fs/ext4/super.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/ext4/super.c
   4  *
   5  * Copyright (C) 1992, 1993, 1994, 1995
   6  * Remy Card (card@masi.ibp.fr)
   7  * Laboratoire MASI - Institut Blaise Pascal
   8  * Universite Pierre et Marie Curie (Paris VI)
   9  *
  10  *  from
  11  *
  12  *  linux/fs/minix/inode.c
  13  *
  14  *  Copyright (C) 1991, 1992  Linus Torvalds
  15  *
  16  *  Big-endian to little-endian byte-swapping/bitmaps by
  17  *        David S. Miller (davem@caip.rutgers.edu), 1995
  18  */
  19
  20 #include <linux/module.h>
  21 #include <linux/string.h>
  22 #include <linux/fs.h>
  23 #include <linux/time.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/backing-dev.h>
  29 #include <linux/parser.h>
  30 #include <linux/buffer_head.h>
  31 #include <linux/exportfs.h>
  32 #include <linux/vfs.h>
  33 #include <linux/random.h>
  34 #include <linux/mount.h>
  35 #include <linux/namei.h>
  36 #include <linux/quotaops.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/dax.h>
  42 #include <linux/uaccess.h>
  43 #include <linux/iversion.h>
  44 #include <linux/unicode.h>
  45 #include <linux/part_stat.h>
  46 #include <linux/kthread.h>
  47 #include <linux/freezer.h>
  48 #include <linux/fsnotify.h>
  49 #include <linux/fs_context.h>
  50 #include <linux/fs_parser.h>
  51
  52 #include "ext4.h"
  53 #include "ext4_extents.h"       /* Needed for trace points definition */
  54 #include "ext4_jbd2.h"
  55 #include "xattr.h"
  56 #include "acl.h"
  57 #include "mballoc.h"
  58 #include "fsmap.h"
  59
  60 #define CREATE_TRACE_POINTS
  61 #include <trace/events/ext4.h>
  62
  63 static struct ext4_lazy_init *ext4_li_info;
  64 static DEFINE_MUTEX(ext4_li_mtx);
  65 static struct ratelimit_state ext4_mount_msg_ratelimit;
  66
  67 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  68                              unsigned long journal_devnum);
  69 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  70 static void ext4_update_super(struct super_block *sb);
  71 static int ext4_commit_super(struct super_block *sb);
  72 static int ext4_mark_recovery_complete(struct super_block *sb,
  73                                         struct ext4_super_block *es);
  74 static int ext4_clear_journal_err(struct super_block *sb,
  75                                   struct ext4_super_block *es);
  76 static int ext4_sync_fs(struct super_block *sb, int wait);
  77 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  78 static int ext4_unfreeze(struct super_block *sb);
  79 static int ext4_freeze(struct super_block *sb);
  80 static inline int ext2_feature_set_ok(struct super_block *sb);
  81 static inline int ext3_feature_set_ok(struct super_block *sb);
  82 static void ext4_destroy_lazyinit_thread(void);
  83 static void ext4_unregister_li_request(struct super_block *sb);
  84 static void ext4_clear_request_list(void);
  85 static struct inode *ext4_get_journal_inode(struct super_block *sb,
  86                                             unsigned int journal_inum);
  87 static int ext4_validate_options(struct fs_context *fc);
  88 static int ext4_check_opt_consistency(struct fs_context *fc,
  89                                       struct super_block *sb);
  90 static void ext4_apply_options(struct fs_context *fc, struct super_block *sb);
  91 static int ext4_parse_param(struct fs_context *fc, struct fs_parameter *param);
  92 static int ext4_get_tree(struct fs_context *fc);
  93 static int ext4_reconfigure(struct fs_context *fc);
  94 static void ext4_fc_free(struct fs_context *fc);
  95 static int ext4_init_fs_context(struct fs_context *fc);
  96 static const struct fs_parameter_spec ext4_param_specs[];
  97
  98 /*
  99  * Lock ordering
 100  *
 101  * page fault path:
 102  * mmap_lock -> sb_start_pagefault -> invalidate_lock (r) -> transaction start
 103  *   -> page lock -> i_data_sem (rw)
 104  *
 105  * buffered write path:
 106  * sb_start_write -> i_mutex -> mmap_lock
 107  * sb_start_write -> i_mutex -> transaction start -> page lock ->
 108  *   i_data_sem (rw)
 109  *
 110  * truncate:
 111  * sb_start_write -> i_mutex -> invalidate_lock (w) -> i_mmap_rwsem (w) ->
 112  *   page lock
 113  * sb_start_write -> i_mutex -> invalidate_lock (w) -> transaction start ->
 114  *   i_data_sem (rw)
 115  *
 116  * direct IO:
 117  * sb_start_write -> i_mutex -> mmap_lock
 118  * sb_start_write -> i_mutex -> transaction start -> i_data_sem (rw)
 119  *
 120  * writepages:
 121  * transaction start -> page lock(s) -> i_data_sem (rw)
 122  */
 123
 124 static const struct fs_context_operations ext4_context_ops = {
 125         .parse_param    = ext4_parse_param,
 126         .get_tree       = ext4_get_tree,
 127         .reconfigure    = ext4_reconfigure,
 128         .free           = ext4_fc_free,
 129 };
 130
 131
 132 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
 133 static struct file_system_type ext2_fs_type = {
 134         .owner                  = THIS_MODULE,
 135         .name                   = "ext2",
 136         .init_fs_context        = ext4_init_fs_context,
 137         .parameters             = ext4_param_specs,
 138         .kill_sb                = kill_block_super,
 139         .fs_flags               = FS_REQUIRES_DEV,
 140 };
 141 MODULE_ALIAS_FS("ext2");
 142 MODULE_ALIAS("ext2");
 143 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
 144 #else
 145 #define IS_EXT2_SB(sb) (0)
 146 #endif
 147
 148
 149 static struct file_system_type ext3_fs_type = {
 150         .owner                  = THIS_MODULE,
 151         .name                   = "ext3",
 152         .init_fs_context        = ext4_init_fs_context,
 153         .parameters             = ext4_param_specs,
 154         .kill_sb                = kill_block_super,
 155         .fs_flags               = FS_REQUIRES_DEV,
 156 };
 157 MODULE_ALIAS_FS("ext3");
 158 MODULE_ALIAS("ext3");
 159 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 160
 161
 162 static inline void __ext4_read_bh(struct buffer_head *bh, blk_opf_t op_flags,
 163                                   bh_end_io_t *end_io)
 164 {
 165         /*
 166          * buffer's verified bit is no longer valid after reading from
 167          * disk again due to write out error, clear it to make sure we
 168          * recheck the buffer contents.
 169          */
 170         clear_buffer_verified(bh);
 171
 172         bh->b_end_io = end_io ? end_io : end_buffer_read_sync;
 173         get_bh(bh);
 174         submit_bh(REQ_OP_READ | op_flags, bh);
 175 }
 176
 177 void ext4_read_bh_nowait(struct buffer_head *bh, blk_opf_t op_flags,
 178                          bh_end_io_t *end_io)
 179 {
 180         BUG_ON(!buffer_locked(bh));
 181
 182         if (ext4_buffer_uptodate(bh)) {
 183                 unlock_buffer(bh);
 184                 return;
 185         }
 186         __ext4_read_bh(bh, op_flags, end_io);
 187 }
 188
 189 int ext4_read_bh(struct buffer_head *bh, blk_opf_t op_flags, bh_end_io_t *end_io)
 190 {
 191         BUG_ON(!buffer_locked(bh));
 192
 193         if (ext4_buffer_uptodate(bh)) {
 194                 unlock_buffer(bh);
 195                 return 0;
 196         }
 197
 198         __ext4_read_bh(bh, op_flags, end_io);
 199
 200         wait_on_buffer(bh);
 201         if (buffer_uptodate(bh))
 202                 return 0;
 203         return -EIO;
 204 }
 205
 206 int ext4_read_bh_lock(struct buffer_head *bh, blk_opf_t op_flags, bool wait)
 207 {
 208         lock_buffer(bh);
 209         if (!wait) {
 210                 ext4_read_bh_nowait(bh, op_flags, NULL);
 211                 return 0;
 212         }
 213         return ext4_read_bh(bh, op_flags, NULL);
 214 }
 215
 216 /*
 217  * This works like __bread_gfp() except it uses ERR_PTR for error
 218  * returns.  Currently with sb_bread it's impossible to distinguish
 219  * between ENOMEM and EIO situations (since both result in a NULL
 220  * return.
 221  */
 222 static struct buffer_head *__ext4_sb_bread_gfp(struct super_block *sb,
 223                                                sector_t block,
 224                                                blk_opf_t op_flags, gfp_t gfp)
 225 {
 226         struct buffer_head *bh;
 227         int ret;
 228
 229         bh = sb_getblk_gfp(sb, block, gfp);
 230         if (bh == NULL)
 231                 return ERR_PTR(-ENOMEM);
 232         if (ext4_buffer_uptodate(bh))
 233                 return bh;
 234
 235         ret = ext4_read_bh_lock(bh, REQ_META | op_flags, true);
 236         if (ret) {
 237                 put_bh(bh);
 238                 return ERR_PTR(ret);
 239         }
 240         return bh;
 241 }
 242
 243 struct buffer_head *ext4_sb_bread(struct super_block *sb, sector_t block,
 244                                    blk_opf_t op_flags)
 245 {
 246         return __ext4_sb_bread_gfp(sb, block, op_flags, __GFP_MOVABLE);
 247 }
 248
 249 struct buffer_head *ext4_sb_bread_unmovable(struct super_block *sb,
 250                                             sector_t block)
 251 {
 252         return __ext4_sb_bread_gfp(sb, block, 0, 0);
 253 }
 254
 255 void ext4_sb_breadahead_unmovable(struct super_block *sb, sector_t block)
 256 {
 257         struct buffer_head *bh = sb_getblk_gfp(sb, block, 0);
 258
 259         if (likely(bh)) {
 260                 if (trylock_buffer(bh))
 261                         ext4_read_bh_nowait(bh, REQ_RAHEAD, NULL);
 262                 brelse(bh);
 263         }
 264 }
 265
 266 static int ext4_verify_csum_type(struct super_block *sb,
 267                                  struct ext4_super_block *es)
 268 {
 269         if (!ext4_has_feature_metadata_csum(sb))
 270                 return 1;
 271
 272         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 273 }
 274
 275 __le32 ext4_superblock_csum(struct super_block *sb,
 276                             struct ext4_super_block *es)
 277 {
 278         struct ext4_sb_info *sbi = EXT4_SB(sb);
 279         int offset = offsetof(struct ext4_super_block, s_checksum);
 280         __u32 csum;
 281
 282         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 283
 284         return cpu_to_le32(csum);
 285 }
 286
 287 static int ext4_superblock_csum_verify(struct super_block *sb,
 288                                        struct ext4_super_block *es)
 289 {
 290         if (!ext4_has_metadata_csum(sb))
 291                 return 1;
 292
 293         return es->s_checksum == ext4_superblock_csum(sb, es);
 294 }
 295
 296 void ext4_superblock_csum_set(struct super_block *sb)
 297 {
 298         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 299
 300         if (!ext4_has_metadata_csum(sb))
 301                 return;
 302
 303         es->s_checksum = ext4_superblock_csum(sb, es);
 304 }
 305
 306 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 307                                struct ext4_group_desc *bg)
 308 {
 309         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 310                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 311                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 312 }
 313
 314 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 315                                struct ext4_group_desc *bg)
 316 {
 317         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 318                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 319                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 320 }
 321
 322 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 323                               struct ext4_group_desc *bg)
 324 {
 325         return le32_to_cpu(bg->bg_inode_table_lo) |
 326                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 327                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 328 }
 329
 330 __u32 ext4_free_group_clusters(struct super_block *sb,
 331                                struct ext4_group_desc *bg)
 332 {
 333         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 334                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 335                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 336 }
 337
 338 __u32 ext4_free_inodes_count(struct super_block *sb,
 339                               struct ext4_group_desc *bg)
 340 {
 341         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 342                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 343                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 344 }
 345
 346 __u32 ext4_used_dirs_count(struct super_block *sb,
 347                               struct ext4_group_desc *bg)
 348 {
 349         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 350                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 351                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 352 }
 353
 354 __u32 ext4_itable_unused_count(struct super_block *sb,
 355                               struct ext4_group_desc *bg)
 356 {
 357         return le16_to_cpu(bg->bg_itable_unused_lo) |
 358                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 359                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 360 }
 361
 362 void ext4_block_bitmap_set(struct super_block *sb,
 363                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 364 {
 365         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 366         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 367                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 368 }
 369
 370 void ext4_inode_bitmap_set(struct super_block *sb,
 371                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 372 {
 373         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 374         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 375                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 376 }
 377
 378 void ext4_inode_table_set(struct super_block *sb,
 379                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 380 {
 381         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 382         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 383                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 384 }
 385
 386 void ext4_free_group_clusters_set(struct super_block *sb,
 387                                   struct ext4_group_desc *bg, __u32 count)
 388 {
 389         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 390         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 391                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 392 }
 393
 394 void ext4_free_inodes_set(struct super_block *sb,
 395                           struct ext4_group_desc *bg, __u32 count)
 396 {
 397         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 398         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 399                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 400 }
 401
 402 void ext4_used_dirs_set(struct super_block *sb,
 403                           struct ext4_group_desc *bg, __u32 count)
 404 {
 405         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 406         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 407                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 408 }
 409
 410 void ext4_itable_unused_set(struct super_block *sb,
 411                           struct ext4_group_desc *bg, __u32 count)
 412 {
 413         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 414         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 415                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 416 }
 417
 418 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi, time64_t now)
 419 {
 420         now = clamp_val(now, 0, (1ull << 40) - 1);
 421
 422         *lo = cpu_to_le32(lower_32_bits(now));
 423         *hi = upper_32_bits(now);
 424 }
 425
 426 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi)
 427 {
 428         return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo);
 429 }
 430 #define ext4_update_tstamp(es, tstamp) \
 431         __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi, \
 432                              ktime_get_real_seconds())
 433 #define ext4_get_tstamp(es, tstamp) \
 434         __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
 435
 436 /*
 437  * The del_gendisk() function uninitializes the disk-specific data
 438  * structures, including the bdi structure, without telling anyone
 439  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 440  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 441  * This is a kludge to prevent these oops until we can put in a proper
 442  * hook in del_gendisk() to inform the VFS and file system layers.
