsigned v, max = (sbi->s_inode_size -
4248                                            EXT4_GOOD_OLD_INODE_SIZE);
4249
4250                         v = le16_to_cpu(es->s_want_extra_isize);
4251                         if (v > max) {
4252                                 ext4_msg(sb, KERN_ERR,
4253                                          "bad s_want_extra_isize: %d", v);
4254                                 goto failed_mount;
4255                         }
4256                         if (sbi->s_want_extra_isize < v)
4257                                 sbi->s_want_extra_isize = v;
4258
4259                         v = le16_to_cpu(es->s_min_extra_isize);
4260                         if (v > max) {
4261                                 ext4_msg(sb, KERN_ERR,
4262                                          "bad s_min_extra_isize: %d", v);
4263                                 goto failed_mount;
4264                         }
4265                         if (sbi->s_want_extra_isize < v)
4266                                 sbi->s_want_extra_isize = v;
4267                 }
4268         }
4269
4270         if (sbi->s_es->s_mount_opts[0]) {
4271                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
4272                                               sizeof(sbi->s_es->s_mount_opts),
4273                                               GFP_KERNEL);
4274                 if (!s_mount_opts)
4275                         goto failed_mount;
4276                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
4277                                    &journal_ioprio, 0)) {
4278                         ext4_msg(sb, KERN_WARNING,
4279                                  "failed to parse options in superblock: %s",
4280                                  s_mount_opts);
4281                 }
4282                 kfree(s_mount_opts);
4283         }
4284         sbi->s_def_mount_opt = sbi->s_mount_opt;
4285         if (!parse_options((char *) data, sb, &journal_devnum,
4286                            &journal_ioprio, 0))
4287                 goto failed_mount;
4288
4289 #ifdef CONFIG_UNICODE
4290         if (ext4_has_feature_casefold(sb) && !sb->s_encoding) {
4291                 const struct ext4_sb_encodings *encoding_info;
4292                 struct unicode_map *encoding;
4293                 __u16 encoding_flags;
4294
4295                 if (ext4_has_feature_encrypt(sb)) {
4296                         ext4_msg(sb, KERN_ERR,
4297                                  "Can't mount with encoding and encryption");
4298                         goto failed_mount;
4299                 }
4300
4301                 if (ext4_sb_read_encoding(es, &encoding_info,
4302                                           &encoding_flags)) {
4303                         ext4_msg(sb, KERN_ERR,
4304                                  "Encoding requested by superblock is unknown");
4305                         goto failed_mount;
4306                 }
4307
4308                 encoding = utf8_load(encoding_info->version);
4309                 if (IS_ERR(encoding)) {
4310                         ext4_msg(sb, KERN_ERR,
4311                                  "can't mount with superblock charset: %s-%s "
4312                                  "not supported by the kernel. flags: 0x%x.",
4313                                  encoding_info->name, encoding_info->version,
4314                                  encoding_flags);
4315                         goto failed_mount;
4316                 }
4317                 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: "
4318                          "%s-%s with flags 0x%hx", encoding_info->name,
4319                          encoding_info->version?:"\b", encoding_flags);
4320
4321                 sb->s_encoding = encoding;
4322                 sb->s_encoding_flags = encoding_flags;
4323         }
4324 #endif
4325
4326         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4327                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with data=journal disables delayed allocation, dioread_nolock, O_DIRECT and fast_commit support!\n");
4328                 /* can't mount with both data=journal and dioread_nolock. */
4329                 clear_opt(sb, DIOREAD_NOLOCK);
4330                 clear_opt2(sb, JOURNAL_FAST_COMMIT);
4331                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4332                         ext4_msg(sb, KERN_ERR, "can't mount with "
4333                                  "both data=journal and delalloc");
4334                         goto failed_mount;
4335                 }
4336                 if (test_opt(sb, DAX_ALWAYS)) {
4337                         ext4_msg(sb, KERN_ERR, "can't mount with "
4338                                  "both data=journal and dax");
4339                         goto failed_mount;
4340                 }
4341                 if (ext4_has_feature_encrypt(sb)) {
4342                         ext4_msg(sb, KERN_WARNING,
4343                                  "encrypted files will use data=ordered "
4344                                  "instead of data journaling mode");
4345                 }
4346                 if (test_opt(sb, DELALLOC))
4347                         clear_opt(sb, DELALLOC);
4348         } else {
4349                 sb->s_iflags |= SB_I_CGROUPWB;
4350         }
4351
4352         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
4353                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
4354
4355         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
4356             (ext4_has_compat_features(sb) ||
4357              ext4_has_ro_compat_features(sb) ||
4358              ext4_has_incompat_features(sb)))
4359                 ext4_msg(sb, KERN_WARNING,
4360                        "feature flags set on rev 0 fs, "
4361                        "running e2fsck is recommended");
4362
4363         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
4364                 set_opt2(sb, HURD_COMPAT);
4365                 if (ext4_has_feature_64bit(sb)) {
4366                         ext4_msg(sb, KERN_ERR,
4367                                  "The Hurd can't support 64-bit file systems");
4368                         goto failed_mount;
4369                 }
4370
4371                 /*
4372                  * ea_inode feature uses l_i_version field which is not
4373                  * available in HURD_COMPAT mode.
4374                  */
4375                 if (ext4_has_feature_ea_inode(sb)) {
4376                         ext4_msg(sb, KERN_ERR,
4377                                  "ea_inode feature is not supported for Hurd");
4378                         goto failed_mount;
4379                 }
4380         }
4381
4382         if (IS_EXT2_SB(sb)) {
4383                 if (ext2_feature_set_ok(sb))
4384                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
4385                                  "using the ext4 subsystem");
4386                 else {
4387                         /*
4388                          * If we're probing be silent, if this looks like
4389                          * it's actually an ext[34] filesystem.
4390                          */
4391                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4392                                 goto failed_mount;
4393                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
4394                                  "to feature incompatibilities");
4395                         goto failed_mount;
4396                 }
4397         }
4398
4399         if (IS_EXT3_SB(sb)) {
4400                 if (ext3_feature_set_ok(sb))
4401                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
4402                                  "using the ext4 subsystem");
4403                 else {
4404                         /*
4405                          * If we're probing be silent, if this looks like
4406                          * it's actually an ext4 filesystem.
4407                          */
4408                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4409                                 goto failed_mount;
4410                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
4411                                  "to feature incompatibilities");
4412                         goto failed_mount;
4413                 }
4414         }
4415
4416         /*
4417          * Check feature flags regardless of the revision level, since we
4418          * previously didn't change the revision level when setting the flags,
4419          * so there is a chance incompat flags are set on a rev 0 filesystem.
4420          */
4421         if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
4422                 goto failed_mount;
4423
4424         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
4425                 ext4_msg(sb, KERN_ERR,
4426                          "Number of reserved GDT blocks insanely large: %d",
4427                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
4428                 goto failed_mount;
4429         }
4430
4431         if (bdev_dax_supported(sb->s_bdev, blocksize))
4432                 set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags);
4433
4434         if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) {
4435                 if (ext4_has_feature_inline_data(sb)) {
4436                         ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
4437                                         " that may contain inline data");
4438                         goto failed_mount;
4439                 }
4440                 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) {
4441                         ext4_msg(sb, KERN_ERR,
4442                                 "DAX unsupported by block device.");
4443                         goto failed_mount;
4444                 }
4445         }
4446
4447         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
4448                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
4449                          es->s_encryption_level);
4450                 goto failed_mount;
4451         }
4452
4453         if (sb->s_blocksize != blocksize) {
4454                 /* Validate the filesystem blocksize */
4455                 if (!sb_set_blocksize(sb, blocksize)) {
4456                         ext4_msg(sb, KERN_ERR, "bad block size %d",
4457                                         blocksize);
4458                         goto failed_mount;
4459                 }
4460
4461                 brelse(bh);
4462                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
4463                 offset = do_div(logical_sb_block, blocksize);
4464                 bh = ext4_sb_bread_unmovable(sb, logical_sb_block);
4465                 if (IS_ERR(bh)) {
4466                         ext4_msg(sb, KERN_ERR,
4467                                "Can't read superblock on 2nd try");
4468                         ret = PTR_ERR(bh);
4469                         bh = NULL;
4470                         goto failed_mount;
4471                 }
4472                 es = (struct ext4_super_block *)(bh->b_data + offset);
4473                 sbi->s_es = es;
4474                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
4475                         ext4_msg(sb, KERN_ERR,
4476                                "Magic mismatch, very weird!");
4477                         goto failed_mount;
4478                 }
4479         }
4480
4481         has_huge_files = ext4_has_feature_huge_file(sb);
4482         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
4483                                                       has_huge_files);
4484         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
4485
4486         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
4487         if (ext4_has_feature_64bit(sb)) {
4488                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
4489                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
4490                     !is_power_of_2(sbi->s_desc_size)) {
4491                         ext4_msg(sb, KERN_ERR,
4492                                "unsupported descriptor size %lu",
4493                                sbi->s_desc_size);
4494                         goto failed_mount;
4495                 }
4496         } else
4497                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
4498
4499         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
4500         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
4501
4502         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
4503         if (sbi->s_inodes_per_block == 0)
4504                 goto cantfind_ext4;
4505         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
4506             sbi->s_inodes_per_group > blocksize * 8) {
4507                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
4508                          sbi->s_inodes_per_group);
4509                 goto failed_mount;
4510         }
4511         sbi->s_itb_per_group = sbi->s_inodes_per_group /
4512                                         sbi->s_inodes_per_block;
4513         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
4514         sbi->s_sbh = bh;
4515         sbi->s_mount_state = le16_to_cpu(es->s_state);
4516         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
4517         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
4518
4519         for (i = 0; i < 4; i++)
4520                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
4521         sbi->s_def_hash_version = es->s_def_hash_version;
4522         if (ext4_has_feature_dir_index(sb)) {
4523                 i = le32_to_cpu(es->s_flags);
4524                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
4525                         sbi->s_hash_unsigned = 3;
4526                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
4527 #ifdef __CHAR_UNSIGNED__
4528                         if (!sb_rdonly(sb))
4529                                 es->s_flags |=
4530                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
4531                         sbi->s_hash_unsigned = 3;
4532 #else
4533                         if (!sb_rdonly(sb))
4534                                 es->s_flags |=
4535                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
4536 #endif
4537                 }
4538         }
4539
4540         /* Handle clustersize */
4541         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
4542         if (ext4_has_feature_bigalloc(sb)) {
4543                 if (clustersize < blocksize) {
4544                         ext4_msg(sb, KERN_ERR,
4545                                  "cluster size (%d) smaller than "
4546                                  "block size (%d)", clustersize, blocksize);
4547                         goto failed_mount;
4548                 }
4549                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
4550                         le32_to_cpu(es->s_log_block_size);
4551                 sbi->s_clusters_per_group =
4552                         le32_to_cpu(es->s_clusters_per_group);
4553                 if (sbi->s_clusters_per_group > blocksize * 8) {
4554                         ext4_msg(sb, KERN_ERR,
4555                                  "#clusters per group too big: %lu",
4556                                  sbi->s_clusters_per_group);
4557                         goto failed_mount;
4558                 }
4559                 if (sbi->s_blocks_per_group !=
4560                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
4561                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
4562                                  "clusters per group (%lu) inconsistent",
4563                                  sbi->s_blocks_per_group,
4564                                  sbi->s_clusters_per_group);
4565                         goto failed_mount;
4566                 }
4567         } else {
4568                 if (clustersize != blocksize) {
4569                         ext4_msg(sb, KERN_ERR,
4570                                  "fragment/cluster size (%d) != "
4571                                  "block size (%d)", clustersize, blocksize);
4572                         goto failed_mount;
4573                 }
4574                 if (sbi->s_blocks_per_group > blocksize * 8) {
4575                         ext4_msg(sb, KERN_ERR,
4576                                  "#blocks per group too big: %lu",
4577                                  sbi->s_blocks_per_group);
4578                         goto failed_mount;
4579                 }
4580                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
4581                 sbi->s_cluster_bits = 0;
4582         }
4583         sbi->s_cluster_ratio = clustersize / blocksize;
4584
4585         /* Do we have standard group size of clustersize * 8 blocks ? */
4586         if (sbi->s_blocks_per_group == clustersize << 3)
4587                 set_opt2(sb, STD_GROUP_SIZE);
4588
4589         /*
4590          * Test whether we have more sectors than will fit in sector_t,
4591          * and whether the max offset is addressable by the page cache.