 443  */
 444 static int block_device_ejected(struct super_block *sb)
 445 {
 446         struct inode *bd_inode = sb->s_bdev->bd_inode;
 447         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 448
 449         return bdi->dev == NULL;
 450 }
 451
 452 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 453 {
 454         struct super_block              *sb = journal->j_private;
 455         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 456         int                             error = is_journal_aborted(journal);
 457         struct ext4_journal_cb_entry    *jce;
 458
 459         BUG_ON(txn->t_state == T_FINISHED);
 460
 461         ext4_process_freed_data(sb, txn->t_tid);
 462
 463         spin_lock(&sbi->s_md_lock);
 464         while (!list_empty(&txn->t_private_list)) {
 465                 jce = list_entry(txn->t_private_list.next,
 466                                  struct ext4_journal_cb_entry, jce_list);
 467                 list_del_init(&jce->jce_list);
 468                 spin_unlock(&sbi->s_md_lock);
 469                 jce->jce_func(sb, jce, error);
 470                 spin_lock(&sbi->s_md_lock);
 471         }
 472         spin_unlock(&sbi->s_md_lock);
 473 }
 474
 475 /*
 476  * This writepage callback for write_cache_pages()
 477  * takes care of a few cases after page cleaning.
 478  *
 479  * write_cache_pages() already checks for dirty pages
 480  * and calls clear_page_dirty_for_io(), which we want,
 481  * to write protect the pages.
 482  *
 483  * However, we may have to redirty a page (see below.)
 484  */
 485 static int ext4_journalled_writepage_callback(struct folio *folio,
 486                                               struct writeback_control *wbc,
 487                                               void *data)
 488 {
 489         transaction_t *transaction = (transaction_t *) data;
 490         struct buffer_head *bh, *head;
 491         struct journal_head *jh;
 492
 493         bh = head = folio_buffers(folio);
 494         do {
 495                 /*
 496                  * We have to redirty a page in these cases:
 497                  * 1) If buffer is dirty, it means the page was dirty because it
 498                  * contains a buffer that needs checkpointing. So the dirty bit
 499                  * needs to be preserved so that checkpointing writes the buffer
 500                  * properly.
 501                  * 2) If buffer is not part of the committing transaction
 502                  * (we may have just accidentally come across this buffer because
 503                  * inode range tracking is not exact) or if the currently running
 504                  * transaction already contains this buffer as well, dirty bit
 505                  * needs to be preserved so that the buffer gets writeprotected
 506                  * properly on running transaction's commit.
 507                  */
 508                 jh = bh2jh(bh);
 509                 if (buffer_dirty(bh) ||
 510                     (jh && (jh->b_transaction != transaction ||
 511                             jh->b_next_transaction))) {
 512                         folio_redirty_for_writepage(wbc, folio);
 513                         goto out;
 514                 }
 515         } while ((bh = bh->b_this_page) != head);
 516
 517 out:
 518         return AOP_WRITEPAGE_ACTIVATE;
 519 }
 520
 521 static int ext4_journalled_submit_inode_data_buffers(struct jbd2_inode *jinode)
 522 {
 523         struct address_space *mapping = jinode->i_vfs_inode->i_mapping;
 524         struct writeback_control wbc = {
 525                 .sync_mode =  WB_SYNC_ALL,
 526                 .nr_to_write = LONG_MAX,
 527                 .range_start = jinode->i_dirty_start,
 528                 .range_end = jinode->i_dirty_end,
 529         };
 530
 531         return write_cache_pages(mapping, &wbc,
 532                                  ext4_journalled_writepage_callback,
 533                                  jinode->i_transaction);
 534 }
 535
 536 static int ext4_journal_submit_inode_data_buffers(struct jbd2_inode *jinode)
 537 {
 538         int ret;
 539
 540         if (ext4_should_journal_data(jinode->i_vfs_inode))
 541                 ret = ext4_journalled_submit_inode_data_buffers(jinode);
 542         else
 543                 ret = ext4_normal_submit_inode_data_buffers(jinode);
 544         return ret;
 545 }
 546
 547 static int ext4_journal_finish_inode_data_buffers(struct jbd2_inode *jinode)
 548 {
 549         int ret = 0;
 550
 551         if (!ext4_should_journal_data(jinode->i_vfs_inode))
 552                 ret = jbd2_journal_finish_inode_data_buffers(jinode);
 553
 554         return ret;
 555 }
 556
 557 static bool system_going_down(void)
 558 {
 559         return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF
 560                 || system_state == SYSTEM_RESTART;
 561 }
 562
 563 struct ext4_err_translation {
 564         int code;
 565         int errno;
 566 };
 567
 568 #define EXT4_ERR_TRANSLATE(err) { .code = EXT4_ERR_##err, .errno = err }
 569
 570 static struct ext4_err_translation err_translation[] = {
 571         EXT4_ERR_TRANSLATE(EIO),
 572         EXT4_ERR_TRANSLATE(ENOMEM),
 573         EXT4_ERR_TRANSLATE(EFSBADCRC),
 574         EXT4_ERR_TRANSLATE(EFSCORRUPTED),
 575         EXT4_ERR_TRANSLATE(ENOSPC),
 576         EXT4_ERR_TRANSLATE(ENOKEY),
 577         EXT4_ERR_TRANSLATE(EROFS),
 578         EXT4_ERR_TRANSLATE(EFBIG),
 579         EXT4_ERR_TRANSLATE(EEXIST),
 580         EXT4_ERR_TRANSLATE(ERANGE),
 581         EXT4_ERR_TRANSLATE(EOVERFLOW),
 582         EXT4_ERR_TRANSLATE(EBUSY),
 583         EXT4_ERR_TRANSLATE(ENOTDIR),
 584         EXT4_ERR_TRANSLATE(ENOTEMPTY),
 585         EXT4_ERR_TRANSLATE(ESHUTDOWN),
 586         EXT4_ERR_TRANSLATE(EFAULT),
 587 };
 588
 589 static int ext4_errno_to_code(int errno)
 590 {
 591         int i;
 592
 593         for (i = 0; i < ARRAY_SIZE(err_translation); i++)
 594                 if (err_translation[i].errno == errno)
 595                         return err_translation[i].code;
 596         return EXT4_ERR_UNKNOWN;
 597 }
 598
 599 static void save_error_info(struct super_block *sb, int error,
 600                             __u32 ino, __u64 block,
 601                             const char *func, unsigned int line)
 602 {
 603         struct ext4_sb_info *sbi = EXT4_SB(sb);
 604
 605         /* We default to EFSCORRUPTED error... */
 606         if (error == 0)
 607                 error = EFSCORRUPTED;
 608
 609         spin_lock(&sbi->s_error_lock);
 610         sbi->s_add_error_count++;
 611         sbi->s_last_error_code = error;
 612         sbi->s_last_error_line = line;
 613         sbi->s_last_error_ino = ino;
 614         sbi->s_last_error_block = block;
 615         sbi->s_last_error_func = func;
 616         sbi->s_last_error_time = ktime_get_real_seconds();
 617         if (!sbi->s_first_error_time) {
 618                 sbi->s_first_error_code = error;
 619                 sbi->s_first_error_line = line;
 620                 sbi->s_first_error_ino = ino;
 621                 sbi->s_first_error_block = block;
 622                 sbi->s_first_error_func = func;
 623                 sbi->s_first_error_time = sbi->s_last_error_time;
 624         }
 625         spin_unlock(&sbi->s_error_lock);
 626 }
 627
 628 /* Deal with the reporting of failure conditions on a filesystem such as
 629  * inconsistencies detected or read IO failures.
 630  *
 631  * On ext2, we can store the error state of the filesystem in the
 632  * superblock.  That is not possible on ext4, because we may have other
 633  * write ordering constraints on the superblock which prevent us from
 634  * writing it out straight away; and given that the journal is about to
 635  * be aborted, we can't rely on the current, or future, transactions to
 636  * write out the superblock safely.
 637  *
 638  * We'll just use the jbd2_journal_abort() error code to record an error in
 639  * the journal instead.  On recovery, the journal will complain about
 640  * that error until we've noted it down and cleared it.
 641  *
 642  * If force_ro is set, we unconditionally force the filesystem into an
 643  * ABORT|READONLY state, unless the error response on the fs has been set to
 644  * panic in which case we take the easy way out and panic immediately. This is
 645  * used to deal with unrecoverable failures such as journal IO errors or ENOMEM
 646  * at a critical moment in log management.
 647  */
 648 static void ext4_handle_error(struct super_block *sb, bool force_ro, int error,
 649                               __u32 ino, __u64 block,
 650                               const char *func, unsigned int line)
 651 {
 652         journal_t *journal = EXT4_SB(sb)->s_journal;
 653         bool continue_fs = !force_ro && test_opt(sb, ERRORS_CONT);
 654
 655         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 656         if (test_opt(sb, WARN_ON_ERROR))
 657                 WARN_ON_ONCE(1);
 658
 659         if (!continue_fs && !sb_rdonly(sb)) {
 660                 ext4_set_mount_flag(sb, EXT4_MF_FS_ABORTED);
 661                 if (journal)
 662                         jbd2_journal_abort(journal, -EIO);
 663         }
 664
 665         if (!bdev_read_only(sb->s_bdev)) {
 666                 save_error_info(sb, error, ino, block, func, line);
 667                 /*
 668                  * In case the fs should keep running, we need to writeout
 669                  * superblock through the journal. Due to lock ordering
 670                  * constraints, it may not be safe to do it right here so we
 671                  * defer superblock flushing to a workqueue.
 672                  */
 673                 if (continue_fs && journal)
 674                         schedule_work(&EXT4_SB(sb)->s_error_work);
 675                 else
 676                         ext4_commit_super(sb);
 677         }
 678
 679         /*
 680          * We force ERRORS_RO behavior when system is rebooting. Otherwise we
 681          * could panic during 'reboot -f' as the underlying device got already
 682          * disabled.
 683          */
 684         if (test_opt(sb, ERRORS_PANIC) && !system_going_down()) {
 685                 panic("EXT4-fs (device %s): panic forced after error\n",
 686                         sb->s_id);
 687         }
 688
 689         if (sb_rdonly(sb) || continue_fs)
 690                 return;
 691
 692         ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 693         /*
 694          * Make sure updated value of ->s_mount_flags will be visible before
 695          * ->s_flags update
 696          */
 697         smp_wmb();
 698         sb->s_flags |= SB_RDONLY;
 699 }
 700
 701 static void flush_stashed_error_work(struct work_struct *work)
 702 {
 703         struct ext4_sb_info *sbi = container_of(work, struct ext4_sb_info,
 704                                                 s_error_work);
 705         journal_t *journal = sbi->s_journal;
 706         handle_t *handle;
 707
 708         /*
 709          * If the journal is still running, we have to write out superblock
 710          * through the journal to avoid collisions of other journalled sb
 711          * updates.
 712          *
 713          * We use directly jbd2 functions here to avoid recursing back into
 714          * ext4 error handling code during handling of previous errors.
 715          */
 716         if (!sb_rdonly(sbi->s_sb) && journal) {
 717                 struct buffer_head *sbh = sbi->s_sbh;
 718                 handle = jbd2_journal_start(journal, 1);
 719                 if (IS_ERR(handle))
 720                         goto write_directly;
 721                 if (jbd2_journal_get_write_access(handle, sbh)) {
 722                         jbd2_journal_stop(handle);
 723                         goto write_directly;
 724                 }
 725                 ext4_update_super(sbi->s_sb);
 726                 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
 727                         ext4_msg(sbi->s_sb, KERN_ERR, "previous I/O error to "
 728                                  "superblock detected");
 729                         clear_buffer_write_io_error(sbh);
 730                         set_buffer_uptodate(sbh);
 731                 }
 732
 733                 if (jbd2_journal_dirty_metadata(handle, sbh)) {
 734                         jbd2_journal_stop(handle);
 735                         goto write_directly;
 736                 }
 737                 jbd2_journal_stop(handle);
 738                 ext4_notify_error_sysfs(sbi);
 739                 return;
 740         }
 741 write_directly:
 742         /*
 743          * Write through journal failed. Write sb directly to get error info
 744          * out and hope for the best.
 745          */
 746         ext4_commit_super(sbi->s_sb);
 747         ext4_notify_error_sysfs(sbi);
 748 }
 749
 750 #define ext4_error_ratelimit(sb)                                        \
 751                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 752                              "EXT4-fs error")
 753
 754 void __ext4_error(struct super_block *sb, const char *function,
 755                   unsigned int line, bool force_ro, int error, __u64 block,
 756                   const char *fmt, ...)
 757 {
 758         struct va_format vaf;
 759         va_list args;
 760
 761         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 762                 return;
 763
 764         trace_ext4_error(sb, function, line);
 765         if (ext4_error_ratelimit(sb)) {
 766                 va_start(args, fmt);
 767                 vaf.fmt = fmt;
 768                 vaf.va = &args;
 769                 printk(KERN_CRIT
 770                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 771                        sb->s_id, function, line, current->comm, &vaf);
 772                 va_end(args);
 773         }
 774         fsnotify_sb_error(sb, NULL, error ? error : EFSCORRUPTED);
 775
 776         ext4_handle_error(sb, force_ro, error, 0, block, function, line);
 777 }
 778
 779 void __ext4_error_inode(struct inode *inode, const char *function,
 780                         unsigned int line, ext4_fsblk_t block, int error,
 781                         const char *fmt, ...)
 782 {
 783         va_list args;
 784         struct va_format vaf;
 785
 786         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 787                 return;
 788
 789         trace_ext4_error(inode->i_sb, function, line);
 790         if (ext4_error_ratelimit(inode->i_sb)) {
 791                 va_start(args, fmt);
 792                 vaf.fmt = fmt;
 793                 vaf.va = &args;
 794                 if (block)
 795                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 796                                "inode #%lu: block %llu: comm %s: %pV\n",
 797                                inode->i_sb->s_id, function, line, inode->i_ino,
 798                                block, current->comm, &vaf);
 799                 else
 800                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 801                                "inode #%lu: comm %s: %pV\n",
 802                                inode->i_sb->s_id, function, line, inode->i_ino,
 803                                current->comm, &vaf);
 804                 va_end(args);
 805         }
 806         fsnotify_sb_error(inode->i_sb, inode, error ? error : EFSCORRUPTED);
 807
 808         ext4_handle_error(inode->i_sb, false, error, inode->i_ino, block,
 809                           function, line);
 810 }
 811
 812 void __ext4_error_file(struct file *file, const char *function,
 813                        unsigned int line, ext4_fsblk_t block,
 814                        const char *fmt, ...)
 815 {
 816         va_list args;
 817         struct va_format vaf;
 818         struct inode *inode = file_inode(file);
 819         char pathname[80], *path;
 820
 821         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 822                 return;
 823
 824         trace_ext4_error(inode->i_sb, function, line);
 825         if (ext4_error_ratelimit(inode->i_sb)) {
 826                 path = file_path(file, pathname, sizeof(pathname));
 827                 if (IS_ERR(path))
 828                         path = "(unknown)";
 829                 va_start(args, fmt);
 830                 vaf.fmt = fmt;
 831                 vaf.va = &args;
 832                 if (block)
 833                         printk(KERN_CRIT
 834                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 835                                "block %llu: comm %s: path %s: %pV\n",
 836                                inode->i_sb->s_id, function, line, inode->i_ino,
 837                                block, current->comm, path, &vaf);
 838                 else
 839                         printk(KERN_CRIT
 840                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 841                                "comm %s: path %s: %pV\n",
 842                                inode->i_sb->s_id, function, line, inode->i_ino,
 843                                current->comm, path, &vaf);
 844                 va_end(args);
 845         }
 846         fsnotify_sb_error(inode->i_sb, inode, EFSCORRUPTED);
 847
 848         ext4_handle_error(inode->i_sb, false, EFSCORRUPTED, inode->i_ino, block,
 849                           function, line);
 850 }
 851
 852 const char *ext4_decode_error(struct super_block *sb, int errno,
 853                               char nbuf[16])
 854 {
 855         char *errstr = NULL;
 856
 857         switch (errno) {
 858         case -EFSCORRUPTED:
 859                 errstr = "Corrupt filesystem";
 860                 break;
 861         case -EFSBADCRC:
 862                 errstr = "Filesystem failed CRC";
 863                 break;
 864         case -EIO:
 865                 errstr = "IO failure";
 866                 break;
 867         case -ENOMEM:
 868                 errstr = "Out of memory";
 869                 break;
 870         case -EROFS:
 871                 if (!sb || (EXT4_SB(sb)->s_journal &&
 872                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 873                         errstr = "Journal has aborted";
 874                 else
 875                         errstr = "Readonly filesystem";
 876                 break;
 877         default:
 878                 /* If the caller passed in an extra buffer for unknown
 879                  * errors, textualise them now.  Else we just return
 880                  * NULL. */
 881                 if (nbuf) {
 882                         /* Check for truncated error codes... */
 883                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 884                                 errstr = nbuf;
 885                 }
 886                 break;
 887         }
 888
 889         return errstr;
 890 }
 891
 892 /* __ext4_std_error decodes expected errors from journaling functions
 893  * automatically and invokes the appropriate error response.  */
 894
 895 void __ext4_std_error(struct super_block *sb, const char *function,
 896                       unsigned int line, int errno)
 897 {
 898         char nbuf[16];
 899         const char *errstr;
 900
 901         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 902                 return;
 903
 904         /* Special case: if the error is EROFS, and we're not already
 905          * inside a transaction, then there's really no point in logging
 906          * an error. */
 907         if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb))
 908                 return;
 909
 910         if (ext4_error_ratelimit(sb)) {
 911                 errstr = ext4_decode_error(sb, errno, nbuf);
 912                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 913                        sb->s_id, function, line, errstr);
 914         }
 915         fsnotify_sb_error(sb, NULL, errno ? errno : EFSCORRUPTED);
 916
 917         ext4_handle_error(sb, false, -errno, 0, 0, function, line);
 918 }
 919
 920 void __ext4_msg(struct super_block *sb,
 921                 const char *prefix, const char *fmt, ...)
 922 {
 923         struct va_format vaf;
 924         va_list args;
 925
 926         if (sb) {
 927                 atomic_inc(&EXT4_SB(sb)->s_msg_count);
 928                 if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state),
 929                                   "EXT4-fs"))
 930                         return;
 931         }
 932
 933         va_start(args, fmt);
 934         vaf.fmt = fmt;
 935         vaf.va = &args;
 936         if (sb)
 937                 printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 938         else
 939                 printk("%sEXT4-fs: %pV\n", prefix, &vaf);
 940         va_end(args);
 941 }
 942
 943 static int ext4_warning_ratelimit(struct super_block *sb)
 944 {
 945         atomic_inc(&EXT4_SB(sb)->s_warning_count);
 946         return ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state),
 947                             "EXT4-fs warning");
 948 }
 949
 950 void __ext4_warning(struct super_block *sb, const char *function,
 951                     unsigned int line, const char *fmt, ...)
 952 {
 953         struct va_format vaf;
 954         va_list args;
 955
 956         if (!ext4_warning_ratelimit(sb))
 957                 return;
 958
 959         va_start(args, fmt);
 960         vaf.fmt = fmt;
 961         vaf.va = &args;
 962         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 963                sb->s_id, function, line, &vaf);
 964         va_end(args);
 965 }
 966
 967 void __ext4_warning_inode(const struct inode *inode, const char *function,
 968                           unsigned int line, const char *fmt, ...)
 969 {
 970         struct va_format vaf;
 971         va_list args;
 972
 973         if (!ext4_warning_ratelimit(inode->i_sb))
 974                 return;
 975
 976         va_start(args, fmt);
 977         vaf.fmt = fmt;
 978         vaf.va = &args;
 979         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 980                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 981                function, line, inode->i_ino, current->comm, &vaf);
 982         va_end(args);
 983 }
 984
 985 void __ext4_grp_locked_error(const char *function, unsigned int line,
 986                              struct super_block *sb, ext4_group_t grp,
 987                              unsigned long ino, ext4_fsblk_t block,
 988                              const char *fmt, ...)
 989 __releases(bitlock)
 990 __acquires(bitlock)
 991 {
 992         struct va_format vaf;
 993         va_list args;
 994
 995         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 996                 return;
 997
 998         trace_ext4_error(sb, function, line);
 999         if (ext4_error_ratelimit(sb)) {
1000                 va_start(args, fmt);
1001                 vaf.fmt = fmt;
1002                 vaf.va = &args;
1003                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
1004                        sb->s_id, function, line, grp);
1005                 if (ino)
1006                         printk(KERN_CONT "inode %lu: ", ino);
1007                 if (block)
1008                         printk(KERN_CONT "block %llu:",
1009                                (unsigned long long) block);
1010                 printk(KERN_CONT "%pV\n", &vaf);
1011                 va_end(args);
1012         }
1013
1014         if (test_opt(sb, ERRORS_CONT)) {
1015                 if (test_opt(sb, WARN_ON_ERROR))
1016                         WARN_ON_ONCE(1);
1017                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
1018                 if (!bdev_read_only(sb->s_bdev)) {
1019                         save_error_info(sb, EFSCORRUPTED, ino, block, function,
1020                                         line);
1021                         schedule_work(&EXT4_SB(sb)->s_error_work);
1022                 }
1023                 return;
1024         }
1025         ext4_unlock_group(sb, grp);
1026         ext4_handle_error(sb, false, EFSCORRUPTED, ino, block, function, line);
1027         /*
1028          * We only get here in the ERRORS_RO case; relocking the group
1029          * may be dangerous, but nothing bad will happen since the
1030          * filesystem will have already been marked read/only and the
1031          * journal has been aborted.  We return 1 as a hint to callers
1032          * who might what to use the return value from
1033          * ext4_grp_locked_error() to distinguish between the
1034          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
1035          * aggressively from the ext4 function in question, with a
1036          * more appropriate error code.
1037          */
1038         ext4_lock_group(sb, grp);
1039         return;
1040 }
1041
1042 void ext4_mark_group_bitmap_corrupted(struct super_block *sb,
1043                                      ext4_group_t group,
1044                                      unsigned int flags)
1045 {
1046         struct ext4_sb_info *sbi = EXT4_SB(sb);
1047         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1048         struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
1049         int ret;
1050
1051         if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) {
1052                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT,
1053                                             &grp->bb_state);
1054                 if (!ret)
1055                         percpu_counter_sub(&sbi->s_freeclusters_counter,
1056                                            grp->bb_free);
1057         }
1058
1059         if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) {
1060                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT,
1061                                             &grp->bb_state);
1062                 if (!ret && gdp) {
1063                         int count;
1064
1065                         count = ext4_free_inodes_count(sb, gdp);
1066                         percpu_counter_sub(&sbi->s_freeinodes_counter,
1067                                            count);
1068                 }
1069         }
1070 }
1071
1072 void ext4_update_dynamic_rev(struct super_block *sb)
1073 {
1074         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1075
1076         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
1077                 return;
1078
1079         ext4_warning(sb,
1080                      "updating to rev %d because of new feature flag, "
1081                      "running e2fsck is recommended",
1082                      EXT4_DYNAMIC_REV);
1083
1084         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
1085         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
1086         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
1087         /* leave es->s_feature_*compat flags alone */
1088         /* es->s_uuid will be set by e2fsck if empty */
1089
1090         /*
1091          * The rest of the superblock fields should be zero, and if not it
1092          * means they are likely already in use, so leave them alone.  We
1093          * can leave it up to e2fsck to clean up any inconsistencies there.
1094          */
1095 }
1096
1097 /*
1098  * Open the external journal device
1099  */
1100 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
1101 {
1102         struct block_device *bdev;
1103
1104         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
1105         if (IS_ERR(bdev))
1106                 goto fail;
1107         return bdev;
1108
1109 fail:
1110         ext4_msg(sb, KERN_ERR,
1111                  "failed to open journal device unknown-block(%u,%u) %ld",
1112                  MAJOR(dev), MINOR(dev), PTR_ERR(bdev));
1113         return NULL;
1114 }
1115
1116 /*
1117  * Release the journal device
1118  */
1119 static void ext4_blkdev_put(struct block_device *bdev)
1120 {
1121         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1122 }
1123
1124 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
1125 {
1126         struct block_device *bdev;
1127         bdev = sbi->s_journal_bdev;
1128         if (bdev) {
1129                 ext4_blkdev_put(bdev);
1130                 sbi->s_journal_bdev = NULL;
1131         }
1132 }
1133
1134 static inline struct inode *orphan_list_entry(struct list_head *l)
1135 {
1136         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
1137 }
1138
1139 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
1140 {
1141         struct list_head *l;
1142
1143         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
1144                  le32_to_cpu(sbi->s_es->s_last_orphan));
1145
1146         printk(KERN_ERR "sb_info orphan list:\n");
1147         list_for_each(l, &sbi->s_orphan) {
1148                 struct inode *inode = orphan_list_entry(l);
1149                 printk(KERN_ERR "  "
1150                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
1151                        inode->i_sb->s_id, inode->i_ino, inode,
1152                        inode->i_mode, inode->i_nlink,
1153                        NEXT_ORPHAN(inode));
1154         }
1155 }
1156
1157 #ifdef CONFIG_QUOTA
1158 static int ext4_quota_off(struct super_block *sb, int type);
1159
1160 static inline void ext4_quota_off_umount(struct super_block *sb)
1161 {
1162         int type;
1163
1164         /* Use our quota_off function to clear inode flags etc. */
1165         for (type = 0; type < EXT4_MAXQUOTAS; type++)
1166                 ext4_quota_off(sb, type);
1167 }
1168
1169 /*
1170  * This is a helper function which is used in the mount/remount
1171  * codepaths (which holds s_umount) to fetch the quota file name.
1172  */
1173 static inline char *get_qf_name(struct super_block *sb,
1174                                 struct ext4_sb_info *sbi,
1175                                 int type)
1176 {
1177         return rcu_dereference_protected(sbi->s_qf_names[type],
1178                                          lockdep_is_held(&sb->s_umount));
1179 }
1180 #else
1181 static inline void ext4_quota_off_umount(struct super_block *sb)
1182 {
1183 }
1184 #endif
1185
1186 static int ext4_percpu_param_init(struct ext4_sb_info *sbi)
1187 {
1188         ext4_fsblk_t block;
1189         int err;
1190
1191         block = ext4_count_free_clusters(sbi->s_sb);
1192         ext4_free_blocks_count_set(sbi->s_es, EXT4_C2B(sbi, block));
1193         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
1194                                   GFP_KERNEL);
1195         if (!err) {
1196                 unsigned long freei = ext4_count_free_inodes(sbi->s_sb);
1197                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
1198                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
1199                                           GFP_KERNEL);
1200         }
1201         if (!err)
1202                 err = percpu_counter_init(&sbi->s_dirs_counter,
1203                                           ext4_count_dirs(sbi->s_sb), GFP_KERNEL);
1204         if (!err)
1205                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
1206                                           GFP_KERNEL);
1207         if (!err)
1208                 err = percpu_counter_init(&sbi->s_sra_exceeded_retry_limit, 0,
1209                                           GFP_KERNEL);
1210         if (!err)
1211                 err = percpu_init_rwsem(&sbi->s_writepages_rwsem);
1212
1213         if (err)
1214                 ext4_msg(sbi->s_sb, KERN_ERR, "insufficient memory");
1215
1216         return err;
1217 }
1218
1219 static void ext4_percpu_param_destroy(struct ext4_sb_info *sbi)
1220 {
1221         percpu_counter_destroy(&sbi->s_freeclusters_counter);
1222         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1223         percpu_counter_destroy(&sbi->s_dirs_counter);
1224         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
1225         percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit);
1226         percpu_free_rwsem(&sbi->s_writepages_rwsem);
1227 }
1228
1229 static void ext4_group_desc_free(struct ext4_sb_info *sbi)
1230 {
1231         struct buffer_head **group_desc;
1232         int i;
1233
1234         rcu_read_lock();
1235         group_desc = rcu_dereference(sbi->s_group_desc);
1236         for (i = 0; i < sbi->s_gdb_count; i++)
1237                 brelse(group_desc[i]);
1238         kvfree(group_desc);
1239         rcu_read_unlock();
1240 }
1241
1242 static void ext4_flex_groups_free(struct ext4_sb_info *sbi)
1243 {
1244         struct flex_groups **flex_groups;
1245         int i;
1246
1247         rcu_read_lock();
1248         flex_groups = rcu_dereference(sbi->s_flex_groups);
1249         if (flex_groups) {
1250                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
1251                         kvfree(flex_groups[i]);
1252                 kvfree(flex_groups);
1253         }
1254         rcu_read_unlock();
1255 }
1256
1257 static void ext4_put_super(struct super_block *sb)
1258 {
1259         struct ext4_sb_info *sbi = EXT4_SB(sb);
1260         struct ext4_super_block *es = sbi->s_es;
1261         int aborted = 0;
1262         int i, err;
1263
1264         /*
1265          * Unregister sysfs before destroying jbd2 journal.