4592          */
4593         err = generic_check_addressable(sb->s_blocksize_bits,
4594                                         ext4_blocks_count(es));
4595         if (err) {
4596                 ext4_msg(sb, KERN_ERR, "filesystem"
4597                          " too large to mount safely on this system");
4598                 goto failed_mount;
4599         }
4600
4601         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
4602                 goto cantfind_ext4;
4603
4604         /* check blocks count against device size */
4605         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
4606         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
4607                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
4608                        "exceeds size of device (%llu blocks)",
4609                        ext4_blocks_count(es), blocks_count);
4610                 goto failed_mount;
4611         }
4612
4613         /*
4614          * It makes no sense for the first data block to be beyond the end
4615          * of the filesystem.
4616          */
4617         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
4618                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4619                          "block %u is beyond end of filesystem (%llu)",
4620                          le32_to_cpu(es->s_first_data_block),
4621                          ext4_blocks_count(es));
4622                 goto failed_mount;
4623         }
4624         if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) &&
4625             (sbi->s_cluster_ratio == 1)) {
4626                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4627                          "block is 0 with a 1k block and cluster size");
4628                 goto failed_mount;
4629         }
4630
4631         blocks_count = (ext4_blocks_count(es) -
4632                         le32_to_cpu(es->s_first_data_block) +
4633                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
4634         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
4635         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
4636                 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu "
4637                        "(block count %llu, first data block %u, "
4638                        "blocks per group %lu)", blocks_count,
4639                        ext4_blocks_count(es),
4640                        le32_to_cpu(es->s_first_data_block),
4641                        EXT4_BLOCKS_PER_GROUP(sb));
4642                 goto failed_mount;
4643         }
4644         sbi->s_groups_count = blocks_count;
4645         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
4646                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
4647         if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) !=
4648             le32_to_cpu(es->s_inodes_count)) {
4649                 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu",
4650                          le32_to_cpu(es->s_inodes_count),
4651                          ((u64)sbi->s_groups_count * sbi->s_inodes_per_group));
4652                 ret = -EINVAL;
4653                 goto failed_mount;
4654         }
4655         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
4656                    EXT4_DESC_PER_BLOCK(sb);
4657         if (ext4_has_feature_meta_bg(sb)) {
4658                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
4659                         ext4_msg(sb, KERN_WARNING,
4660                                  "first meta block group too large: %u "
4661                                  "(group descriptor block count %u)",
4662                                  le32_to_cpu(es->s_first_meta_bg), db_count);
4663                         goto failed_mount;
4664                 }
4665         }
4666         rcu_assign_pointer(sbi->s_group_desc,
4667                            kvmalloc_array(db_count,
4668                                           sizeof(struct buffer_head *),
4669                                           GFP_KERNEL));
4670         if (sbi->s_group_desc == NULL) {
4671                 ext4_msg(sb, KERN_ERR, "not enough memory");
4672                 ret = -ENOMEM;
4673                 goto failed_mount;
4674         }
4675
4676         bgl_lock_init(sbi->s_blockgroup_lock);
4677
4678         /* Pre-read the descriptors into the buffer cache */
4679         for (i = 0; i < db_count; i++) {
4680                 block = descriptor_loc(sb, logical_sb_block, i);
4681                 ext4_sb_breadahead_unmovable(sb, block);
4682         }
4683
4684         for (i = 0; i < db_count; i++) {
4685                 struct buffer_head *bh;
4686
4687                 block = descriptor_loc(sb, logical_sb_block, i);
4688                 bh = ext4_sb_bread_unmovable(sb, block);
4689                 if (IS_ERR(bh)) {
4690                         ext4_msg(sb, KERN_ERR,
4691                                "can't read group descriptor %d", i);
4692                         db_count = i;
4693                         ret = PTR_ERR(bh);
4694                         goto failed_mount2;
4695                 }
4696                 rcu_read_lock();
4697                 rcu_dereference(sbi->s_group_desc)[i] = bh;
4698                 rcu_read_unlock();
4699         }
4700         sbi->s_gdb_count = db_count;
4701         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
4702                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
4703                 ret = -EFSCORRUPTED;
4704                 goto failed_mount2;
4705         }
4706
4707         timer_setup(&sbi->s_err_report, print_daily_error_info, 0);
4708         spin_lock_init(&sbi->s_error_lock);
4709         INIT_WORK(&sbi->s_error_work, flush_stashed_error_work);
4710
4711         /* Register extent status tree shrinker */
4712         if (ext4_es_register_shrinker(sbi))
4713                 goto failed_mount3;
4714
4715         sbi->s_stripe = ext4_get_stripe_size(sbi);
4716         sbi->s_extent_max_zeroout_kb = 32;
4717
4718         /*
4719          * set up enough so that it can read an inode
4720          */
4721         sb->s_op = &ext4_sops;
4722         sb->s_export_op = &ext4_export_ops;
4723         sb->s_xattr = ext4_xattr_handlers;
4724 #ifdef CONFIG_FS_ENCRYPTION
4725         sb->s_cop = &ext4_cryptops;
4726 #endif
4727 #ifdef CONFIG_FS_VERITY
4728         sb->s_vop = &ext4_verityops;
4729 #endif
4730 #ifdef CONFIG_QUOTA
4731         sb->dq_op = &ext4_quota_operations;
4732         if (ext4_has_feature_quota(sb))
4733                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4734         else
4735                 sb->s_qcop = &ext4_qctl_operations;
4736         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
4737 #endif
4738         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
4739
4740         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
4741         mutex_init(&sbi->s_orphan_lock);
4742
4743         /* Initialize fast commit stuff */
4744         atomic_set(&sbi->s_fc_subtid, 0);
4745         atomic_set(&sbi->s_fc_ineligible_updates, 0);
4746         INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_MAIN]);
4747         INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_STAGING]);
4748         INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_MAIN]);
4749         INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_STAGING]);
4750         sbi->s_fc_bytes = 0;
4751         ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE);
4752         ext4_clear_mount_flag(sb, EXT4_MF_FC_COMMITTING);
4753         spin_lock_init(&sbi->s_fc_lock);
4754         memset(&sbi->s_fc_stats, 0, sizeof(sbi->s_fc_stats));
4755         sbi->s_fc_replay_state.fc_regions = NULL;
4756         sbi->s_fc_replay_state.fc_regions_size = 0;
4757         sbi->s_fc_replay_state.fc_regions_used = 0;
4758         sbi->s_fc_replay_state.fc_regions_valid = 0;
4759         sbi->s_fc_replay_state.fc_modified_inodes = NULL;
4760         sbi->s_fc_replay_state.fc_modified_inodes_size = 0;
4761         sbi->s_fc_replay_state.fc_modified_inodes_used = 0;
4762
4763         sb->s_root = NULL;
4764
4765         needs_recovery = (es->s_last_orphan != 0 ||
4766                           ext4_has_feature_journal_needs_recovery(sb));
4767
4768         if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb))
4769                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
4770                         goto failed_mount3a;
4771
4772         /*
4773          * The first inode we look at is the journal inode.  Don't try
4774          * root first: it may be modified in the journal!