1266          * Since we could still access attr_journal_task attribute via sysfs
1267          * path which could have sbi->s_journal->j_task as NULL
1268          * Unregister sysfs before flush sbi->s_error_work.
1269          * Since user may read /proc/fs/ext4/xx/mb_groups during umount, If
1270          * read metadata verify failed then will queue error work.
1271          * flush_stashed_error_work will call start_this_handle may trigger
1272          * BUG_ON.
1273          */
1274         ext4_unregister_sysfs(sb);
1275
1276         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs unmount"))
1277                 ext4_msg(sb, KERN_INFO, "unmounting filesystem %pU.",
1278                          &sb->s_uuid);
1279
1280         ext4_unregister_li_request(sb);
1281         ext4_quota_off_umount(sb);
1282
1283         flush_work(&sbi->s_error_work);
1284         destroy_workqueue(sbi->rsv_conversion_wq);
1285         ext4_release_orphan_info(sb);
1286
1287         if (sbi->s_journal) {
1288                 aborted = is_journal_aborted(sbi->s_journal);
1289                 err = jbd2_journal_destroy(sbi->s_journal);
1290                 sbi->s_journal = NULL;
1291                 if ((err < 0) && !aborted) {
1292                         ext4_abort(sb, -err, "Couldn't clean up the journal");
1293                 }
1294         }
1295
1296         ext4_es_unregister_shrinker(sbi);
1297         timer_shutdown_sync(&sbi->s_err_report);
1298         ext4_release_system_zone(sb);
1299         ext4_mb_release(sb);
1300         ext4_ext_release(sb);
1301
1302         if (!sb_rdonly(sb) && !aborted) {
1303                 ext4_clear_feature_journal_needs_recovery(sb);
1304                 ext4_clear_feature_orphan_present(sb);
1305                 es->s_state = cpu_to_le16(sbi->s_mount_state);
1306         }
1307         if (!sb_rdonly(sb))
1308                 ext4_commit_super(sb);
1309
1310         ext4_group_desc_free(sbi);
1311         ext4_flex_groups_free(sbi);
1312         ext4_percpu_param_destroy(sbi);
1313 #ifdef CONFIG_QUOTA
1314         for (i = 0; i < EXT4_MAXQUOTAS; i++)
1315                 kfree(get_qf_name(sb, sbi, i));
1316 #endif
1317
1318         /* Debugging code just in case the in-memory inode orphan list
1319          * isn't empty.  The on-disk one can be non-empty if we've
1320          * detected an error and taken the fs readonly, but the
1321          * in-memory list had better be clean by this point. */
1322         if (!list_empty(&sbi->s_orphan))
1323                 dump_orphan_list(sb, sbi);
1324         ASSERT(list_empty(&sbi->s_orphan));
1325
1326         sync_blockdev(sb->s_bdev);
1327         invalidate_bdev(sb->s_bdev);
1328         if (sbi->s_journal_bdev && sbi->s_journal_bdev != sb->s_bdev) {
1329                 /*
1330                  * Invalidate the journal device's buffers.  We don't want them
1331                  * floating about in memory - the physical journal device may
1332                  * hotswapped, and it breaks the `ro-after' testing code.
1333                  */
1334                 sync_blockdev(sbi->s_journal_bdev);
1335                 invalidate_bdev(sbi->s_journal_bdev);
1336                 ext4_blkdev_remove(sbi);
1337         }
1338
1339         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
1340         sbi->s_ea_inode_cache = NULL;
1341
1342         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
1343         sbi->s_ea_block_cache = NULL;
1344
1345         ext4_stop_mmpd(sbi);
1346
1347         brelse(sbi->s_sbh);
1348         sb->s_fs_info = NULL;
1349         /*
1350          * Now that we are completely done shutting down the
1351          * superblock, we need to actually destroy the kobject.
1352          */
1353         kobject_put(&sbi->s_kobj);
1354         wait_for_completion(&sbi->s_kobj_unregister);
1355         if (sbi->s_chksum_driver)
1356                 crypto_free_shash(sbi->s_chksum_driver);
1357         kfree(sbi->s_blockgroup_lock);
1358         fs_put_dax(sbi->s_daxdev, NULL);
1359         fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy);
1360 #if IS_ENABLED(CONFIG_UNICODE)
1361         utf8_unload(sb->s_encoding);
1362 #endif
1363         kfree(sbi);
1364 }
1365
1366 static struct kmem_cache *ext4_inode_cachep;
1367
1368 /*
1369  * Called inside transaction, so use GFP_NOFS
1370  */
1371 static struct inode *ext4_alloc_inode(struct super_block *sb)
1372 {
1373         struct ext4_inode_info *ei;
1374
1375         ei = alloc_inode_sb(sb, ext4_inode_cachep, GFP_NOFS);
1376         if (!ei)
1377                 return NULL;
1378
1379         inode_set_iversion(&ei->vfs_inode, 1);
1380         ei->i_flags = 0;
1381         spin_lock_init(&ei->i_raw_lock);
1382         ei->i_prealloc_node = RB_ROOT;
1383         atomic_set(&ei->i_prealloc_active, 0);
1384         rwlock_init(&ei->i_prealloc_lock);
1385         ext4_es_init_tree(&ei->i_es_tree);
1386         rwlock_init(&ei->i_es_lock);
1387         INIT_LIST_HEAD(&ei->i_es_list);
1388         ei->i_es_all_nr = 0;
1389         ei->i_es_shk_nr = 0;
1390         ei->i_es_shrink_lblk = 0;
1391         ei->i_reserved_data_blocks = 0;
1392         spin_lock_init(&(ei->i_block_reservation_lock));
1393         ext4_init_pending_tree(&ei->i_pending_tree);
1394 #ifdef CONFIG_QUOTA
1395         ei->i_reserved_quota = 0;
1396         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
1397 #endif
1398         ei->jinode = NULL;
1399         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
1400         spin_lock_init(&ei->i_completed_io_lock);
1401         ei->i_sync_tid = 0;
1402         ei->i_datasync_tid = 0;
1403         atomic_set(&ei->i_unwritten, 0);
1404         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
1405         ext4_fc_init_inode(&ei->vfs_inode);
1406         mutex_init(&ei->i_fc_lock);
1407         return &ei->vfs_inode;
1408 }
1409
1410 static int ext4_drop_inode(struct inode *inode)
1411 {
1412         int drop = generic_drop_inode(inode);
1413
1414         if (!drop)
1415                 drop = fscrypt_drop_inode(inode);
1416
1417         trace_ext4_drop_inode(inode, drop);
1418         return drop;
1419 }
1420
1421 static void ext4_free_in_core_inode(struct inode *inode)
1422 {
1423         fscrypt_free_inode(inode);
1424         if (!list_empty(&(EXT4_I(inode)->i_fc_list))) {
1425                 pr_warn("%s: inode %ld still in fc list",
1426                         __func__, inode->i_ino);
1427         }
1428         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1429 }
1430
1431 static void ext4_destroy_inode(struct inode *inode)
1432 {
1433         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1434                 ext4_msg(inode->i_sb, KERN_ERR,
1435                          "Inode %lu (%p): orphan list check failed!",
1436                          inode->i_ino, EXT4_I(inode));
1437                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1438                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1439                                 true);
1440                 dump_stack();
1441         }
1442
1443         if (EXT4_I(inode)->i_reserved_data_blocks)
1444                 ext4_msg(inode->i_sb, KERN_ERR,
1445                          "Inode %lu (%p): i_reserved_data_blocks (%u) not cleared!",
1446                          inode->i_ino, EXT4_I(inode),
1447                          EXT4_I(inode)->i_reserved_data_blocks);
1448 }
1449
1450 static void init_once(void *foo)
1451 {
1452         struct ext4_inode_info *ei = foo;
1453
1454         INIT_LIST_HEAD(&ei->i_orphan);
1455         init_rwsem(&ei->xattr_sem);
1456         init_rwsem(&ei->i_data_sem);
1457         inode_init_once(&ei->vfs_inode);
1458         ext4_fc_init_inode(&ei->vfs_inode);
1459 }
1460
1461 static int __init init_inodecache(void)
1462 {
1463         ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache",
1464                                 sizeof(struct ext4_inode_info), 0,
1465                                 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
1466                                         SLAB_ACCOUNT),
1467                                 offsetof(struct ext4_inode_info, i_data),
1468                                 sizeof_field(struct ext4_inode_info, i_data),
1469                                 init_once);
1470         if (ext4_inode_cachep == NULL)
1471                 return -ENOMEM;
1472         return 0;
1473 }
1474
1475 static void destroy_inodecache(void)
1476 {
1477         /*
1478          * Make sure all delayed rcu free inodes are flushed before we
1479          * destroy cache.
1480          */
1481         rcu_barrier();
1482         kmem_cache_destroy(ext4_inode_cachep);
1483 }
1484
1485 void ext4_clear_inode(struct inode *inode)
1486 {
1487         ext4_fc_del(inode);
1488         invalidate_inode_buffers(inode);
1489         clear_inode(inode);
1490         ext4_discard_preallocations(inode, 0);
1491         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1492         dquot_drop(inode);
1493         if (EXT4_I(inode)->jinode) {
1494                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1495                                                EXT4_I(inode)->jinode);
1496                 jbd2_free_inode(EXT4_I(inode)->jinode);
1497                 EXT4_I(inode)->jinode = NULL;
1498         }
1499         fscrypt_put_encryption_info(inode);
1500         fsverity_cleanup_inode(inode);
1501 }
1502
1503 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1504                                         u64 ino, u32 generation)
1505 {
1506         struct inode *inode;
1507
1508         /*
1509          * Currently we don't know the generation for parent directory, so
1510          * a generation of 0 means "accept any"
1511          */
1512         inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE);
1513         if (IS_ERR(inode))
1514                 return ERR_CAST(inode);
1515         if (generation && inode->i_generation != generation) {
1516                 iput(inode);
1517                 return ERR_PTR(-ESTALE);
1518         }
1519
1520         return inode;
1521 }
1522
1523 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1524                                         int fh_len, int fh_type)
1525 {
1526         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1527                                     ext4_nfs_get_inode);
1528 }
1529
1530 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1531                                         int fh_len, int fh_type)
1532 {
1533         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1534                                     ext4_nfs_get_inode);
1535 }
1536
1537 static int ext4_nfs_commit_metadata(struct inode *inode)
1538 {
1539         struct writeback_control wbc = {
1540                 .sync_mode = WB_SYNC_ALL
1541         };
1542
1543         trace_ext4_nfs_commit_metadata(inode);
1544         return ext4_write_inode(inode, &wbc);
1545 }
1546
1547 #ifdef CONFIG_QUOTA
1548 static const char * const quotatypes[] = INITQFNAMES;
1549 #define QTYPE2NAME(t) (quotatypes[t])
1550
1551 static int ext4_write_dquot(struct dquot *dquot);
1552 static int ext4_acquire_dquot(struct dquot *dquot);
1553 static int ext4_release_dquot(struct dquot *dquot);
1554 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1555 static int ext4_write_info(struct super_block *sb, int type);
1556 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1557                          const struct path *path);
1558 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1559                                size_t len, loff_t off);
1560 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1561                                 const char *data, size_t len, loff_t off);
1562 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1563                              unsigned int flags);
1564
1565 static struct dquot **ext4_get_dquots(struct inode *inode)
1566 {
1567         return EXT4_I(inode)->i_dquot;
1568 }
1569
1570 static const struct dquot_operations ext4_quota_operations = {
1571         .get_reserved_space     = ext4_get_reserved_space,
1572         .write_dquot            = ext4_write_dquot,
1573         .acquire_dquot          = ext4_acquire_dquot,
1574         .release_dquot          = ext4_release_dquot,
1575         .mark_dirty             = ext4_mark_dquot_dirty,
1576         .write_info             = ext4_write_info,
1577         .alloc_dquot            = dquot_alloc,
1578         .destroy_dquot          = dquot_destroy,
1579         .get_projid             = ext4_get_projid,
1580         .get_inode_usage        = ext4_get_inode_usage,
1581         .get_next_id            = dquot_get_next_id,
1582 };
1583
1584 static const struct quotactl_ops ext4_qctl_operations = {
1585         .quota_on       = ext4_quota_on,
1586         .quota_off      = ext4_quota_off,
1587         .quota_sync     = dquot_quota_sync,
1588         .get_state      = dquot_get_state,
1589         .set_info       = dquot_set_dqinfo,
1590         .get_dqblk      = dquot_get_dqblk,
1591         .set_dqblk      = dquot_set_dqblk,
1592         .get_nextdqblk  = dquot_get_next_dqblk,
1593 };
1594 #endif
1595
1596 static const struct super_operations ext4_sops = {
1597         .alloc_inode    = ext4_alloc_inode,
1598         .free_inode     = ext4_free_in_core_inode,
1599         .destroy_inode  = ext4_destroy_inode,
1600         .write_inode    = ext4_write_inode,
1601         .dirty_inode    = ext4_dirty_inode,
1602         .drop_inode     = ext4_drop_inode,
1603         .evict_inode    = ext4_evict_inode,
1604         .put_super      = ext4_put_super,
1605         .sync_fs        = ext4_sync_fs,
1606         .freeze_fs      = ext4_freeze,
1607         .unfreeze_fs    = ext4_unfreeze,
1608         .statfs         = ext4_statfs,
1609         .show_options   = ext4_show_options,
1610 #ifdef CONFIG_QUOTA
1611         .quota_read     = ext4_quota_read,
1612         .quota_write    = ext4_quota_write,
1613         .get_dquots     = ext4_get_dquots,
1614 #endif
1615 };
1616
1617 static const struct export_operations ext4_export_ops = {
1618         .fh_to_dentry = ext4_fh_to_dentry,
1619         .fh_to_parent = ext4_fh_to_parent,
1620         .get_parent = ext4_get_parent,
1621         .commit_metadata = ext4_nfs_commit_metadata,
1622 };
1623
1624 enum {
1625         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1626         Opt_resgid, Opt_resuid, Opt_sb,
1627         Opt_nouid32, Opt_debug, Opt_removed,
1628         Opt_user_xattr, Opt_acl,
1629         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1630         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1631         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1632         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1633         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1634         Opt_inlinecrypt,
1635         Opt_usrjquota, Opt_grpjquota, Opt_quota,
1636         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1637         Opt_usrquota, Opt_grpquota, Opt_prjquota,
1638         Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never,
1639         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
1640         Opt_nowarn_on_error, Opt_mblk_io_submit, Opt_debug_want_extra_isize,
1641         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1642         Opt_inode_readahead_blks, Opt_journal_ioprio,
1643         Opt_dioread_nolock, Opt_dioread_lock,
1644         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1645         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1646         Opt_no_prefetch_block_bitmaps, Opt_mb_optimize_scan,
1647         Opt_errors, Opt_data, Opt_data_err, Opt_jqfmt, Opt_dax_type,
1648 #ifdef CONFIG_EXT4_DEBUG
1649         Opt_fc_debug_max_replay, Opt_fc_debug_force
1650 #endif
1651 };
1652
1653 static const struct constant_table ext4_param_errors[] = {
1654         {"continue",    EXT4_MOUNT_ERRORS_CONT},
1655         {"panic",       EXT4_MOUNT_ERRORS_PANIC},
1656         {"remount-ro",  EXT4_MOUNT_ERRORS_RO},
1657         {}
1658 };
1659
1660 static const struct constant_table ext4_param_data[] = {
1661         {"journal",     EXT4_MOUNT_JOURNAL_DATA},
1662         {"ordered",     EXT4_MOUNT_ORDERED_DATA},
1663         {"writeback",   EXT4_MOUNT_WRITEBACK_DATA},
1664         {}
1665 };
1666
1667 static const struct constant_table ext4_param_data_err[] = {
1668         {"abort",       Opt_data_err_abort},
1669         {"ignore",      Opt_data_err_ignore},
1670         {}
1671 };
1672
1673 static const struct constant_table ext4_param_jqfmt[] = {
1674         {"vfsold",      QFMT_VFS_OLD},
1675         {"vfsv0",       QFMT_VFS_V0},
1676         {"vfsv1",       QFMT_VFS_V1},
1677         {}
1678 };
1679
1680 static const struct constant_table ext4_param_dax[] = {
1681         {"always",      Opt_dax_always},
1682         {"inode",       Opt_dax_inode},
1683         {"never",       Opt_dax_never},
1684         {}
1685 };
1686
1687 /* String parameter that allows empty argument */
1688 #define fsparam_string_empty(NAME, OPT) \
1689         __fsparam(fs_param_is_string, NAME, OPT, fs_param_can_be_empty, NULL)
1690
1691 /*
1692  * Mount option specification
1693  * We don't use fsparam_flag_no because of the way we set the
1694  * options and the way we show them in _ext4_show_options(). To
1695  * keep the changes to a minimum, let's keep the negative options
1696  * separate for now.
1697  */
1698 static const struct fs_parameter_spec ext4_param_specs[] = {
1699         fsparam_flag    ("bsddf",               Opt_bsd_df),
1700         fsparam_flag    ("minixdf",             Opt_minix_df),
1701         fsparam_flag    ("grpid",               Opt_grpid),
1702         fsparam_flag    ("bsdgroups",           Opt_grpid),
1703         fsparam_flag    ("nogrpid",             Opt_nogrpid),
1704         fsparam_flag    ("sysvgroups",          Opt_nogrpid),
1705         fsparam_u32     ("resgid",              Opt_resgid),
1706         fsparam_u32     ("resuid",              Opt_resuid),
1707         fsparam_u32     ("sb",                  Opt_sb),
1708         fsparam_enum    ("errors",              Opt_errors, ext4_param_errors),
1709         fsparam_flag    ("nouid32",             Opt_nouid32),
1710         fsparam_flag    ("debug",               Opt_debug),
1711         fsparam_flag    ("oldalloc",            Opt_removed),
1712         fsparam_flag    ("orlov",               Opt_removed),
1713         fsparam_flag    ("user_xattr",          Opt_user_xattr),
1714         fsparam_flag    ("acl",                 Opt_acl),
1715         fsparam_flag    ("norecovery",          Opt_noload),
1716         fsparam_flag    ("noload",              Opt_noload),
1717         fsparam_flag    ("bh",                  Opt_removed),
1718         fsparam_flag    ("nobh",                Opt_removed),
1719         fsparam_u32     ("commit",              Opt_commit),
1720         fsparam_u32     ("min_batch_time",      Opt_min_batch_time),
1721         fsparam_u32     ("max_batch_time",      Opt_max_batch_time),
1722         fsparam_u32     ("journal_dev",         Opt_journal_dev),
1723         fsparam_bdev    ("journal_path",        Opt_journal_path),
1724         fsparam_flag    ("journal_checksum",    Opt_journal_checksum),
1725         fsparam_flag    ("nojournal_checksum",  Opt_nojournal_checksum),
1726         fsparam_flag    ("journal_async_commit",Opt_journal_async_commit),
1727         fsparam_flag    ("abort",               Opt_abort),
1728         fsparam_enum    ("data",                Opt_data, ext4_param_data),
1729         fsparam_enum    ("data_err",            Opt_data_err,
1730                                                 ext4_param_data_err),
1731         fsparam_string_empty
1732                         ("usrjquota",           Opt_usrjquota),
1733         fsparam_string_empty
1734                         ("grpjquota",           Opt_grpjquota),
1735         fsparam_enum    ("jqfmt",               Opt_jqfmt, ext4_param_jqfmt),
1736         fsparam_flag    ("grpquota",            Opt_grpquota),
1737         fsparam_flag    ("quota",               Opt_quota),
1738         fsparam_flag    ("noquota",             Opt_noquota),
1739         fsparam_flag    ("usrquota",            Opt_usrquota),
1740         fsparam_flag    ("prjquota",            Opt_prjquota),
1741         fsparam_flag    ("barrier",             Opt_barrier),
1742         fsparam_u32     ("barrier",             Opt_barrier),
1743         fsparam_flag    ("nobarrier",           Opt_nobarrier),
1744         fsparam_flag    ("i_version",           Opt_removed),
1745         fsparam_flag    ("dax",                 Opt_dax),
1746         fsparam_enum    ("dax",                 Opt_dax_type, ext4_param_dax),
1747         fsparam_u32     ("stripe",              Opt_stripe),
1748         fsparam_flag    ("delalloc",            Opt_delalloc),
1749         fsparam_flag    ("nodelalloc",          Opt_nodelalloc),
1750         fsparam_flag    ("warn_on_error",       Opt_warn_on_error),
1751         fsparam_flag    ("nowarn_on_error",     Opt_nowarn_on_error),
1752         fsparam_u32     ("debug_want_extra_isize",
1753                                                 Opt_debug_want_extra_isize),
1754         fsparam_flag    ("mblk_io_submit",      Opt_removed),
1755         fsparam_flag    ("nomblk_io_submit",    Opt_removed),
1756         fsparam_flag    ("block_validity",      Opt_block_validity),
1757         fsparam_flag    ("noblock_validity",    Opt_noblock_validity),
1758         fsparam_u32     ("inode_readahead_blks",
1759                                                 Opt_inode_readahead_blks),
1760         fsparam_u32     ("journal_ioprio",      Opt_journal_ioprio),
1761         fsparam_u32     ("auto_da_alloc",       Opt_auto_da_alloc),
1762         fsparam_flag    ("auto_da_alloc",       Opt_auto_da_alloc),
1763         fsparam_flag    ("noauto_da_alloc",     Opt_noauto_da_alloc),
1764         fsparam_flag    ("dioread_nolock",      Opt_dioread_nolock),
1765         fsparam_flag    ("nodioread_nolock",    Opt_dioread_lock),
1766         fsparam_flag    ("dioread_lock",        Opt_dioread_lock),
1767         fsparam_flag    ("discard",             Opt_discard),
1768         fsparam_flag    ("nodiscard",           Opt_nodiscard),
1769         fsparam_u32     ("init_itable",         Opt_init_itable),
1770         fsparam_flag    ("init_itable",         Opt_init_itable),
1771         fsparam_flag    ("noinit_itable",       Opt_noinit_itable),
1772 #ifdef CONFIG_EXT4_DEBUG
1773         fsparam_flag    ("fc_debug_force",      Opt_fc_debug_force),
1774         fsparam_u32     ("fc_debug_max_replay", Opt_fc_debug_max_replay),
1775 #endif
1776         fsparam_u32     ("max_dir_size_kb",     Opt_max_dir_size_kb),
1777         fsparam_flag    ("test_dummy_encryption",
1778                                                 Opt_test_dummy_encryption),
1779         fsparam_string  ("test_dummy_encryption",
1780                                                 Opt_test_dummy_encryption),
1781         fsparam_flag    ("inlinecrypt",         Opt_inlinecrypt),
1782         fsparam_flag    ("nombcache",           Opt_nombcache),
1783         fsparam_flag    ("no_mbcache",          Opt_nombcache), /* for backward compatibility */
1784         fsparam_flag    ("prefetch_block_bitmaps",
1785                                                 Opt_removed),
1786         fsparam_flag    ("no_prefetch_block_bitmaps",
1787                                                 Opt_no_prefetch_block_bitmaps),
1788         fsparam_s32     ("mb_optimize_scan",    Opt_mb_optimize_scan),
1789         fsparam_string  ("check",               Opt_removed),   /* mount option from ext2/3 */
1790         fsparam_flag    ("nocheck",             Opt_removed),   /* mount option from ext2/3 */
1791         fsparam_flag    ("reservation",         Opt_removed),   /* mount option from ext2/3 */
1792         fsparam_flag    ("noreservation",       Opt_removed),   /* mount option from ext2/3 */
1793         fsparam_u32     ("journal",             Opt_removed),   /* mount option from ext2/3 */
1794         {}
1795 };
1796
1797 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1798
1799 #define MOPT_SET        0x0001
1800 #define MOPT_CLEAR      0x0002
1801 #define MOPT_NOSUPPORT  0x0004
1802 #define MOPT_EXPLICIT   0x0008
1803 #ifdef CONFIG_QUOTA
1804 #define MOPT_Q          0
1805 #define MOPT_QFMT       0x0010
1806 #else
1807 #define MOPT_Q          MOPT_NOSUPPORT
1808 #define MOPT_QFMT       MOPT_NOSUPPORT
1809 #endif
1810 #define MOPT_NO_EXT2    0x0020
1811 #define MOPT_NO_EXT3    0x0040
1812 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1813 #define MOPT_SKIP       0x0080
1814 #define MOPT_2          0x0100
1815
1816 static const struct mount_opts {
1817         int     token;
1818         int     mount_opt;
1819         int     flags;
1820 } ext4_mount_opts[] = {
1821         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1822         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1823         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1824         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1825         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1826         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1827         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1828          MOPT_EXT4_ONLY | MOPT_SET},
1829         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1830          MOPT_EXT4_ONLY | MOPT_CLEAR},
1831         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1832         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1833         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1834          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1835         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1836          MOPT_EXT4_ONLY | MOPT_CLEAR},
1837         {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET},
1838         {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR},
1839         {Opt_commit, 0, MOPT_NO_EXT2},
1840         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1841          MOPT_EXT4_ONLY | MOPT_CLEAR},
1842         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1843          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1844         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1845                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1846          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1847         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1848         {Opt_data_err, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_NO_EXT2},
1849         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1850         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1851         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1852         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1853         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1854         {Opt_dax_type, 0, MOPT_EXT4_ONLY},
1855         {Opt_journal_dev, 0, MOPT_NO_EXT2},
1856         {Opt_journal_path, 0, MOPT_NO_EXT2},
1857         {Opt_journal_ioprio, 0, MOPT_NO_EXT2},
1858         {Opt_data, 0, MOPT_NO_EXT2},
1859         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1860 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1861         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1862 #else
1863         {Opt_acl, 0, MOPT_NOSUPPORT},
1864 #endif
1865         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1866         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1867         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1868         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1869                                                         MOPT_SET | MOPT_Q},
1870         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1871                                                         MOPT_SET | MOPT_Q},
1872         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1873                                                         MOPT_SET | MOPT_Q},
1874         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1875                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1876                                                         MOPT_CLEAR | MOPT_Q},
1877         {Opt_usrjquota, 0, MOPT_Q},
1878         {Opt_grpjquota, 0, MOPT_Q},
1879         {Opt_jqfmt, 0, MOPT_QFMT},
1880         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1881         {Opt_no_prefetch_block_bitmaps, EXT4_MOUNT_NO_PREFETCH_BLOCK_BITMAPS,
1882          MOPT_SET},
1883 #ifdef CONFIG_EXT4_DEBUG
1884         {Opt_fc_debug_force, EXT4_MOUNT2_JOURNAL_FAST_COMMIT,
1885          MOPT_SET | MOPT_2 | MOPT_EXT4_ONLY},
1886 #endif
1887         {Opt_err, 0, 0}
1888 };
1889
1890 #if IS_ENABLED(CONFIG_UNICODE)
1891 static const struct ext4_sb_encodings {
1892         __u16 magic;
1893         char *name;
1894         unsigned int version;
1895 } ext4_sb_encoding_map[] = {
1896         {EXT4_ENC_UTF8_12_1, "utf8", UNICODE_AGE(12, 1, 0)},
1897 };
1898
1899 static const struct ext4_sb_encodings *
1900 ext4_sb_read_encoding(const struct ext4_super_block *es)
1901 {
1902         __u16 magic = le16_to_cpu(es->s_encoding);
1903         int i;
1904
1905         for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++)
1906                 if (magic == ext4_sb_encoding_map[i].magic)
1907                         return &ext4_sb_encoding_map[i];
1908
1909         return NULL;
1910 }
1911 #endif
1912
1913 #define EXT4_SPEC_JQUOTA                        (1 <<  0)
1914 #define EXT4_SPEC_JQFMT                         (1 <<  1)
1915 #define EXT4_SPEC_DATAJ                         (1 <<  2)
1916 #define EXT4_SPEC_SB_BLOCK                      (1 <<  3)
1917 #define EXT4_SPEC_JOURNAL_DEV                   (1 <<  4)
1918 #define EXT4_SPEC_JOURNAL_IOPRIO                (1 <<  5)
1919 #define EXT4_SPEC_s_want_extra_isize            (1 <<  7)
1920 #define EXT4_SPEC_s_max_batch_time              (1 <<  8)
1921 #define EXT4_SPEC_s_min_batch_time              (1 <<  9)
1922 #define EXT4_SPEC_s_inode_readahead_blks        (1 << 10)
1923 #define EXT4_SPEC_s_li_wait_mult                (1 << 11)
1924 #define EXT4_SPEC_s_max_dir_size_kb             (1 << 12)
1925 #define EXT4_SPEC_s_stripe                      (1 << 13)
1926 #define EXT4_SPEC_s_resuid                      (1 << 14)
1927 #define EXT4_SPEC_s_resgid                      (1 << 15)
1928 #define EXT4_SPEC_s_commit_interval             (1 << 16)
1929 #define EXT4_SPEC_s_fc_debug_max_replay         (1 << 17)
1930 #define EXT4_SPEC_s_sb_block                    (1 << 18)
1931 #define EXT4_SPEC_mb_optimize_scan              (1 << 19)
1932
1933 struct ext4_fs_context {
1934         char            *s_qf_names[EXT4_MAXQUOTAS];