4775          */
4776         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
4777                 err = ext4_load_journal(sb, es, journal_devnum);
4778                 if (err)
4779                         goto failed_mount3a;
4780         } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
4781                    ext4_has_feature_journal_needs_recovery(sb)) {
4782                 ext4_msg(sb, KERN_ERR, "required journal recovery "
4783                        "suppressed and not mounted read-only");
4784                 goto failed_mount_wq;
4785         } else {
4786                 /* Nojournal mode, all journal mount options are illegal */
4787                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
4788                         ext4_msg(sb, KERN_ERR, "can't mount with "
4789                                  "journal_checksum, fs mounted w/o journal");
4790                         goto failed_mount_wq;
4791                 }
4792                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4793                         ext4_msg(sb, KERN_ERR, "can't mount with "
4794                                  "journal_async_commit, fs mounted w/o journal");
4795                         goto failed_mount_wq;
4796                 }
4797                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
4798                         ext4_msg(sb, KERN_ERR, "can't mount with "
4799                                  "commit=%lu, fs mounted w/o journal",
4800                                  sbi->s_commit_interval / HZ);
4801                         goto failed_mount_wq;
4802                 }
4803                 if (EXT4_MOUNT_DATA_FLAGS &
4804                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
4805                         ext4_msg(sb, KERN_ERR, "can't mount with "
4806                                  "data=, fs mounted w/o journal");
4807                         goto failed_mount_wq;
4808                 }
4809                 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM;
4810                 clear_opt(sb, JOURNAL_CHECKSUM);
4811                 clear_opt(sb, DATA_FLAGS);
4812                 clear_opt2(sb, JOURNAL_FAST_COMMIT);
4813                 sbi->s_journal = NULL;
4814                 needs_recovery = 0;
4815                 goto no_journal;
4816         }
4817
4818         if (ext4_has_feature_64bit(sb) &&
4819             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4820                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4821                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4822                 goto failed_mount_wq;
4823         }
4824
4825         if (!set_journal_csum_feature_set(sb)) {
4826                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4827                          "feature set");
4828                 goto failed_mount_wq;
4829         }
4830
4831         if (test_opt2(sb, JOURNAL_FAST_COMMIT) &&
4832                 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4833                                           JBD2_FEATURE_INCOMPAT_FAST_COMMIT)) {
4834                 ext4_msg(sb, KERN_ERR,
4835                         "Failed to set fast commit journal feature");
4836                 goto failed_mount_wq;
4837         }
4838
4839         /* We have now updated the journal if required, so we can
4840          * validate the data journaling mode. */
4841         switch (test_opt(sb, DATA_FLAGS)) {
4842         case 0:
4843                 /* No mode set, assume a default based on the journal
4844                  * capabilities: ORDERED_DATA if the journal can
4845                  * cope, else JOURNAL_DATA
4846                  */
4847                 if (jbd2_journal_check_available_features
4848                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4849                         set_opt(sb, ORDERED_DATA);
4850                         sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
4851                 } else {
4852                         set_opt(sb, JOURNAL_DATA);
4853                         sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
4854                 }
4855                 break;
4856
4857         case EXT4_MOUNT_ORDERED_DATA:
4858         case EXT4_MOUNT_WRITEBACK_DATA:
4859                 if (!jbd2_journal_check_available_features
4860                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4861                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4862                                "requested data journaling mode");
4863                         goto failed_mount_wq;
4864                 }
4865                 break;
4866         default:
4867                 break;
4868         }
4869
4870         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4871             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4872                 ext4_msg(sb, KERN_ERR, "can't mount with "
4873                         "journal_async_commit in data=ordered mode");
4874                 goto failed_mount_wq;
4875         }
4876
4877         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4878
4879         sbi->s_journal->j_submit_inode_data_buffers =
4880                 ext4_journal_submit_inode_data_buffers;
4881         sbi->s_journal->j_finish_inode_data_buffers =
4882                 ext4_journal_finish_inode_data_buffers;
4883
4884 no_journal:
4885         if (!test_opt(sb, NO_MBCACHE)) {
4886                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
4887                 if (!sbi->s_ea_block_cache) {
4888                         ext4_msg(sb, KERN_ERR,
4889                                  "Failed to create ea_block_cache");
4890                         goto failed_mount_wq;
4891                 }
4892
4893                 if (ext4_has_feature_ea_inode(sb)) {
4894                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
4895                         if (!sbi->s_ea_inode_cache) {
4896                                 ext4_msg(sb, KERN_ERR,
4897                                          "Failed to create ea_inode_cache");
4898                                 goto failed_mount_wq;
4899                         }
4900                 }
4901         }
4902
4903         if (ext4_has_feature_verity(sb) && blocksize != PAGE_SIZE) {
4904                 ext4_msg(sb, KERN_ERR, "Unsupported blocksize for fs-verity");
4905                 goto failed_mount_wq;
4906         }
4907
4908         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) &&
4909             !ext4_has_feature_encrypt(sb)) {
4910                 ext4_set_feature_encrypt(sb);
4911                 ext4_commit_super(sb);
4912         }
4913
4914         /*
4915          * Get the # of file system overhead blocks from the
4916          * superblock if present.
4917          */
4918         if (es->s_overhead_clusters)
4919                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4920         else {
4921                 err = ext4_calculate_overhead(sb);
4922                 if (err)
4923                         goto failed_mount_wq;
4924         }
4925
4926         /*
4927          * The maximum number of concurrent works can be high and
4928          * concurrency isn't really necessary.  Limit it to 1.
4929          */
4930         EXT4_SB(sb)->rsv_conversion_wq =
4931                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4932         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4933                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4934                 ret = -ENOMEM;
4935                 goto failed_mount4;
4936         }
4937
4938         /*
4939          * The jbd2_journal_load will have done any necessary log recovery,
4940          * so we can safely mount the rest of the filesystem now.
4941          */
4942
4943         root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL);
4944         if (IS_ERR(root)) {
4945                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4946                 ret = PTR_ERR(root);
4947                 root = NULL;
4948                 goto failed_mount4;
4949         }
4950         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4951                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4952                 iput(root);
4953                 goto failed_mount4;
4954         }
4955
4956         sb->s_root = d_make_root(root);
4957         if (!sb->s_root) {
4958                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4959                 ret = -ENOMEM;
4960                 goto failed_mount4;
4961         }
4962
4963         ret = ext4_setup_super(sb, es, sb_rdonly(sb));
4964         if (ret == -EROFS) {
4965                 sb->s_flags |= SB_RDONLY;
4966                 ret = 0;
4967         } else if (ret)
4968                 goto failed_mount4a;
4969
4970         ext4_set_resv_clusters(sb);
4971
4972         if (test_opt(sb, BLOCK_VALIDITY)) {
4973                 err = ext4_setup_system_zone(sb);
4974                 if (err) {
4975                         ext4_msg(sb, KERN_ERR, "failed to initialize system "
4976                                  "zone (%d)", err);
4977                         goto failed_mount4a;
4978                 }
4979         }
4980         ext4_fc_replay_cleanup(sb);
4981
4982         ext4_ext_init(sb);
4983         err = ext4_mb_init(sb);
4984         if (err) {
4985                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4986                          err);
4987                 goto failed_mount5;
4988         }
4989
4990         /*
4991          * We can only set up the journal commit callback once
4992          * mballoc is initialized
4993          */
4994         if (sbi->s_journal)
4995                 sbi->s_journal->j_commit_callback =
4996                         ext4_journal_commit_callback;
4997
4998         block = ext4_count_free_clusters(sb);
4999         ext4_free_blocks_count_set(sbi->s_es,
5000                                    EXT4_C2B(sbi, block));
5001         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
5002                                   GFP_KERNEL);
5003         if (!err) {
5004                 unsigned long freei = ext4_count_free_inodes(sb);
5005                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
5006                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
5007                                           GFP_KERNEL);
5008         }
5009         if (!err)
5010                 err = percpu_counter_init(&sbi->s_dirs_counter,
5011                                           ext4_count_dirs(sb), GFP_KERNEL);
5012         if (!err)
5013                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
5014                                           GFP_KERNEL);
5015         if (!err)
5016                 err = percpu_counter_init(&sbi->s_sra_exceeded_retry_limit, 0,
5017                                           GFP_KERNEL);
5018         if (!err)
5019                 err = percpu_init_rwsem(&sbi->s_writepages_rwsem);
5020
5021         if (err) {
5022                 ext4_msg(sb, KERN_ERR, "insufficient memory");
5023                 goto failed_mount6;
5024         }
5025
5026         if (ext4_has_feature_flex_bg(sb))
5027                 if (!ext4_fill_flex_info(sb)) {
5028                         ext4_msg(sb, KERN_ERR,
5029                                "unable to initialize "
5030                                "flex_bg meta info!");
5031                         goto failed_mount6;
5032                 }
5033
5034         err = ext4_register_li_request(sb, first_not_zeroed);
5035         if (err)
5036                 goto failed_mount6;
5037
5038         err = ext4_register_sysfs(sb);
5039         if (err)
5040                 goto failed_mount7;
5041
5042 #ifdef CONFIG_QUOTA
5043         /* Enable quota usage during mount. */
5044         if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
5045                 err = ext4_enable_quotas(sb);
5046                 if (err)
5047                         goto failed_mount8;
5048         }
5049 #endif  /* CONFIG_QUOTA */
5050
5051         /*
5052          * Save the original bdev mapping's wb_err value which could be
5053          * used to detect the metadata async write error.
5054          */
5055         spin_lock_init(&sbi->s_bdev_wb_lock);
5056         errseq_check_and_advance(&sb->s_bdev->bd_inode->i_mapping->wb_err,
5057                                  &sbi->s_bdev_wb_err);
5058         sb->s_bdev->bd_super = sb;
5059         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
5060         ext4_orphan_cleanup(sb, es);
5061         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
5062         if (needs_recovery) {
5063                 ext4_msg(sb, KERN_INFO, "recovery complete");
5064                 err = ext4_mark_recovery_complete(sb, es);
5065                 if (err)
5066                         goto failed_mount8;
5067         }
5068         if (EXT4_SB(sb)->s_journal) {
5069                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
5070                         descr = " journalled data mode";
5071                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
5072                         descr = " ordered data mode";
5073                 else
5074                         descr = " writeback data mode";
5075         } else
5076                 descr = "out journal";
5077
5078         if (test_opt(sb, DISCARD)) {
5079                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
5080                 if (!blk_queue_discard(q))
5081                         ext4_msg(sb, KERN_WARNING,
5082                                  "mounting with \"discard\" option, but "
5083                                  "the device does not support discard");
5084         }
5085
5086         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
5087                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
5088                          "Opts: %.*s%s%s. Quota mode: %s.", descr,
5089                          (int) sizeof(sbi->s_es->s_mount_opts),
5090                          sbi->s_es->s_mount_opts,
5091                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data,
5092                          ext4_quota_mode(sb));
5093
5094         if (es->s_error_count)
5095                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
5096
5097         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
5098         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
5099         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
5100         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
5101         atomic_set(&sbi->s_warning_count, 0);
5102         atomic_set(&sbi->s_msg_count, 0);
5103
5104         kfree(orig_data);
5105         return 0;
5106
5107 cantfind_ext4:
5108         if (!silent)
5109                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
5110         goto failed_mount;
5111
5112 failed_mount8:
5113         ext4_unregister_sysfs(sb);
5114         kobject_put(&sbi->s_kobj);
5115 failed_mount7:
5116         ext4_unregister_li_request(sb);
5117 failed_mount6:
5118         ext4_mb_release(sb);
5119         rcu_read_lock();
5120         flex_groups = rcu_dereference(sbi->s_flex_groups);
5121         if (flex_groups) {
5122                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
5123                         kvfree(flex_groups[i]);
5124                 kvfree(flex_groups);
5125         }
5126         rcu_read_unlock();
5127         percpu_counter_destroy(&sbi->s_freeclusters_counter);
5128         percpu_counter_destroy(&sbi->s_freeinodes_counter);
5129         percpu_counter_destroy(&sbi->s_dirs_counter);
5130         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
5131         percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit);
5132         percpu_free_rwsem(&sbi->s_writepages_rwsem);
5133 failed_mount5:
5134         ext4_ext_release(sb);
5135         ext4_release_system_zone(sb);
5136 failed_mount4a:
5137         dput(sb->s_root);
5138         sb->s_root = NULL;
5139 failed_mount4:
5140         ext4_msg(sb, KERN_ERR, "mount failed");
5141         if (EXT4_SB(sb)->rsv_conversion_wq)
5142                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
5143 failed_mount_wq:
5144         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
5145         sbi->s_ea_inode_cache = NULL;
5146
5147         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
5148         sbi->s_ea_block_cache = NULL;
5149
5150         if (sbi->s_journal) {
5151                 jbd2_journal_destroy(sbi->s_journal);
5152                 sbi->s_journal = NULL;
5153         }
5154 failed_mount3a:
5155         ext4_es_unregister_shrinker(sbi);
5156 failed_mount3:
5157         flush_work(&sbi->s_error_work);
5158         del_timer_sync(&sbi->s_err_report);
5159         if (sbi->s_mmp_tsk)
5160                 kthread_stop(sbi->s_mmp_tsk);
5161 failed_mount2:
5162         rcu_read_lock();
5163         group_desc = rcu_dereference(sbi->s_group_desc);
5164         for (i = 0; i < db_count; i++)
5165                 brelse(group_desc[i]);
5166         kvfree(group_desc);
5167         rcu_read_unlock();
5168 failed_mount:
5169         if (sbi->s_chksum_driver)
5170                 crypto_free_shash(sbi->s_chksum_driver);
5171
5172 #ifdef CONFIG_UNICODE
5173         utf8_unload(sb->s_encoding);
5174 #endif
5175
5176 #ifdef CONFIG_QUOTA
5177         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5178                 kfree(get_qf_name(sb, sbi, i));
5179 #endif
5180         fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy);
5181         ext4_blkdev_remove(sbi);
5182         brelse(bh);
5183 out_fail:
5184         sb->s_fs_info = NULL;
5185         kfree(sbi->s_blockgroup_lock);
5186 out_free_base:
5187         kfree(sbi);
5188         kfree(orig_data);
5189         fs_put_dax(dax_dev);
5190         return err ? err : ret;
5191 }
5192
5193 /*
5194  * Setup any per-fs journal parameters now.  We'll do this both on
5195  * initial mount, once the journal has been initialised but before we've
5196  * done any recovery; and again on any subsequent remount.