1935         struct fscrypt_dummy_policy dummy_enc_policy;
1936         int             s_jquota_fmt;   /* Format of quota to use */
1937 #ifdef CONFIG_EXT4_DEBUG
1938         int s_fc_debug_max_replay;
1939 #endif
1940         unsigned short  qname_spec;
1941         unsigned long   vals_s_flags;   /* Bits to set in s_flags */
1942         unsigned long   mask_s_flags;   /* Bits changed in s_flags */
1943         unsigned long   journal_devnum;
1944         unsigned long   s_commit_interval;
1945         unsigned long   s_stripe;
1946         unsigned int    s_inode_readahead_blks;
1947         unsigned int    s_want_extra_isize;
1948         unsigned int    s_li_wait_mult;
1949         unsigned int    s_max_dir_size_kb;
1950         unsigned int    journal_ioprio;
1951         unsigned int    vals_s_mount_opt;
1952         unsigned int    mask_s_mount_opt;
1953         unsigned int    vals_s_mount_opt2;
1954         unsigned int    mask_s_mount_opt2;
1955         unsigned long   vals_s_mount_flags;
1956         unsigned long   mask_s_mount_flags;
1957         unsigned int    opt_flags;      /* MOPT flags */
1958         unsigned int    spec;
1959         u32             s_max_batch_time;
1960         u32             s_min_batch_time;
1961         kuid_t          s_resuid;
1962         kgid_t          s_resgid;
1963         ext4_fsblk_t    s_sb_block;
1964 };
1965
1966 static void ext4_fc_free(struct fs_context *fc)
1967 {
1968         struct ext4_fs_context *ctx = fc->fs_private;
1969         int i;
1970
1971         if (!ctx)
1972                 return;
1973
1974         for (i = 0; i < EXT4_MAXQUOTAS; i++)
1975                 kfree(ctx->s_qf_names[i]);
1976
1977         fscrypt_free_dummy_policy(&ctx->dummy_enc_policy);
1978         kfree(ctx);
1979 }
1980
1981 int ext4_init_fs_context(struct fs_context *fc)
1982 {
1983         struct ext4_fs_context *ctx;
1984
1985         ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL);
1986         if (!ctx)
1987                 return -ENOMEM;
1988
1989         fc->fs_private = ctx;
1990         fc->ops = &ext4_context_ops;
1991
1992         return 0;
1993 }
1994
1995 #ifdef CONFIG_QUOTA
1996 /*
1997  * Note the name of the specified quota file.
1998  */
1999 static int note_qf_name(struct fs_context *fc, int qtype,
2000                        struct fs_parameter *param)
2001 {
2002         struct ext4_fs_context *ctx = fc->fs_private;
2003         char *qname;
2004
2005         if (param->size < 1) {
2006                 ext4_msg(NULL, KERN_ERR, "Missing quota name");
2007                 return -EINVAL;
2008         }
2009         if (strchr(param->string, '/')) {
2010                 ext4_msg(NULL, KERN_ERR,
2011                          "quotafile must be on filesystem root");
2012                 return -EINVAL;
2013         }
2014         if (ctx->s_qf_names[qtype]) {
2015                 if (strcmp(ctx->s_qf_names[qtype], param->string) != 0) {
2016                         ext4_msg(NULL, KERN_ERR,
2017                                  "%s quota file already specified",
2018                                  QTYPE2NAME(qtype));
2019                         return -EINVAL;
2020                 }
2021                 return 0;
2022         }
2023
2024         qname = kmemdup_nul(param->string, param->size, GFP_KERNEL);
2025         if (!qname) {
2026                 ext4_msg(NULL, KERN_ERR,
2027                          "Not enough memory for storing quotafile name");
2028                 return -ENOMEM;
2029         }
2030         ctx->s_qf_names[qtype] = qname;
2031         ctx->qname_spec |= 1 << qtype;
2032         ctx->spec |= EXT4_SPEC_JQUOTA;
2033         return 0;
2034 }
2035
2036 /*
2037  * Clear the name of the specified quota file.
2038  */
2039 static int unnote_qf_name(struct fs_context *fc, int qtype)
2040 {
2041         struct ext4_fs_context *ctx = fc->fs_private;
2042
2043         if (ctx->s_qf_names[qtype])
2044                 kfree(ctx->s_qf_names[qtype]);
2045
2046         ctx->s_qf_names[qtype] = NULL;
2047         ctx->qname_spec |= 1 << qtype;
2048         ctx->spec |= EXT4_SPEC_JQUOTA;
2049         return 0;
2050 }
2051 #endif
2052
2053 static int ext4_parse_test_dummy_encryption(const struct fs_parameter *param,
2054                                             struct ext4_fs_context *ctx)
2055 {
2056         int err;
2057
2058         if (!IS_ENABLED(CONFIG_FS_ENCRYPTION)) {
2059                 ext4_msg(NULL, KERN_WARNING,
2060                          "test_dummy_encryption option not supported");
2061                 return -EINVAL;
2062         }
2063         err = fscrypt_parse_test_dummy_encryption(param,
2064                                                   &ctx->dummy_enc_policy);
2065         if (err == -EINVAL) {
2066                 ext4_msg(NULL, KERN_WARNING,
2067                          "Value of option \"%s\" is unrecognized", param->key);
2068         } else if (err == -EEXIST) {
2069                 ext4_msg(NULL, KERN_WARNING,
2070                          "Conflicting test_dummy_encryption options");
2071                 return -EINVAL;
2072         }
2073         return err;
2074 }
2075
2076 #define EXT4_SET_CTX(name)                                              \
2077 static inline void ctx_set_##name(struct ext4_fs_context *ctx,          \
2078                                   unsigned long flag)                   \
2079 {                                                                       \
2080         ctx->mask_s_##name |= flag;                                     \
2081         ctx->vals_s_##name |= flag;                                     \
2082 }
2083
2084 #define EXT4_CLEAR_CTX(name)                                            \
2085 static inline void ctx_clear_##name(struct ext4_fs_context *ctx,        \
2086                                     unsigned long flag)                 \
2087 {                                                                       \
2088         ctx->mask_s_##name |= flag;                                     \
2089         ctx->vals_s_##name &= ~flag;                                    \
2090 }
2091
2092 #define EXT4_TEST_CTX(name)                                             \
2093 static inline unsigned long                                             \
2094 ctx_test_##name(struct ext4_fs_context *ctx, unsigned long flag)        \
2095 {                                                                       \
2096         return (ctx->vals_s_##name & flag);                             \
2097 }
2098
2099 EXT4_SET_CTX(flags); /* set only */
2100 EXT4_SET_CTX(mount_opt);
2101 EXT4_CLEAR_CTX(mount_opt);
2102 EXT4_TEST_CTX(mount_opt);
2103 EXT4_SET_CTX(mount_opt2);
2104 EXT4_CLEAR_CTX(mount_opt2);
2105 EXT4_TEST_CTX(mount_opt2);
2106
2107 static inline void ctx_set_mount_flag(struct ext4_fs_context *ctx, int bit)
2108 {
2109         set_bit(bit, &ctx->mask_s_mount_flags);
2110         set_bit(bit, &ctx->vals_s_mount_flags);
2111 }
2112
2113 static int ext4_parse_param(struct fs_context *fc, struct fs_parameter *param)
2114 {
2115         struct ext4_fs_context *ctx = fc->fs_private;
2116         struct fs_parse_result result;
2117         const struct mount_opts *m;
2118         int is_remount;
2119         kuid_t uid;
2120         kgid_t gid;
2121         int token;
2122
2123         token = fs_parse(fc, ext4_param_specs, param, &result);
2124         if (token < 0)
2125                 return token;
2126         is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE;
2127
2128         for (m = ext4_mount_opts; m->token != Opt_err; m++)
2129                 if (token == m->token)
2130                         break;
2131
2132         ctx->opt_flags |= m->flags;
2133
2134         if (m->flags & MOPT_EXPLICIT) {
2135                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
2136                         ctx_set_mount_opt2(ctx, EXT4_MOUNT2_EXPLICIT_DELALLOC);
2137                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
2138                         ctx_set_mount_opt2(ctx,
2139                                        EXT4_MOUNT2_EXPLICIT_JOURNAL_CHECKSUM);
2140                 } else
2141                         return -EINVAL;
2142         }
2143
2144         if (m->flags & MOPT_NOSUPPORT) {
2145                 ext4_msg(NULL, KERN_ERR, "%s option not supported",
2146                          param->key);
2147                 return 0;
2148         }
2149
2150         switch (token) {
2151 #ifdef CONFIG_QUOTA
2152         case Opt_usrjquota:
2153                 if (!*param->string)
2154                         return unnote_qf_name(fc, USRQUOTA);
2155                 else
2156                         return note_qf_name(fc, USRQUOTA, param);
2157         case Opt_grpjquota:
2158                 if (!*param->string)
2159                         return unnote_qf_name(fc, GRPQUOTA);
2160                 else
2161                         return note_qf_name(fc, GRPQUOTA, param);
2162 #endif
2163         case Opt_sb:
2164                 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
2165                         ext4_msg(NULL, KERN_WARNING,
2166                                  "Ignoring %s option on remount", param->key);
2167                 } else {
2168                         ctx->s_sb_block = result.uint_32;
2169                         ctx->spec |= EXT4_SPEC_s_sb_block;
2170                 }
2171                 return 0;
2172         case Opt_removed:
2173                 ext4_msg(NULL, KERN_WARNING, "Ignoring removed %s option",
2174                          param->key);
2175                 return 0;
2176         case Opt_abort:
2177                 ctx_set_mount_flag(ctx, EXT4_MF_FS_ABORTED);
2178                 return 0;
2179         case Opt_inlinecrypt:
2180 #ifdef CONFIG_FS_ENCRYPTION_INLINE_CRYPT
2181                 ctx_set_flags(ctx, SB_INLINECRYPT);
2182 #else
2183                 ext4_msg(NULL, KERN_ERR, "inline encryption not supported");
2184 #endif
2185                 return 0;
2186         case Opt_errors:
2187                 ctx_clear_mount_opt(ctx, EXT4_MOUNT_ERRORS_MASK);
2188                 ctx_set_mount_opt(ctx, result.uint_32);
2189                 return 0;
2190 #ifdef CONFIG_QUOTA
2191         case Opt_jqfmt:
2192                 ctx->s_jquota_fmt = result.uint_32;
2193                 ctx->spec |= EXT4_SPEC_JQFMT;
2194                 return 0;
2195 #endif
2196         case Opt_data:
2197                 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS);
2198                 ctx_set_mount_opt(ctx, result.uint_32);
2199                 ctx->spec |= EXT4_SPEC_DATAJ;
2200                 return 0;
2201         case Opt_commit:
2202                 if (result.uint_32 == 0)
2203                         result.uint_32 = JBD2_DEFAULT_MAX_COMMIT_AGE;
2204                 else if (result.uint_32 > INT_MAX / HZ) {
2205                         ext4_msg(NULL, KERN_ERR,
2206                                  "Invalid commit interval %d, "
2207                                  "must be smaller than %d",
2208                                  result.uint_32, INT_MAX / HZ);
2209                         return -EINVAL;
2210                 }
2211                 ctx->s_commit_interval = HZ * result.uint_32;
2212                 ctx->spec |= EXT4_SPEC_s_commit_interval;
2213                 return 0;
2214         case Opt_debug_want_extra_isize:
2215                 if ((result.uint_32 & 1) || (result.uint_32 < 4)) {
2216                         ext4_msg(NULL, KERN_ERR,
2217                                  "Invalid want_extra_isize %d", result.uint_32);
2218                         return -EINVAL;
2219                 }
2220                 ctx->s_want_extra_isize = result.uint_32;
2221                 ctx->spec |= EXT4_SPEC_s_want_extra_isize;
2222                 return 0;
2223         case Opt_max_batch_time:
2224                 ctx->s_max_batch_time = result.uint_32;
2225                 ctx->spec |= EXT4_SPEC_s_max_batch_time;
2226                 return 0;
2227         case Opt_min_batch_time:
2228                 ctx->s_min_batch_time = result.uint_32;
2229                 ctx->spec |= EXT4_SPEC_s_min_batch_time;
2230                 return 0;
2231         case Opt_inode_readahead_blks:
2232                 if (result.uint_32 &&
2233                     (result.uint_32 > (1 << 30) ||
2234                      !is_power_of_2(result.uint_32))) {
2235                         ext4_msg(NULL, KERN_ERR,
2236                                  "EXT4-fs: inode_readahead_blks must be "
2237                                  "0 or a power of 2 smaller than 2^31");
2238                         return -EINVAL;
2239                 }
2240                 ctx->s_inode_readahead_blks = result.uint_32;
2241                 ctx->spec |= EXT4_SPEC_s_inode_readahead_blks;
2242                 return 0;
2243         case Opt_init_itable:
2244                 ctx_set_mount_opt(ctx, EXT4_MOUNT_INIT_INODE_TABLE);
2245                 ctx->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
2246                 if (param->type == fs_value_is_string)
2247                         ctx->s_li_wait_mult = result.uint_32;
2248                 ctx->spec |= EXT4_SPEC_s_li_wait_mult;
2249                 return 0;
2250         case Opt_max_dir_size_kb:
2251                 ctx->s_max_dir_size_kb = result.uint_32;
2252                 ctx->spec |= EXT4_SPEC_s_max_dir_size_kb;
2253                 return 0;
2254 #ifdef CONFIG_EXT4_DEBUG
2255         case Opt_fc_debug_max_replay:
2256                 ctx->s_fc_debug_max_replay = result.uint_32;
2257                 ctx->spec |= EXT4_SPEC_s_fc_debug_max_replay;
2258                 return 0;
2259 #endif
2260         case Opt_stripe:
2261                 ctx->s_stripe = result.uint_32;
2262                 ctx->spec |= EXT4_SPEC_s_stripe;
2263                 return 0;
2264         case Opt_resuid:
2265                 uid = make_kuid(current_user_ns(), result.uint_32);
2266                 if (!uid_valid(uid)) {
2267                         ext4_msg(NULL, KERN_ERR, "Invalid uid value %d",
2268                                  result.uint_32);
2269                         return -EINVAL;
2270                 }
2271                 ctx->s_resuid = uid;
2272                 ctx->spec |= EXT4_SPEC_s_resuid;
2273                 return 0;
2274         case Opt_resgid:
2275                 gid = make_kgid(current_user_ns(), result.uint_32);
2276                 if (!gid_valid(gid)) {
2277                         ext4_msg(NULL, KERN_ERR, "Invalid gid value %d",
2278                                  result.uint_32);
2279                         return -EINVAL;
2280                 }
2281                 ctx->s_resgid = gid;
2282                 ctx->spec |= EXT4_SPEC_s_resgid;
2283                 return 0;
2284         case Opt_journal_dev:
2285                 if (is_remount) {
2286                         ext4_msg(NULL, KERN_ERR,
2287                                  "Cannot specify journal on remount");
2288                         return -EINVAL;
2289                 }
2290                 ctx->journal_devnum = result.uint_32;
2291                 ctx->spec |= EXT4_SPEC_JOURNAL_DEV;
2292                 return 0;
2293         case Opt_journal_path:
2294         {
2295                 struct inode *journal_inode;
2296                 struct path path;
2297                 int error;
2298
2299                 if (is_remount) {
2300                         ext4_msg(NULL, KERN_ERR,
2301                                  "Cannot specify journal on remount");
2302                         return -EINVAL;
2303                 }
2304
2305                 error = fs_lookup_param(fc, param, 1, LOOKUP_FOLLOW, &path);
2306                 if (error) {
2307                         ext4_msg(NULL, KERN_ERR, "error: could not find "
2308                                  "journal device path");
2309                         return -EINVAL;
2310                 }
2311
2312                 journal_inode = d_inode(path.dentry);
2313                 ctx->journal_devnum = new_encode_dev(journal_inode->i_rdev);
2314                 ctx->spec |= EXT4_SPEC_JOURNAL_DEV;
2315                 path_put(&path);
2316                 return 0;
2317         }
2318         case Opt_journal_ioprio:
2319                 if (result.uint_32 > 7) {
2320                         ext4_msg(NULL, KERN_ERR, "Invalid journal IO priority"
2321                                  " (must be 0-7)");
2322                         return -EINVAL;
2323                 }
2324                 ctx->journal_ioprio =
2325                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, result.uint_32);
2326                 ctx->spec |= EXT4_SPEC_JOURNAL_IOPRIO;
2327                 return 0;
2328         case Opt_test_dummy_encryption:
2329                 return ext4_parse_test_dummy_encryption(param, ctx);
2330         case Opt_dax:
2331         case Opt_dax_type:
2332 #ifdef CONFIG_FS_DAX
2333         {
2334                 int type = (token == Opt_dax) ?
2335                            Opt_dax : result.uint_32;
2336
2337                 switch (type) {
2338                 case Opt_dax:
2339                 case Opt_dax_always:
2340                         ctx_set_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS);
2341                         ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER);
2342                         break;
2343                 case Opt_dax_never:
2344                         ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER);
2345                         ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS);
2346                         break;
2347                 case Opt_dax_inode:
2348                         ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS);
2349                         ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER);
2350                         /* Strictly for printing options */
2351                         ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE);
2352                         break;
2353                 }
2354                 return 0;
2355         }
2356 #else
2357                 ext4_msg(NULL, KERN_INFO, "dax option not supported");
2358                 return -EINVAL;
2359 #endif
2360         case Opt_data_err:
2361                 if (result.uint_32 == Opt_data_err_abort)
2362                         ctx_set_mount_opt(ctx, m->mount_opt);
2363                 else if (result.uint_32 == Opt_data_err_ignore)
2364                         ctx_clear_mount_opt(ctx, m->mount_opt);
2365                 return 0;
2366         case Opt_mb_optimize_scan:
2367                 if (result.int_32 == 1) {
2368                         ctx_set_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN);
2369                         ctx->spec |= EXT4_SPEC_mb_optimize_scan;
2370                 } else if (result.int_32 == 0) {
2371                         ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN);
2372                         ctx->spec |= EXT4_SPEC_mb_optimize_scan;
2373                 } else {
2374                         ext4_msg(NULL, KERN_WARNING,
2375                                  "mb_optimize_scan should be set to 0 or 1.");
2376                         return -EINVAL;
2377                 }
2378                 return 0;
2379         }
2380
2381         /*
2382          * At this point we should only be getting options requiring MOPT_SET,
2383          * or MOPT_CLEAR. Anything else is a bug
2384          */
2385         if (m->token == Opt_err) {
2386                 ext4_msg(NULL, KERN_WARNING, "buggy handling of option %s",
2387                          param->key);
2388                 WARN_ON(1);
2389                 return -EINVAL;
2390         }
2391
2392         else {
2393                 unsigned int set = 0;
2394
2395                 if ((param->type == fs_value_is_flag) ||
2396                     result.uint_32 > 0)
2397                         set = 1;
2398
2399                 if (m->flags & MOPT_CLEAR)
2400                         set = !set;
2401                 else if (unlikely(!(m->flags & MOPT_SET))) {
2402                         ext4_msg(NULL, KERN_WARNING,
2403                                  "buggy handling of option %s",
2404                                  param->key);
2405                         WARN_ON(1);
2406                         return -EINVAL;
2407                 }
2408                 if (m->flags & MOPT_2) {
2409                         if (set != 0)
2410                                 ctx_set_mount_opt2(ctx, m->mount_opt);
2411                         else
2412                                 ctx_clear_mount_opt2(ctx, m->mount_opt);
2413                 } else {
2414                         if (set != 0)
2415                                 ctx_set_mount_opt(ctx, m->mount_opt);
2416                         else
2417                                 ctx_clear_mount_opt(ctx, m->mount_opt);
2418                 }
2419         }
2420
2421         return 0;
2422 }
2423
2424 static int parse_options(struct fs_context *fc, char *options)
2425 {
2426         struct fs_parameter param;
2427         int ret;
2428         char *key;
2429
2430         if (!options)
2431                 return 0;
2432
2433         while ((key = strsep(&options, ",")) != NULL) {
2434                 if (*key) {
2435                         size_t v_len = 0;
2436                         char *value = strchr(key, '=');
2437
2438                         param.type = fs_value_is_flag;
2439                         param.string = NULL;
2440
2441                         if (value) {
2442                                 if (value == key)
2443                                         continue;
2444
2445                                 *value++ = 0;
2446                                 v_len = strlen(value);
2447                                 param.string = kmemdup_nul(value, v_len,
2448                                                            GFP_KERNEL);
2449                                 if (!param.string)
2450                                         return -ENOMEM;
2451                                 param.type = fs_value_is_string;
2452                         }
2453
2454                         param.key = key;
2455                         param.size = v_len;
2456
2457                         ret = ext4_parse_param(fc, &param);
2458                         if (param.string)
2459                                 kfree(param.string);
2460                         if (ret < 0)
2461                                 return ret;
2462                 }
2463         }
2464
2465         ret = ext4_validate_options(fc);
2466         if (ret < 0)
2467                 return ret;
2468
2469         return 0;
2470 }
2471
2472 static int parse_apply_sb_mount_options(struct super_block *sb,
2473                                         struct ext4_fs_context *m_ctx)
2474 {
2475         struct ext4_sb_info *sbi = EXT4_SB(sb);
2476         char *s_mount_opts = NULL;
2477         struct ext4_fs_context *s_ctx = NULL;
2478         struct fs_context *fc = NULL;
2479         int ret = -ENOMEM;
2480
2481         if (!sbi->s_es->s_mount_opts[0])
2482                 return 0;
2483
2484         s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
2485                                 sizeof(sbi->s_es->s_mount_opts),
2486                                 GFP_KERNEL);
2487         if (!s_mount_opts)
2488                 return ret;
2489
2490         fc = kzalloc(sizeof(struct fs_context), GFP_KERNEL);
2491         if (!fc)
2492                 goto out_free;
2493
2494         s_ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL);
2495         if (!s_ctx)
2496                 goto out_free;
2497
2498         fc->fs_private = s_ctx;
2499         fc->s_fs_info = sbi;
2500
2501         ret = parse_options(fc, s_mount_opts);
2502         if (ret < 0)
2503                 goto parse_failed;
2504
2505         ret = ext4_check_opt_consistency(fc, sb);
2506         if (ret < 0) {
2507 parse_failed:
2508                 ext4_msg(sb, KERN_WARNING,
2509                          "failed to parse options in superblock: %s",
2510                          s_mount_opts);
2511                 ret = 0;
2512                 goto out_free;
2513         }
2514
2515         if (s_ctx->spec & EXT4_SPEC_JOURNAL_DEV)
2516                 m_ctx->journal_devnum = s_ctx->journal_devnum;
2517         if (s_ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)
2518                 m_ctx->journal_ioprio = s_ctx->journal_ioprio;
2519
2520         ext4_apply_options(fc, sb);
2521         ret = 0;
2522
2523 out_free:
2524         if (fc) {
2525                 ext4_fc_free(fc);
2526                 kfree(fc);
2527         }
2528         kfree(s_mount_opts);
2529         return ret;
2530 }
2531
2532 static void ext4_apply_quota_options(struct fs_context *fc,
2533                                      struct super_block *sb)
2534 {
2535 #ifdef CONFIG_QUOTA
2536         bool quota_feature = ext4_has_feature_quota(sb);
2537         struct ext4_fs_context *ctx = fc->fs_private;
2538         struct ext4_sb_info *sbi = EXT4_SB(sb);
2539         char *qname;
2540         int i;
2541
2542         if (quota_feature)
2543                 return;
2544
2545         if (ctx->spec & EXT4_SPEC_JQUOTA) {
2546                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2547                         if (!(ctx->qname_spec & (1 << i)))
2548                                 continue;
2549
2550                         qname = ctx->s_qf_names[i]; /* May be NULL */
2551                         if (qname)
2552                                 set_opt(sb, QUOTA);
2553                         ctx->s_qf_names[i] = NULL;
2554                         qname = rcu_replace_pointer(sbi->s_qf_names[i], qname,
2555                                                 lockdep_is_held(&sb->s_umount));
2556                         if (qname)
2557                                 kfree_rcu_mightsleep(qname);
2558                 }
2559         }
2560
2561         if (ctx->spec & EXT4_SPEC_JQFMT)
2562                 sbi->s_jquota_fmt = ctx->s_jquota_fmt;
2563 #endif
2564 }
2565
2566 /*
2567  * Check quota settings consistency.
2568  */
2569 static int ext4_check_quota_consistency(struct fs_context *fc,
2570                                         struct super_block *sb)
2571 {
2572 #ifdef CONFIG_QUOTA
2573         struct ext4_fs_context *ctx = fc->fs_private;
2574         struct ext4_sb_info *sbi = EXT4_SB(sb);
2575         bool quota_feature = ext4_has_feature_quota(sb);
2576         bool quota_loaded = sb_any_quota_loaded(sb);
2577         bool usr_qf_name, grp_qf_name, usrquota, grpquota;
2578         int quota_flags, i;
2579
2580         /*
2581          * We do the test below only for project quotas. 'usrquota' and
2582          * 'grpquota' mount options are allowed even without quota feature
2583          * to support legacy quotas in quota files.