5197  */
5198 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
5199 {
5200         struct ext4_sb_info *sbi = EXT4_SB(sb);
5201
5202         journal->j_commit_interval = sbi->s_commit_interval;
5203         journal->j_min_batch_time = sbi->s_min_batch_time;
5204         journal->j_max_batch_time = sbi->s_max_batch_time;
5205         ext4_fc_init(sb, journal);
5206
5207         write_lock(&journal->j_state_lock);
5208         if (test_opt(sb, BARRIER))
5209                 journal->j_flags |= JBD2_BARRIER;
5210         else
5211                 journal->j_flags &= ~JBD2_BARRIER;
5212         if (test_opt(sb, DATA_ERR_ABORT))
5213                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
5214         else
5215                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
5216         write_unlock(&journal->j_state_lock);
5217 }
5218
5219 static struct inode *ext4_get_journal_inode(struct super_block *sb,
5220                                              unsigned int journal_inum)
5221 {
5222         struct inode *journal_inode;
5223
5224         /*
5225          * Test for the existence of a valid inode on disk.  Bad things
5226          * happen if we iget() an unused inode, as the subsequent iput()
5227          * will try to delete it.
5228          */
5229         journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL);
5230         if (IS_ERR(journal_inode)) {
5231                 ext4_msg(sb, KERN_ERR, "no journal found");
5232                 return NULL;
5233         }
5234         if (!journal_inode->i_nlink) {
5235                 make_bad_inode(journal_inode);
5236                 iput(journal_inode);
5237                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
5238                 return NULL;
5239         }
5240
5241         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
5242                   journal_inode, journal_inode->i_size);
5243         if (!S_ISREG(journal_inode->i_mode)) {
5244                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
5245                 iput(journal_inode);
5246                 return NULL;
5247         }
5248         return journal_inode;
5249 }
5250
5251 static journal_t *ext4_get_journal(struct super_block *sb,
5252                                    unsigned int journal_inum)
5253 {
5254         struct inode *journal_inode;
5255         journal_t *journal;
5256
5257         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5258                 return NULL;
5259
5260         journal_inode = ext4_get_journal_inode(sb, journal_inum);
5261         if (!journal_inode)
5262                 return NULL;
5263
5264         journal = jbd2_journal_init_inode(journal_inode);
5265         if (!journal) {
5266                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
5267                 iput(journal_inode);
5268                 return NULL;
5269         }
5270         journal->j_private = sb;
5271         ext4_init_journal_params(sb, journal);
5272         return journal;
5273 }
5274
5275 static journal_t *ext4_get_dev_journal(struct super_block *sb,
5276                                        dev_t j_dev)
5277 {
5278         struct buffer_head *bh;
5279         journal_t *journal;
5280         ext4_fsblk_t start;
5281         ext4_fsblk_t len;
5282         int hblock, blocksize;
5283         ext4_fsblk_t sb_block;
5284         unsigned long offset;
5285         struct ext4_super_block *es;
5286         struct block_device *bdev;
5287
5288         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5289                 return NULL;
5290
5291         bdev = ext4_blkdev_get(j_dev, sb);
5292         if (bdev == NULL)
5293                 return NULL;
5294
5295         blocksize = sb->s_blocksize;
5296         hblock = bdev_logical_block_size(bdev);
5297         if (blocksize < hblock) {
5298                 ext4_msg(sb, KERN_ERR,
5299                         "blocksize too small for journal device");
5300                 goto out_bdev;
5301         }
5302
5303         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
5304         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
5305         set_blocksize(bdev, blocksize);
5306         if (!(bh = __bread(bdev, sb_block, blocksize))) {
5307                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
5308                        "external journal");
5309                 goto out_bdev;
5310         }
5311
5312         es = (struct ext4_super_block *) (bh->b_data + offset);
5313         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
5314             !(le32_to_cpu(es->s_feature_incompat) &
5315               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
5316                 ext4_msg(sb, KERN_ERR, "external journal has "
5317                                         "bad superblock");
5318                 brelse(bh);
5319                 goto out_bdev;
5320         }
5321
5322         if ((le32_to_cpu(es->s_feature_ro_compat) &
5323              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
5324             es->s_checksum != ext4_superblock_csum(sb, es)) {
5325                 ext4_msg(sb, KERN_ERR, "external journal has "
5326                                        "corrupt superblock");
5327                 brelse(bh);
5328                 goto out_bdev;
5329         }
5330
5331         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
5332                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
5333                 brelse(bh);
5334                 goto out_bdev;
5335         }
5336
5337         len = ext4_blocks_count(es);
5338         start = sb_block + 1;
5339         brelse(bh);     /* we're done with the superblock */
5340
5341         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
5342                                         start, len, blocksize);
5343         if (!journal) {
5344                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
5345                 goto out_bdev;
5346         }
5347         journal->j_private = sb;
5348         if (ext4_read_bh_lock(journal->j_sb_buffer, REQ_META | REQ_PRIO, true)) {
5349                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
5350                 goto out_journal;
5351         }
5352         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
5353                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
5354                                         "user (unsupported) - %d",
5355                         be32_to_cpu(journal->j_superblock->s_nr_users));
5356                 goto out_journal;
5357         }
5358         EXT4_SB(sb)->s_journal_bdev = bdev;
5359         ext4_init_journal_params(sb, journal);
5360         return journal;
5361
5362 out_journal:
5363         jbd2_journal_destroy(journal);
5364 out_bdev:
5365         ext4_blkdev_put(bdev);
5366         return NULL;
5367 }
5368
5369 static int ext4_load_journal(struct super_block *sb,
5370                              struct ext4_super_block *es,
5371                              unsigned long journal_devnum)
5372 {
5373         journal_t *journal;
5374         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
5375         dev_t journal_dev;
5376         int err = 0;
5377         int really_read_only;
5378         int journal_dev_ro;
5379
5380         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5381                 return -EFSCORRUPTED;
5382
5383         if (journal_devnum &&
5384             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5385                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
5386                         "numbers have changed");
5387                 journal_dev = new_decode_dev(journal_devnum);
5388         } else
5389                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
5390
5391         if (journal_inum && journal_dev) {
5392                 ext4_msg(sb, KERN_ERR,
5393                          "filesystem has both journal inode and journal device!");
5394                 return -EINVAL;
5395         }
5396
5397         if (journal_inum) {
5398                 journal = ext4_get_journal(sb, journal_inum);
5399                 if (!journal)
5400                         return -EINVAL;
5401         } else {
5402                 journal = ext4_get_dev_journal(sb, journal_dev);
5403                 if (!journal)
5404                         return -EINVAL;
5405         }
5406
5407         journal_dev_ro = bdev_read_only(journal->j_dev);
5408         really_read_only = bdev_read_only(sb->s_bdev) | journal_dev_ro;
5409
5410         if (journal_dev_ro && !sb_rdonly(sb)) {
5411                 ext4_msg(sb, KERN_ERR,
5412                          "journal device read-only, try mounting with '-o ro'");
5413                 err = -EROFS;
5414                 goto err_out;
5415         }
5416
5417         /*
5418          * Are we loading a blank journal or performing recovery after a
5419          * crash?  For recovery, we need to check in advance whether we
5420          * can get read-write access to the device.
5421          */
5422         if (ext4_has_feature_journal_needs_recovery(sb)) {
5423                 if (sb_rdonly(sb)) {
5424                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
5425                                         "required on readonly filesystem");
5426                         if (really_read_only) {
5427                                 ext4_msg(sb, KERN_ERR, "write access "
5428                                         "unavailable, cannot proceed "
5429                                         "(try mounting with noload)");
5430                                 err = -EROFS;
5431                                 goto err_out;
5432                         }
5433                         ext4_msg(sb, KERN_INFO, "write access will "
5434                                "be enabled during recovery");
5435                 }
5436         }
5437
5438         if (!(journal->j_flags & JBD2_BARRIER))
5439                 ext4_msg(sb, KERN_INFO, "barriers disabled");
5440
5441         if (!ext4_has_feature_journal_needs_recovery(sb))
5442                 err = jbd2_journal_wipe(journal, !really_read_only);
5443         if (!err) {
5444                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
5445                 if (save)
5446                         memcpy(save, ((char *) es) +
5447                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
5448                 err = jbd2_journal_load(journal);
5449                 if (save)
5450                         memcpy(((char *) es) + EXT4_S_ERR_START,
5451                                save, EXT4_S_ERR_LEN);
5452                 kfree(save);
5453         }
5454
5455         if (err) {
5456                 ext4_msg(sb, KERN_ERR, "error loading journal");
5457                 goto err_out;
5458         }
5459
5460         EXT4_SB(sb)->s_journal = journal;
5461         err = ext4_clear_journal_err(sb, es);
5462         if (err) {
5463                 EXT4_SB(sb)->s_journal = NULL;
5464                 jbd2_journal_destroy(journal);
5465                 return err;
5466         }
5467
5468         if (!really_read_only && journal_devnum &&
5469             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5470                 es->s_journal_dev = cpu_to_le32(journal_devnum);
5471
5472                 /* Make sure we flush the recovery flag to disk. */
5473                 ext4_commit_super(sb);
5474         }
5475
5476         return 0;
5477
5478 err_out:
5479         jbd2_journal_destroy(journal);
5480         return err;
5481 }
5482
5483 /* Copy state of EXT4_SB(sb) into buffer for on-disk superblock */
5484 static void ext4_update_super(struct super_block *sb)
5485 {
5486         struct ext4_sb_info *sbi = EXT4_SB(sb);
5487         struct ext4_super_block *es = sbi->s_es;
5488         struct buffer_head *sbh = sbi->s_sbh;
5489
5490         lock_buffer(sbh);
5491         /*
5492          * If the file system is mounted read-only, don't update the
5493          * superblock write time.  This avoids updating the superblock
5494          * write time when we are mounting the root file system
5495          * read/only but we need to replay the journal; at that point,
5496          * for people who are east of GMT and who make their clock
5497          * tick in localtime for Windows bug-for-bug compatibility,
5498          * the clock is set in the future, and this will cause e2fsck
5499          * to complain and force a full file system check.
5500          */
5501         if (!(sb->s_flags & SB_RDONLY))
5502                 ext4_update_tstamp(es, s_wtime);
5503         es->s_kbytes_written =
5504                 cpu_to_le64(sbi->s_kbytes_written +
5505                     ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) -
5506                       sbi->s_sectors_written_start) >> 1));
5507         if (percpu_counter_initialized(&sbi->s_freeclusters_counter))
5508                 ext4_free_blocks_count_set(es,
5509                         EXT4_C2B(sbi, percpu_counter_sum_positive(
5510                                 &sbi->s_freeclusters_counter)));
5511         if (percpu_counter_initialized(&sbi->s_freeinodes_counter))
5512                 es->s_free_inodes_count =
5513                         cpu_to_le32(percpu_counter_sum_positive(
5514                                 &sbi->s_freeinodes_counter));
5515         /* Copy error information to the on-disk superblock */
5516         spin_lock(&sbi->s_error_lock);
5517         if (sbi->s_add_error_count > 0) {
5518                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
5519                 if (!es->s_first_error_time && !es->s_first_error_time_hi) {
5520                         __ext4_update_tstamp(&es->s_first_error_time,
5521                                              &es->s_first_error_time_hi,
5522                                              sbi->s_first_error_time);
5523                         strncpy(es->s_first_error_func, sbi->s_first_error_func,
5524                                 sizeof(es->s_first_error_func));
5525                         es->s_first_error_line =
5526                                 cpu_to_le32(sbi->s_first_error_line);
5527                         es->s_first_error_ino =
5528                                 cpu_to_le32(sbi->s_first_error_ino);
5529                         es->s_first_error_block =
5530                                 cpu_to_le64(sbi->s_first_error_block);
5531                         es->s_first_error_errcode =
5532                                 ext4_errno_to_code(sbi->s_first_error_code);
5533                 }
5534                 __ext4_update_tstamp(&es->s_last_error_time,
5535                                      &es->s_last_error_time_hi,
5536                                      sbi->s_last_error_time);
5537                 strncpy(es->s_last_error_func, sbi->s_last_error_func,
5538                         sizeof(es->s_last_error_func));
5539                 es->s_last_error_line = cpu_to_le32(sbi->s_last_error_line);
5540                 es->s_last_error_ino = cpu_to_le32(sbi->s_last_error_ino);
5541                 es->s_last_error_block = cpu_to_le64(sbi->s_last_error_block);
5542                 es->s_last_error_errcode =
5543                                 ext4_errno_to_code(sbi->s_last_error_code);
5544                 /*
5545                  * Start the daily error reporting function if it hasn't been
5546                  * started already
5547                  */
5548                 if (!es->s_error_count)
5549                         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);
5550                 le32_add_cpu(&es->s_error_count, sbi->s_add_error_count);
5551                 sbi->s_add_error_count = 0;
5552         }
5553         spin_unlock(&sbi->s_error_lock);
5554
5555         ext4_superblock_csum_set(sb);
5556         unlock_buffer(sbh);
5557 }
5558
5559 static int ext4_commit_super(struct super_block *sb)
5560 {
5561         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
5562         int error = 0;
5563
5564         if (!sbh || block_device_ejected(sb))
5565                 return error;
5566
5567         ext4_update_super(sb);
5568
5569         if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
5570                 /*
5571                  * Oh, dear.  A previous attempt to write the
5572                  * superblock failed.  This could happen because the
5573                  * USB device was yanked out.  Or it could happen to
5574                  * be a transient write error and maybe the block will
5575                  * be remapped.  Nothing we can do but to retry the
5576                  * write and hope for the best.
5577                  */
5578                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
5579                        "superblock detected");
5580                 clear_buffer_write_io_error(sbh);
5581                 set_buffer_uptodate(sbh);
5582         }
5583         BUFFER_TRACE(sbh, "marking dirty");
5584         mark_buffer_dirty(sbh);
5585         error = __sync_dirty_buffer(sbh,
5586                 REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
5587         if (buffer_write_io_error(sbh)) {
5588                 ext4_msg(sb, KERN_ERR, "I/O error while writing "
5589                        "superblock");
5590                 clear_buffer_write_io_error(sbh);
5591                 set_buffer_uptodate(sbh);
5592         }
5593         return error;
5594 }
5595
5596 /*
5597  * Have we just finished recovery?  If so, and if we are mounting (or
5598  * remounting) the filesystem readonly, then we will end up with a
5599  * consistent fs on disk.  Record that fact.
5600  */
5601 static int ext4_mark_recovery_complete(struct super_block *sb,
5602                                        struct ext4_super_block *es)
5603 {
5604         int err;
5605         journal_t *journal = EXT4_SB(sb)->s_journal;
5606
5607         if (!ext4_has_feature_journal(sb)) {
5608                 if (journal != NULL) {
5609                         ext4_error(sb, "Journal got removed while the fs was "
5610                                    "mounted!");
5611                         return -EFSCORRUPTED;
5612                 }
5613                 return 0;
5614         }
5615         jbd2_journal_lock_updates(journal);
5616         err = jbd2_journal_flush(journal);
5617         if (err < 0)
5618                 goto out;
5619
5620         if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) {
5621                 ext4_clear_feature_journal_needs_recovery(sb);
5622                 ext4_commit_super(sb);
5623         }
5624 out:
5625         jbd2_journal_unlock_updates(journal);
5626         return err;
5627 }
5628
5629 /*
5630  * If we are mounting (or read-write remounting) a filesystem whose journal
5631  * has recorded an error from a previous lifetime, move that error to the
5632  * main filesystem now.
5633  */
5634 static int ext4_clear_journal_err(struct super_block *sb,
5635                                    struct ext4_super_block *es)
5636 {
5637         journal_t *journal;
5638         int j_errno;
5639         const char *errstr;
5640
5641         if (!ext4_has_feature_journal(sb)) {
5642                 ext4_error(sb, "Journal got removed while the fs was mounted!");
5643                 return -EFSCORRUPTED;
5644         }
5645
5646         journal = EXT4_SB(sb)->s_journal;
5647
5648         /*
5649          * Now check for any error status which may have been recorded in the
5650          * journal by a prior ext4_error() or ext4_abort()
5651          */
5652
5653         j_errno = jbd2_journal_errno(journal);
5654         if (j_errno) {
5655                 char nbuf[16];
5656
5657                 errstr = ext4_decode_error(sb, j_errno, nbuf);
5658                 ext4_warning(sb, "Filesystem error recorded "
5659                              "from previous mount: %s", errstr);
5660                 ext4_warning(sb, "Marking fs in need of filesystem check.");
5661
5662                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
5663                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
5664                 ext4_commit_super(sb);
5665
5666                 jbd2_journal_clear_err(journal);
5667                 jbd2_journal_update_sb_errno(journal);
5668         }
5669         return 0;
5670 }
5671
5672 /*
5673  * Force the running and committing transactions to commit,
5674  * and wait on the commit.
5675  */
5676 int ext4_force_commit(struct super_block *sb)
5677 {
5678         journal_t *journal;
5679
5680         if (sb_rdonly(sb))
5681                 return 0;
5682
5683         journal = EXT4_SB(sb)->s_journal;
5684         return ext4_journal_force_commit(journal);
5685 }
5686
5687 static int ext4_sync_fs(struct super_block *sb, int wait)
5688 {
5689         int ret = 0;
5690         tid_t target;
5691         bool needs_barrier = false;
5692         struct ext4_sb_info *sbi = EXT4_SB(sb);
5693
5694         if (unlikely(ext4_forced_shutdown(sbi)))
5695                 return 0;
5696
5697         trace_ext4_sync_fs(sb, wait);
5698         flush_workqueue(sbi->rsv_conversion_wq);
5699         /*
5700          * Writeback quota in non-journalled quota case - journalled quota has
5701          * no dirty dquots
5702          */
5703         dquot_writeback_dquots(sb, -1);
5704         /*
5705          * Data writeback is possible w/o journal transaction, so barrier must
5706          * being sent at the end of the function. But we can skip it if
5707          * transaction_commit will do it for us.
5708          */
5709         if (sbi->s_journal) {
5710                 target = jbd2_get_latest_transaction(sbi->s_journal);
5711                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
5712                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
5713                         needs_barrier = true;
5714
5715                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
5716                         if (wait)
5717                                 ret = jbd2_log_wait_commit(sbi->s_journal,
5718                                                            target);
5719                 }
5720         } else if (wait && test_opt(sb, BARRIER))
5721                 needs_barrier = true;
5722         if (needs_barrier) {
5723                 int err;
5724                 err = blkdev_issue_flush(sb->s_bdev);
5725                 if (!ret)
5726                         ret = err;
5727         }
5728
5729         return ret;
5730 }
5731
5732 /*
5733  * LVM calls this function before a (read-only) snapshot is created.  This
5734  * gives us a chance to flush the journal completely and mark the fs clean.
5735  *
5736  * Note that only this function cannot bring a filesystem to be in a clean
5737  * state independently. It relies on upper layer to stop all data & metadata
5738  * modifications.
5739  */
5740 static int ext4_freeze(struct super_block *sb)
5741 {
5742         int error = 0;
5743         journal_t *journal;
5744
5745         if (sb_rdonly(sb))
5746                 return 0;
5747
5748         journal = EXT4_SB(sb)->s_journal;
5749
5750         if (journal) {
5751                 /* Now we set up the journal barrier. */
5752                 jbd2_journal_lock_updates(journal);
5753
5754                 /*
5755                  * Don't clear the needs_recovery flag if we failed to
5756                  * flush the journal.
5757                  */
5758                 error = jbd2_journal_flush(journal);
5759                 if (error < 0)
5760                         goto out;
5761
5762                 /* Journal blocked and flushed, clear needs_recovery flag. */
5763                 ext4_clear_feature_journal_needs_recovery(sb);
5764         }
5765
5766         error = ext4_commit_super(sb);
5767 out:
5768         if (journal)
5769                 /* we rely on upper layer to stop further updates */
5770                 jbd2_journal_unlock_updates(journal);
5771         return error;
5772 }
5773
5774 /*
5775  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
5776  * flag here, even though the filesystem is not technically dirty yet.
5777  */
5778 static int ext4_unfreeze(struct super_block *sb)
5779 {
5780         if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
5781                 return 0;
5782
5783         if (EXT4_SB(sb)->s_journal) {
5784                 /* Reset the needs_recovery flag before the fs is unlocked. */
5785                 ext4_set_feature_journal_needs_recovery(sb);
5786         }
5787
5788         ext4_commit_super(sb);
5789         return 0;
5790 }
5791
5792 /*
5793  * Structure to save mount options for ext4_remount's benefit
5794  */
5795 struct ext4_mount_options {
5796         unsigned long s_mount_opt;
5797         unsigned long s_mount_opt2;
5798         kuid_t s_resuid;
5799         kgid_t s_resgid;
5800         unsigned long s_commit_interval;
5801         u32 s_min_batch_time, s_max_batch_time;
5802 #ifdef CONFIG_QUOTA
5803         int s_jquota_fmt;
5804         char *s_qf_names[EXT4_MAXQUOTAS];
5805 #endif
5806 };
5807
5808 static int ext4_remount(struct super_block *sb, int *flags, char *data)
5809 {
5810         struct ext4_super_block *es;
5811         struct ext4_sb_info *sbi = EXT4_SB(sb);
5812         unsigned long old_sb_flags, vfs_flags;
5813         struct ext4_mount_options old_opts;
5814         int enable_quota = 0;
5815         ext4_group_t g;
5816         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
5817         int err = 0;
5818 #ifdef CONFIG_QUOTA
5819         int i, j;
5820         char *to_free[EXT4_MAXQUOTAS];
5821 #endif
5822         char *orig_data = kstrdup(data, GFP_KERNEL);
5823
5824         if (data && !orig_data)
5825                 return -ENOMEM;
5826
5827         /* Store the original options */
5828         old_sb_flags = sb->s_flags;
5829         old_opts.s_mount_opt = sbi->s_mount_opt;
5830         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
5831         old_opts.s_resuid = sbi->s_resuid;
5832         old_opts.s_resgid = sbi->s_resgid;
5833         old_opts.s_commit_interval = sbi->s_commit_interval;
5834         old_opts.s_min_batch_time = sbi->s_min_batch_time;
5835         old_opts.s_max_batch_time = sbi->s_max_batch_time;
5836 #ifdef CONFIG_QUOTA
5837         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
5838         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5839                 if (sbi->s_qf_names[i]) {
5840                         char *qf_name = get_qf_name(sb, sbi, i);
5841
5842                         old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL);
5843                         if (!old_opts.s_qf_names[i]) {
5844                                 for (j = 0; j < i; j++)
5845                                         kfree(old_opts.s_qf_names[j]);
5846                                 kfree(orig_data);
5847                                 return -ENOMEM;
5848                         }
5849                 } else
5850                         old_opts.s_qf_names[i] = NULL;
5851 #endif
5852         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
5853                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
5854
5855         /*
5856          * Some options can be enabled by ext4 and/or by VFS mount flag
5857          * either way we need to make sure it matches in both *flags and
5858          * s_flags. Copy those selected flags from *flags to s_flags
5859          */
5860         vfs_flags = SB_LAZYTIME | SB_I_VERSION;
5861         sb->s_flags = (sb->s_flags & ~vfs_flags) | (*flags & vfs_flags);
5862
5863         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
5864                 err = -EINVAL;
5865                 goto restore_opts;
5866         }
5867
5868         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
5869             test_opt(sb, JOURNAL_CHECKSUM)) {
5870                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
5871                          "during remount not supported; ignoring");
5872                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
5873         }
5874
5875         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
5876                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
5877                         ext4_msg(sb, KERN_ERR, "can't mount with "
5878                                  "both data=journal and delalloc");
5879                         err = -EINVAL;
5880                         goto restore_opts;
5881                 }
5882                 if (test_opt(sb, DIOREAD_NOLOCK)) {
5883                         ext4_msg(sb, KERN_ERR, "can't mount with "
5884                                  "both data=journal and dioread_nolock");
5885                         err = -EINVAL;
5886                         goto restore_opts;
5887                 }
5888         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
5889                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5890                         ext4_msg(sb, KERN_ERR, "can't mount with "
5891                                 "journal_async_commit in data=ordered mode");
5892                         err = -EINVAL;
5893                         goto restore_opts;
5894                 }
5895         }
5896
5897         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
5898                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
5899                 err = -EINVAL;
5900                 goto restore_opts;
5901         }
5902
5903         if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED))
5904                 ext4_abort(sb, EXT4_ERR_ESHUTDOWN, "Abort forced by user");
5905
5906         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
5907                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
5908
5909         es = sbi->s_es;
5910
5911         if (sbi->s_journal) {
5912                 ext4_init_journal_params(sb, sbi->s_journal);
5913                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
5914         }
5915
5916         /* Flush outstanding errors before changing fs state */
5917         flush_work(&sbi->s_error_work);
5918
5919         if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) {
5920                 if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED)) {
5921                         err = -EROFS;
5922                         goto restore_opts;
5923                 }
5924
5925                 if (*flags & SB_RDONLY) {
5926                         err = sync_filesystem(sb);
5927                         if (err < 0)
5928                                 goto restore_opts;
5929                         err = dquot_suspend(sb, -1);
5930                         if (err < 0)
5931                                 goto restore_opts;
5932
5933                         /*
5934                          * First of all, the unconditional stuff we have to do
5935                          * to disable replay of the journal when we next remount
5936                          */
5937                         sb->s_flags |= SB_RDONLY;
5938
5939                         /*
5940                          * OK, test if we are remounting a valid rw partition
5941                          * readonly, and if so set the rdonly flag and then
5942                          * mark the partition as valid again.
5943                          */
5944                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5945                             (sbi->s_mount_state & EXT4_VALID_FS))
5946                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5947
5948                         if (sbi->s_journal) {
5949                                 /*
5950                                  * We let remount-ro finish even if marking fs
5951                                  * as clean failed...
5952                                  */
5953                                 ext4_mark_recovery_complete(sb, es);
5954                         }
5955                         if (sbi->s_mmp_tsk)
5956                                 kthread_stop(sbi->s_mmp_tsk);
5957                 } else {
5958                         /* Make sure we can mount this feature set readwrite */
5959                         if (ext4_has_feature_readonly(sb) ||
5960                             !ext4_feature_set_ok(sb, 0)) {
5961                                 err = -EROFS;
5962                                 goto restore_opts;
5963                         }
5964                         /*
5965                          * Make sure the group descriptor checksums
5966                          * are sane.  If they aren't, refuse to remount r/w.
5967                          */
5968                         for (g = 0; g < sbi->s_groups_count; g++) {
5969                                 struct ext4_group_desc *gdp =
5970                                         ext4_get_group_desc(sb, g, NULL);
5971
5972                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5973                                         ext4_msg(sb, KERN_ERR,
5974                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5975                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
5976                                                le16_to_cpu(gdp->bg_checksum));
5977                                         err = -EFSBADCRC;
5978                                         goto restore_opts;
5979                                 }
5980                         }
5981
5982                         /*
5983                          * If we have an unprocessed orphan list hanging
5984                          * around from a previously readonly bdev mount,
5985                          * require a full umount/remount for now.
5986                          */
5987                         if (es->s_last_orphan) {
5988                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5989                                        "remount RDWR because of unprocessed "
5990                                        "orphan inode list.  Please "
5991                                        "umount/remount instead");
5992                                 err = -EINVAL;
5993                                 goto restore_opts;
5994                         }
5995
5996                         /*
5997                          * Mounting a RDONLY partition read-write, so reread
5998                          * and store the current valid flag.  (It may have
5999                          * been changed by e2fsck since we originally mounted
6000                          * the partition.)
6001                          */
6002                         if (sbi->s_journal) {
6003                                 err = ext4_clear_journal_err(sb, es);
6004                                 if (err)
6005                                         goto restore_opts;
6006                         }
6007                         sbi->s_mount_state = le16_to_cpu(es->s_state);
6008
6009                         err = ext4_setup_super(sb, es, 0);
6010                         if (err)
6011                                 goto restore_opts;
6012
6013                         sb->s_flags &= ~SB_RDONLY;
6014                         if (ext4_has_feature_mmp(sb))
6015                                 if (ext4_multi_mount_protect(sb,
6016                                                 le64_to_cpu(es->s_mmp_block))) {
6017                                         err = -EROFS;
6018                                         goto restore_opts;
6019                                 }
6020                         enable_quota = 1;
6021                 }
6022         }
6023
6024         /*
6025          * Reinitialize lazy itable initialization thread based on
6026          * current settings
6027          */
6028         if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
6029                 ext4_unregister_li_request(sb);
6030         else {
6031                 ext4_group_t first_not_zeroed;
6032                 first_not_zeroed = ext4_has_uninit_itable(sb);
6033                 ext4_register_li_request(sb, first_not_zeroed);
6034         }
6035
6036         /*
6037          * Handle creation of system zone data early because it can fail.
6038          * Releasing of existing data is done when we are sure remount will
6039          * succeed.
6040          */
6041         if (test_opt(sb, BLOCK_VALIDITY) && !sbi->s_system_blks) {
6042                 err = ext4_setup_system_zone(sb);
6043                 if (err)
6044                         goto restore_opts;
6045         }
6046
6047         if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) {
6048                 err = ext4_commit_super(sb);
6049                 if (err)
6050                         goto restore_opts;
6051         }
6052
6053 #ifdef CONFIG_QUOTA
6054         /* Release old quota file names */
6055         for (i = 0; i < EXT4_MAXQUOTAS; i++)
6056                 kfree(old_opts.s_qf_names[i]);
6057         if (enable_quota) {
6058                 if (sb_any_quota_suspended(sb))
6059                         dquot_resume(sb, -1);
6060                 else if (ext4_has_feature_quota(sb)) {
6061                         err = ext4_enable_quotas(sb);
6062                         if (err)
6063                                 goto restore_opts;
6064                 }
6065         }
6066 #endif
6067         if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks)
6068                 ext4_release_system_zone(sb);
6069
6070         /*
6071          * Some options can be enabled by ext4 and/or by VFS mount flag
6072          * either way we need to make sure it matches in both *flags and
6073          * s_flags. Copy those selected flags from s_flags to *flags
6074          */
6075         *flags = (*flags & ~vfs_flags) | (sb->s_flags & vfs_flags);
6076
6077         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s. Quota mode: %s.",
6078                  orig_data, ext4_quota_mode(sb));
6079         kfree(orig_data);
6080         return 0;
6081
6082 restore_opts:
6083         sb->s_flags = old_sb_flags;
6084         sbi->s_mount_opt = old_opts.s_mount_opt;
6085         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
6086         sbi->s_resuid = old_opts.s_resuid;
6087         sbi->s_resgid = old_opts.s_resgid;
6088         sbi->s_commit_interval = old_opts.s_commit_interval;
6089         sbi->s_min_batch_time = old_opts.s_min_batch_time;
6090         sbi->s_max_batch_time = old_opts.s_max_batch_time;
6091         if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks)
6092                 ext4_release_system_zone(sb);
6093 #ifdef CONFIG_QUOTA
6094         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
6095         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
6096                 to_free[i] = get_qf_name(sb, sbi, i);
6097                 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]);
6098         }
6099         synchronize_rcu();
6100         for (i = 0; i < EXT4_MAXQUOTAS; i++)
6101                 kfree(to_free[i]);
6102 #endif
6103         kfree(orig_data);
6104         return err;
6105 }
6106
6107 #ifdef CONFIG_QUOTA
6108 static int ext4_statfs_project(struct super_block *sb,
6109                                kprojid_t projid, struct kstatfs *buf)
6110 {
6111         struct kqid qid;
6112         struct dquot *dquot;
6113         u64 limit;
6114         u64 curblock;
6115
6116         qid = make_kqid_projid(projid);
6117         dquot = dqget(sb, qid);
6118         if (IS_ERR(dquot))
6119                 return PTR_ERR(dquot);
6120         spin_lock(&dquot->dq_dqb_lock);
6121
6122         limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit,
6123                              dquot->dq_dqb.dqb_bhardlimit);
6124         limit >>= sb->s_blocksize_bits;
6125
6126         if (limit && buf->f_blocks > limit) {
6127                 curblock = (dquot->dq_dqb.dqb_curspace +
6128                             dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits;
6129                 buf->f_blocks = limit;
6130                 buf->f_bfree = buf->f_bavail =
6131                         (buf->f_blocks > curblock) ?
6132                          (buf->f_blocks - curblock) : 0;
6133         }
6134
6135         limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit,
6136                              dquot->dq_dqb.dqb_ihardlimit);
6137         if (limit && buf->f_files > limit) {
6138                 buf->f_files = limit;
6139                 buf->f_ffree =
6140                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
6141                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
6142         }
6143
6144         spin_unlock(&dquot->dq_dqb_lock);
6145         dqput(dquot);
6146         return 0;
6147 }
6148 #endif
6149
6150 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
6151 {
6152         struct super_block *sb = dentry->d_sb;
6153         struct ext4_sb_info *sbi = EXT4_SB(sb);
6154         struct ext4_super_block *es = sbi->s_es;
6155         ext4_fsblk_t overhead = 0, resv_blocks;
6156         s64 bfree;
6157         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
6158
6159         if (!test_opt(sb, MINIX_DF))
6160                 overhead = sbi->s_overhead;
6161
6162         buf->f_type = EXT4_SUPER_MAGIC;
6163         buf->f_bsize = sb->s_blocksize;
6164         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
6165         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
6166                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
6167         /* prevent underflow in case that few free space is available */
6168         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
6169         buf->f_bavail = buf->f_bfree -
6170                         (ext4_r_blocks_count(es) + resv_blocks);
6171         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
6172                 buf->f_bavail = 0;
6173         buf->f_files = le32_to_cpu(es->s_inodes_count);
6174         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
6175         buf->f_namelen = EXT4_NAME_LEN;
6176         buf->f_fsid = uuid_to_fsid(es->s_uuid);
6177
6178 #ifdef CONFIG_QUOTA
6179         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
6180             sb_has_quota_limits_enabled(sb, PRJQUOTA))
6181                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
6182 #endif
6183         return 0;
6184 }
6185
6186
6187 #ifdef CONFIG_QUOTA
6188
6189 /*
6190  * Helper functions so that transaction is started before we acquire dqio_sem
6191  * to keep correct lock ordering of transaction > dqio_sem
6192  */
6193 static inline struct inode *dquot_to_inode(struct dquot *dquot)
6194 {
6195         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
6196 }
6197
6198 static int ext4_write_dquot(struct dquot *dquot)
6199 {
6200         int ret, err;
6201         handle_t *handle;
6202         struct inode *inode;
6203
6204         inode = dquot_to_inode(dquot);
6205         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
6206                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
6207         if (IS_ERR(handle))
6208                 return PTR_ERR(handle);
6209         ret = dquot_commit(dquot);
6210         err = ext4_journal_stop(handle);
6211         if (!ret)
6212                 ret = err;
6213         return ret;
6214 }
6215
6216 static int ext4_acquire_dquot(struct dquot *dquot)
6217 {
6218         int ret, err;
6219         handle_t *handle;
6220
6221         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
6222                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
6223         if (IS_ERR(handle))
6224                 return PTR_ERR(handle);
6225         ret = dquot_acquire(dquot);
6226         err = ext4_journal_stop(handle);
6227         if (!ret)
6228                 ret = err;
6229         return ret;
6230 }
6231
6232 static int ext4_release_dquot(struct dquot *dquot)
6233 {
6234         int ret, err;
6235         handle_t *handle;
6236
6237         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
6238                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
6239         if (IS_ERR(handle)) {
6240                 /* Release dquot anyway to avoid endless cycle in dqput() */
6241                 dquot_release(dquot);
6242                 return PTR_ERR(handle);
6243         }
6244         ret = dquot_release(dquot);
6245         err = ext4_journal_stop(handle);
6246         if (!ret)
6247                 ret = err;
6248         return ret;
6249 }
6250
6251 static int ext4_mark_dquot_dirty(struct dquot *dquot)
6252 {
6253         struct super_block *sb = dquot->dq_sb;
6254
6255         if (ext4_is_quota_journalled(sb)) {
6256                 dquot_mark_dquot_dirty(dquot);
6257                 return ext4_write_dquot(dquot);
6258         } else {
6259                 return dquot_mark_dquot_dirty(dquot);
6260         }
6261 }
6262
6263 static int ext4_write_info(struct super_block *sb, int type)
6264 {
6265         int ret, err;
6266         handle_t *handle;
6267
6268         /* Data block + inode block */
6269         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
6270         if (IS_ERR(handle))
6271                 return PTR_ERR(handle);
6272         ret = dquot_commit_info(sb, type);
6273         err = ext4_journal_stop(handle);
6274         if (!ret)
6275                 ret = err;
6276         return ret;
6277 }
6278
6279 /*
6280  * Turn on quotas during mount time - we need to find
6281  * the quota file and such...
6282  */
6283 static int ext4_quota_on_mount(struct super_block *sb, int type)
6284 {
6285         return dquot_quota_on_mount(sb, get_qf_name(sb, EXT4_SB(sb), type),
6286                                         EXT4_SB(sb)->s_jquota_fmt, type);
6287 }
6288
6289 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
6290 {
6291         struct ext4_inode_info *ei = EXT4_I(inode);
6292
6293         /* The first argument of lockdep_set_subclass has to be
6294          * *exactly* the same as the argument to init_rwsem() --- in
6295          * this case, in init_once() --- or lockdep gets unhappy
6296          * because the name of the lock is set using the
6297          * stringification of the argument to init_rwsem().
6298          */
6299         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
6300         lockdep_set_subclass(&ei->i_data_sem, subclass);
6301 }
6302
6303 /*
6304  * Standard function to be called on quota_on
6305  */
6306 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
6307                          const struct path *path)
6308 {
6309         int err;
6310
6311         if (!test_opt(sb, QUOTA))
6312                 return -EINVAL;
6313
6314         /* Quotafile not on the same filesystem? */
6315         if (path->dentry->d_sb != sb)
6316                 return -EXDEV;
6317
6318         /* Quota already enabled for this file? */
6319         if (IS_NOQUOTA(d_inode(path->dentry)))
6320                 return -EBUSY;
6321
6322         /* Journaling quota? */
6323         if (EXT4_SB(sb)->s_qf_names[type]) {
6324                 /* Quotafile not in fs root? */
6325                 if (path->dentry->d_parent != sb->s_root)
6326                         ext4_msg(sb, KERN_WARNING,
6327                                 "Quota file not on filesystem root. "
6328                                 "Journaled quota will not work");
6329                 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
6330         } else {
6331                 /*
6332                  * Clear the flag just in case mount options changed since
6333                  * last time.
6334                  */
6335                 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
6336         }
6337
6338         /*
6339          * When we journal data on quota file, we have to flush journal to see
6340          * all updates to the file when we bypass pagecache...
6341          */
6342         if (EXT4_SB(sb)->s_journal &&
6343             ext4_should_journal_data(d_inode(path->dentry))) {
6344                 /*
6345                  * We don't need to lock updates but journal_flush() could
6346                  * otherwise be livelocked...
6347                  */
6348                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
6349                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
6350                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
6351                 if (err)
6352                         return err;
6353         }
6354
6355         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
6356         err = dquot_quota_on(sb, type, format_id, path);
6357         if (err) {
6358                 lockdep_set_quota_inode(path->dentry->d_inode,
6359                                              I_DATA_SEM_NORMAL);
6360         } else {
6361                 struct inode *inode = d_inode(path->dentry);
6362                 handle_t *handle;
6363
6364                 /*
6365                  * Set inode flags to prevent userspace from messing with quota
6366                  * files. If this fails, we return success anyway since quotas
6367                  * are already enabled and this is not a hard failure.
6368                  */
6369                 inode_lock(inode);
6370                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
6371                 if (IS_ERR(handle))
6372                         goto unlock_inode;
6373                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
6374                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
6375                                 S_NOATIME | S_IMMUTABLE);
6376                 err = ext4_mark_inode_dirty(handle, inode);
6377                 ext4_journal_stop(handle);
6378         unlock_inode:
6379                 inode_unlock(inode);
6380         }
6381         return err;
6382 }
6383
6384 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
6385                              unsigned int flags)
6386 {
6387         int err;
6388         struct inode *qf_inode;
6389         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
6390                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
6391                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
6392                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
6393         };
6394
6395         BUG_ON(!ext4_has_feature_quota(sb));
6396
6397         if (!qf_inums[type])
6398                 return -EPERM;
6399
6400         qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL);
6401         if (IS_ERR(qf_inode)) {
6402                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
6403                 return PTR_ERR(qf_inode);
6404         }
6405
6406         /* Don't account quota for quota files to avoid recursion */
6407         qf_inode->i_flags |= S_NOQUOTA;
6408         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
6409         err = dquot_load_quota_inode(qf_inode, type, format_id, flags);
6410         if (err)
6411                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
6412         iput(qf_inode);
6413
6414         return err;
6415 }
6416
6417 /* Enable usage tracking for all quota types. */
6418 static int ext4_enable_quotas(struct super_block *sb)
6419 {
6420         int type, err = 0;
6421         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
6422                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
6423                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
6424                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
6425         };
6426         bool quota_mopt[EXT4_MAXQUOTAS] = {
6427                 test_opt(sb, USRQUOTA),
6428                 test_opt(sb, GRPQUOTA),
6429                 test_opt(sb, PRJQUOTA),
6430         };
6431
6432         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
6433         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
6434                 if (qf_inums[type]) {
6435                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
6436                                 DQUOT_USAGE_ENABLED |
6437                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
6438                         if (err) {
6439                                 ext4_warning(sb,
6440                                         "Failed to enable quota tracking "
6441                                         "(type=%d, err=%d). Please run "
6442                                         "e2fsck to fix.", type, err);
6443                                 for (type--; type >= 0; type--)
6444                                         dquot_quota_off(sb, type);
6445
6446                                 return err;
6447                         }
6448                 }
6449         }
6450         return 0;
6451 }
6452
6453 static int ext4_quota_off(struct super_block *sb, int type)
6454 {
6455         struct inode *inode = sb_dqopt(sb)->files[type];
6456         handle_t *handle;
6457         int err;
6458
6459         /* Force all delayed allocation blocks to be allocated.
6460          * Caller already holds s_umount sem */
6461         if (test_opt(sb, DELALLOC))
6462                 sync_filesystem(sb);
6463
6464         if (!inode || !igrab(inode))
6465                 goto out;
6466
6467         err = dquot_quota_off(sb, type);
6468         if (err || ext4_has_feature_quota(sb))
6469                 goto out_put;
6470
6471         inode_lock(inode);
6472         /*
6473          * Update modification times of quota files when userspace can
6474          * start looking at them. If we fail, we return success anyway since
6475          * this is not a hard failure and quotas are already disabled.
6476          */
6477         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
6478         if (IS_ERR(handle)) {
6479                 err = PTR_ERR(handle);
6480                 goto out_unlock;
6481         }
6482         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
6483         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
6484         inode->i_mtime = inode->i_ctime = current_time(inode);
6485         err = ext4_mark_inode_dirty(handle, inode);
6486         ext4_journal_stop(handle);
6487 out_unlock:
6488         inode_unlock(inode);
6489 out_put:
6490         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
6491         iput(inode);
6492         return err;
6493 out:
6494         return dquot_quota_off(sb, type);
6495 }
6496
6497 /* Read data from quotafile - avoid pagecache and such because we cannot afford
6498  * acquiring the locks... As quota files are never truncated and quota code
6499  * itself serializes the operations (and no one else should touch the files)
6500  * we don't have to be afraid of races */
6501 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
6502                                size_t len, loff_t off)
6503 {
6504         struct inode *inode = sb_dqopt(sb)->files[type];
6505         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
6506         int offset = off & (sb->s_blocksize - 1);
6507         int tocopy;
6508         size_t toread;
6509         struct buffer_head *bh;
6510         loff_t i_size = i_size_read(inode);
6511
6512         if (off > i_size)
6513                 return 0;
6514         if (off+len > i_size)
6515                 len = i_size-off;
6516         toread = len;
6517         while (toread > 0) {
6518                 tocopy = sb->s_blocksize - offset < toread ?
6519                                 sb->s_blocksize - offset : toread;
6520                 bh = ext4_bread(NULL, inode, blk, 0);
6521                 if (IS_ERR(bh))
6522                         return PTR_ERR(bh);
6523                 if (!bh)        /* A hole? */
6524                         memset(data, 0, tocopy);
6525                 else
6526                         memcpy(data, bh->b_data+offset, tocopy);
6527                 brelse(bh);
6528                 offset = 0;
6529                 toread -= tocopy;
6530                 data += tocopy;
6531                 blk++;
6532         }
6533         return len;
6534 }
6535
6536 /* Write to quotafile (we know the transaction is already started and has
6537  * enough credits) */
6538 static ssize_t ext4_quota_write(struct super_block *sb, int type,
6539                                 const char *data, size_t len, loff_t off)
6540 {
6541         struct inode *inode = sb_dqopt(sb)->files[type];
6542         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
6543         int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1);
6544         int retries = 0;
6545         struct buffer_head *bh;
6546         handle_t *handle = journal_current_handle();
6547
6548         if (EXT4_SB(sb)->s_journal && !handle) {
6549                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
6550                         " cancelled because transaction is not started",
6551                         (unsigned long long)off, (unsigned long long)len);
6552                 return -EIO;
6553         }
6554         /*
6555          * Since we account only one data block in transaction credits,
6556          * then it is impossible to cross a block boundary.
6557          */
6558         if (sb->s_blocksize - offset < len) {
6559                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
6560                         " cancelled because not block aligned",
6561                         (unsigned long long)off, (unsigned long long)len);
6562                 return -EIO;
6563         }
6564
6565         do {
6566                 bh = ext4_bread(handle, inode, blk,
6567                                 EXT4_GET_BLOCKS_CREATE |
6568                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
6569         } while (PTR_ERR(bh) == -ENOSPC &&
6570                  ext4_should_retry_alloc(inode->i_sb, &retries));
6571         if (IS_ERR(bh))
6572                 return PTR_ERR(bh);
6573         if (!bh)
6574                 goto out;
6575         BUFFER_TRACE(bh, "get write access");
6576         err = ext4_journal_get_write_access(handle, bh);
6577         if (err) {
6578                 brelse(bh);
6579                 return err;
6580         }
6581         lock_buffer(bh);
6582         memcpy(bh->b_data+offset, data, len);
6583         flush_dcache_page(bh->b_page);
6584         unlock_buffer(bh);
6585         err = ext4_handle_dirty_metadata(handle, NULL, bh);
6586         brelse(bh);
6587 out:
6588         if (inode->i_size < off + len) {
6589                 i_size_write(inode, off + len);
6590                 EXT4_I(inode)->i_disksize = inode->i_size;
6591                 err2 = ext4_mark_inode_dirty(handle, inode);
6592                 if (unlikely(err2 && !err))
6593                         err = err2;
6594         }
6595         return err ? err : len;
6596 }
6597 #endif
6598
6599 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
6600                        const char *dev_name, void *data)
6601 {
6602         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
6603 }
6604
6605 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
6606 static inline void register_as_ext2(void)
6607 {
6608         int err = register_filesystem(&ext2_fs_type);
6609         if (err)
6610                 printk(KERN_WARNING
6611                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
6612 }
6613
6614 static inline void unregister_as_ext2(void)
6615 {
6616         unregister_filesystem(&ext2_fs_type);
6617 }
6618
6619 static inline int ext2_feature_set_ok(struct super_block *sb)
6620 {
6621         if (ext4_has_unknown_ext2_incompat_features(sb))
6622                 return 0;
6623         if (sb_rdonly(sb))
6624                 return 1;
6625         if (ext4_has_unknown_ext2_ro_compat_features(sb))
6626                 return 0;
6627         return 1;
6628 }
6629 #else
6630 static inline void register_as_ext2(void) { }
6631 static inline void unregister_as_ext2(void) { }
6632 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
6633 #endif
6634
6635 static inline void register_as_ext3(void)
6636 {
6637         int err = register_filesystem(&ext3_fs_type);
6638         if (err)
6639                 printk(KERN_WARNING
6640                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
6641 }
6642
6643 static inline void unregister_as_ext3(void)
6644 {
6645         unregister_filesystem(&ext3_fs_type);
6646 }
6647
6648 static inline int ext3_feature_set_ok(struct super_block *sb)
6649 {
6650         if (ext4_has_unknown_ext3_incompat_features(sb))
6651                 return 0;
6652         if (!ext4_has_feature_journal(sb))
6653                 return 0;
6654         if (sb_rdonly(sb))
6655                 return 1;
6656         if (ext4_has_unknown_ext3_ro_compat_features(sb))
6657                 return 0;
6658         return 1;
6659 }
6660
6661 static struct file_system_type ext4_fs_type = {
6662         .owner          = THIS_MODULE,
6663         .name           = "ext4",
6664         .mount          = ext4_mount,
6665         .kill_sb        = kill_block_super,
6666         .fs_flags       = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
6667 };
6668 MODULE_ALIAS_FS("ext4");
6669
6670 /* Shared across all ext4 file systems */
6671 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
6672
6673 static int __init ext4_init_fs(void)
6674 {
6675         int i, err;
6676
6677         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
6678         ext4_li_info = NULL;
6679
6680         /* Build-time check for flags consistency */
6681         ext4_check_flag_values();
6682
6683         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
6684                 init_waitqueue_head(&ext4__ioend_wq[i]);
6685
6686         err = ext4_init_es();
6687         if (err)
6688                 return err;
6689
6690         err = ext4_init_pending();
6691         if (err)
6692                 goto out7;
6693
6694         err = ext4_init_post_read_processing();
6695         if (err)
6696                 goto out6;
6697
6698         err = ext4_init_pageio();
6699         if (err)
6700                 goto out5;
6701
6702         err = ext4_init_system_zone();
6703         if (err)
6704                 goto out4;
6705
6706         err = ext4_init_sysfs();
6707         if (err)
6708                 goto out3;
6709
6710         err = ext4_init_mballoc();
6711         if (err)
6712                 goto out2;
6713         err = init_inodecache();
6714         if (err)
6715                 goto out1;
6716
6717         err = ext4_fc_init_dentry_cache();
6718         if (err)
6719                 goto out05;
6720
6721         register_as_ext3();
6722         register_as_ext2();
6723         err = register_filesystem(&ext4_fs_type);
6724         if (err)
6725                 goto out;
6726
6727         return 0;
6728 out:
6729         unregister_as_ext2();
6730         unregister_as_ext3();
6731 out05:
6732         destroy_inodecache();
6733 out1:
6734         ext4_exit_mballoc();
6735 out2:
6736         ext4_exit_sysfs();
6737 out3:
6738         ext4_exit_system_zone();
6739 out4:
6740         ext4_exit_pageio();
6741 out5:
6742         ext4_exit_post_read_processing();
6743 out6:
6744         ext4_exit_pending();
6745 out7:
6746         ext4_exit_es();
6747
6748         return err;
6749 }
6750
6751 static void __exit ext4_exit_fs(void)
6752 {
6753         ext4_destroy_lazyinit_thread();
6754         unregister_as_ext2();
6755         unregister_as_ext3();
6756         unregister_filesystem(&ext4_fs_type);
6757         destroy_inodecache();
6758         ext4_exit_mballoc();
6759         ext4_exit_sysfs();
6760         ext4_exit_system_zone();
6761         ext4_exit_pageio();
6762         ext4_exit_post_read_processing();
6763         ext4_exit_es();
6764         ext4_exit_pending();
6765 }
6766
6767 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
6768 MODULE_DESCRIPTION("Fourth Extended Filesystem");
6769 MODULE_LICENSE("GPL");
6770 MODULE_SOFTDEP("pre: crc32c");
6771 module_init(ext4_init_fs)
6772 module_exit(ext4_exit_fs)