2584          */
2585         if (ctx_test_mount_opt(ctx, EXT4_MOUNT_PRJQUOTA) &&
2586             !ext4_has_feature_project(sb)) {
2587                 ext4_msg(NULL, KERN_ERR, "Project quota feature not enabled. "
2588                          "Cannot enable project quota enforcement.");
2589                 return -EINVAL;
2590         }
2591
2592         quota_flags = EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
2593                       EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA;
2594         if (quota_loaded &&
2595             ctx->mask_s_mount_opt & quota_flags &&
2596             !ctx_test_mount_opt(ctx, quota_flags))
2597                 goto err_quota_change;
2598
2599         if (ctx->spec & EXT4_SPEC_JQUOTA) {
2600
2601                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2602                         if (!(ctx->qname_spec & (1 << i)))
2603                                 continue;
2604
2605                         if (quota_loaded &&
2606                             !!sbi->s_qf_names[i] != !!ctx->s_qf_names[i])
2607                                 goto err_jquota_change;
2608
2609                         if (sbi->s_qf_names[i] && ctx->s_qf_names[i] &&
2610                             strcmp(get_qf_name(sb, sbi, i),
2611                                    ctx->s_qf_names[i]) != 0)
2612                                 goto err_jquota_specified;
2613                 }
2614
2615                 if (quota_feature) {
2616                         ext4_msg(NULL, KERN_INFO,
2617                                  "Journaled quota options ignored when "
2618                                  "QUOTA feature is enabled");
2619                         return 0;
2620                 }
2621         }
2622
2623         if (ctx->spec & EXT4_SPEC_JQFMT) {
2624                 if (sbi->s_jquota_fmt != ctx->s_jquota_fmt && quota_loaded)
2625                         goto err_jquota_change;
2626                 if (quota_feature) {
2627                         ext4_msg(NULL, KERN_INFO, "Quota format mount options "
2628                                  "ignored when QUOTA feature is enabled");
2629                         return 0;
2630                 }
2631         }
2632
2633         /* Make sure we don't mix old and new quota format */
2634         usr_qf_name = (get_qf_name(sb, sbi, USRQUOTA) ||
2635                        ctx->s_qf_names[USRQUOTA]);
2636         grp_qf_name = (get_qf_name(sb, sbi, GRPQUOTA) ||
2637                        ctx->s_qf_names[GRPQUOTA]);
2638
2639         usrquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) ||
2640                     test_opt(sb, USRQUOTA));
2641
2642         grpquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) ||
2643                     test_opt(sb, GRPQUOTA));
2644
2645         if (usr_qf_name) {
2646                 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA);
2647                 usrquota = false;
2648         }
2649         if (grp_qf_name) {
2650                 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA);
2651                 grpquota = false;
2652         }
2653
2654         if (usr_qf_name || grp_qf_name) {
2655                 if (usrquota || grpquota) {
2656                         ext4_msg(NULL, KERN_ERR, "old and new quota "
2657                                  "format mixing");
2658                         return -EINVAL;
2659                 }
2660
2661                 if (!(ctx->spec & EXT4_SPEC_JQFMT || sbi->s_jquota_fmt)) {
2662                         ext4_msg(NULL, KERN_ERR, "journaled quota format "
2663                                  "not specified");
2664                         return -EINVAL;
2665                 }
2666         }
2667
2668         return 0;
2669
2670 err_quota_change:
2671         ext4_msg(NULL, KERN_ERR,
2672                  "Cannot change quota options when quota turned on");
2673         return -EINVAL;
2674 err_jquota_change:
2675         ext4_msg(NULL, KERN_ERR, "Cannot change journaled quota "
2676                  "options when quota turned on");
2677         return -EINVAL;
2678 err_jquota_specified:
2679         ext4_msg(NULL, KERN_ERR, "%s quota file already specified",
2680                  QTYPE2NAME(i));
2681         return -EINVAL;
2682 #else
2683         return 0;
2684 #endif
2685 }
2686
2687 static int ext4_check_test_dummy_encryption(const struct fs_context *fc,
2688                                             struct super_block *sb)
2689 {
2690         const struct ext4_fs_context *ctx = fc->fs_private;
2691         const struct ext4_sb_info *sbi = EXT4_SB(sb);
2692
2693         if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy))
2694                 return 0;
2695
2696         if (!ext4_has_feature_encrypt(sb)) {
2697                 ext4_msg(NULL, KERN_WARNING,
2698                          "test_dummy_encryption requires encrypt feature");
2699                 return -EINVAL;
2700         }
2701         /*
2702          * This mount option is just for testing, and it's not worthwhile to
2703          * implement the extra complexity (e.g. RCU protection) that would be
2704          * needed to allow it to be set or changed during remount.  We do allow
2705          * it to be specified during remount, but only if there is no change.
2706          */
2707         if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
2708                 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy,
2709                                                  &ctx->dummy_enc_policy))
2710                         return 0;
2711                 ext4_msg(NULL, KERN_WARNING,
2712                          "Can't set or change test_dummy_encryption on remount");
2713                 return -EINVAL;
2714         }
2715         /* Also make sure s_mount_opts didn't contain a conflicting value. */
2716         if (fscrypt_is_dummy_policy_set(&sbi->s_dummy_enc_policy)) {
2717                 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy,
2718                                                  &ctx->dummy_enc_policy))
2719                         return 0;
2720                 ext4_msg(NULL, KERN_WARNING,
2721                          "Conflicting test_dummy_encryption options");
2722                 return -EINVAL;
2723         }
2724         return 0;
2725 }
2726
2727 static void ext4_apply_test_dummy_encryption(struct ext4_fs_context *ctx,
2728                                              struct super_block *sb)
2729 {
2730         if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy) ||
2731             /* if already set, it was already verified to be the same */
2732             fscrypt_is_dummy_policy_set(&EXT4_SB(sb)->s_dummy_enc_policy))
2733                 return;
2734         EXT4_SB(sb)->s_dummy_enc_policy = ctx->dummy_enc_policy;
2735         memset(&ctx->dummy_enc_policy, 0, sizeof(ctx->dummy_enc_policy));
2736         ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled");
2737 }
2738
2739 static int ext4_check_opt_consistency(struct fs_context *fc,
2740                                       struct super_block *sb)
2741 {
2742         struct ext4_fs_context *ctx = fc->fs_private;
2743         struct ext4_sb_info *sbi = fc->s_fs_info;
2744         int is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE;
2745         int err;
2746
2747         if ((ctx->opt_flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
2748                 ext4_msg(NULL, KERN_ERR,
2749                          "Mount option(s) incompatible with ext2");
2750                 return -EINVAL;
2751         }
2752         if ((ctx->opt_flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
2753                 ext4_msg(NULL, KERN_ERR,
2754                          "Mount option(s) incompatible with ext3");
2755                 return -EINVAL;
2756         }
2757
2758         if (ctx->s_want_extra_isize >
2759             (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE)) {
2760                 ext4_msg(NULL, KERN_ERR,
2761                          "Invalid want_extra_isize %d",
2762                          ctx->s_want_extra_isize);
2763                 return -EINVAL;
2764         }
2765
2766         if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DIOREAD_NOLOCK)) {
2767                 int blocksize =
2768                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
2769                 if (blocksize < PAGE_SIZE)
2770                         ext4_msg(NULL, KERN_WARNING, "Warning: mounting with an "
2771                                  "experimental mount option 'dioread_nolock' "
2772                                  "for blocksize < PAGE_SIZE");
2773         }
2774
2775         err = ext4_check_test_dummy_encryption(fc, sb);
2776         if (err)
2777                 return err;
2778
2779         if ((ctx->spec & EXT4_SPEC_DATAJ) && is_remount) {
2780                 if (!sbi->s_journal) {
2781                         ext4_msg(NULL, KERN_WARNING,
2782                                  "Remounting file system with no journal "
2783                                  "so ignoring journalled data option");
2784                         ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS);
2785                 } else if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS) !=
2786                            test_opt(sb, DATA_FLAGS)) {
2787                         ext4_msg(NULL, KERN_ERR, "Cannot change data mode "
2788                                  "on remount");
2789                         return -EINVAL;
2790                 }
2791         }
2792
2793         if (is_remount) {
2794                 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) &&
2795                     (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) {
2796                         ext4_msg(NULL, KERN_ERR, "can't mount with "
2797                                  "both data=journal and dax");
2798                         return -EINVAL;
2799                 }
2800
2801                 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) &&
2802                     (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2803                      (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) {
2804 fail_dax_change_remount:
2805                         ext4_msg(NULL, KERN_ERR, "can't change "
2806                                  "dax mount option while remounting");
2807                         return -EINVAL;
2808                 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER) &&
2809                          (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2810                           (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS))) {
2811                         goto fail_dax_change_remount;
2812                 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE) &&
2813                            ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2814                             (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2815                             !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE))) {
2816                         goto fail_dax_change_remount;
2817                 }
2818         }
2819
2820         return ext4_check_quota_consistency(fc, sb);
2821 }
2822
2823 static void ext4_apply_options(struct fs_context *fc, struct super_block *sb)
2824 {
2825         struct ext4_fs_context *ctx = fc->fs_private;
2826         struct ext4_sb_info *sbi = fc->s_fs_info;
2827
2828         sbi->s_mount_opt &= ~ctx->mask_s_mount_opt;
2829         sbi->s_mount_opt |= ctx->vals_s_mount_opt;
2830         sbi->s_mount_opt2 &= ~ctx->mask_s_mount_opt2;
2831         sbi->s_mount_opt2 |= ctx->vals_s_mount_opt2;
2832         sbi->s_mount_flags &= ~ctx->mask_s_mount_flags;
2833         sbi->s_mount_flags |= ctx->vals_s_mount_flags;
2834         sb->s_flags &= ~ctx->mask_s_flags;
2835         sb->s_flags |= ctx->vals_s_flags;
2836
2837 #define APPLY(X) ({ if (ctx->spec & EXT4_SPEC_##X) sbi->X = ctx->X; })
2838         APPLY(s_commit_interval);
2839         APPLY(s_stripe);
2840         APPLY(s_max_batch_time);
2841         APPLY(s_min_batch_time);
2842         APPLY(s_want_extra_isize);
2843         APPLY(s_inode_readahead_blks);
2844         APPLY(s_max_dir_size_kb);
2845         APPLY(s_li_wait_mult);
2846         APPLY(s_resgid);
2847         APPLY(s_resuid);
2848
2849 #ifdef CONFIG_EXT4_DEBUG
2850         APPLY(s_fc_debug_max_replay);
2851 #endif
2852
2853         ext4_apply_quota_options(fc, sb);
2854         ext4_apply_test_dummy_encryption(ctx, sb);
2855 }
2856
2857
2858 static int ext4_validate_options(struct fs_context *fc)
2859 {
2860 #ifdef CONFIG_QUOTA
2861         struct ext4_fs_context *ctx = fc->fs_private;
2862         char *usr_qf_name, *grp_qf_name;
2863
2864         usr_qf_name = ctx->s_qf_names[USRQUOTA];
2865         grp_qf_name = ctx->s_qf_names[GRPQUOTA];
2866
2867         if (usr_qf_name || grp_qf_name) {
2868                 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) && usr_qf_name)
2869                         ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA);
2870
2871                 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) && grp_qf_name)
2872                         ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA);
2873
2874                 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) ||
2875                     ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA)) {
2876                         ext4_msg(NULL, KERN_ERR, "old and new quota "
2877                                  "format mixing");
2878                         return -EINVAL;
2879                 }
2880         }
2881 #endif
2882         return 1;
2883 }
2884
2885 static inline void ext4_show_quota_options(struct seq_file *seq,
2886                                            struct super_block *sb)
2887 {
2888 #if defined(CONFIG_QUOTA)
2889         struct ext4_sb_info *sbi = EXT4_SB(sb);
2890         char *usr_qf_name, *grp_qf_name;
2891
2892         if (sbi->s_jquota_fmt) {
2893                 char *fmtname = "";
2894
2895                 switch (sbi->s_jquota_fmt) {
2896                 case QFMT_VFS_OLD:
2897                         fmtname = "vfsold";
2898                         break;
2899                 case QFMT_VFS_V0:
2900                         fmtname = "vfsv0";
2901                         break;
2902                 case QFMT_VFS_V1:
2903                         fmtname = "vfsv1";
2904                         break;
2905                 }
2906                 seq_printf(seq, ",jqfmt=%s", fmtname);
2907         }
2908
2909         rcu_read_lock();
2910         usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]);
2911         grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]);
2912         if (usr_qf_name)
2913                 seq_show_option(seq, "usrjquota", usr_qf_name);
2914         if (grp_qf_name)
2915                 seq_show_option(seq, "grpjquota", grp_qf_name);
2916         rcu_read_unlock();
2917 #endif
2918 }
2919
2920 static const char *token2str(int token)
2921 {
2922         const struct fs_parameter_spec *spec;
2923
2924         for (spec = ext4_param_specs; spec->name != NULL; spec++)
2925                 if (spec->opt == token && !spec->type)
2926                         break;
2927         return spec->name;
2928 }
2929
2930 /*
2931  * Show an option if
2932  *  - it's set to a non-default value OR
2933  *  - if the per-sb default is different from the global default
2934  */
2935 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2936                               int nodefs)
2937 {
2938         struct ext4_sb_info *sbi = EXT4_SB(sb);
2939         struct ext4_super_block *es = sbi->s_es;
2940         int def_errors;
2941         const struct mount_opts *m;
2942         char sep = nodefs ? '\n' : ',';
2943
2944 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2945 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2946
2947         if (sbi->s_sb_block != 1)
2948                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2949
2950         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2951                 int want_set = m->flags & MOPT_SET;
2952                 int opt_2 = m->flags & MOPT_2;
2953                 unsigned int mount_opt, def_mount_opt;
2954
2955                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2956                     m->flags & MOPT_SKIP)
2957                         continue;
2958
2959                 if (opt_2) {
2960                         mount_opt = sbi->s_mount_opt2;
2961                         def_mount_opt = sbi->s_def_mount_opt2;
2962                 } else {
2963                         mount_opt = sbi->s_mount_opt;
2964                         def_mount_opt = sbi->s_def_mount_opt;
2965                 }
2966                 /* skip if same as the default */
2967                 if (!nodefs && !(m->mount_opt & (mount_opt ^ def_mount_opt)))
2968                         continue;
2969                 /* select Opt_noFoo vs Opt_Foo */
2970                 if ((want_set &&
2971                      (mount_opt & m->mount_opt) != m->mount_opt) ||
2972                     (!want_set && (mount_opt & m->mount_opt)))
2973                         continue;
2974                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2975         }
2976
2977         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2978             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2979                 SEQ_OPTS_PRINT("resuid=%u",
2980                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2981         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2982             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2983                 SEQ_OPTS_PRINT("resgid=%u",
2984                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2985         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2986         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2987                 SEQ_OPTS_PUTS("errors=remount-ro");
2988         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2989                 SEQ_OPTS_PUTS("errors=continue");
2990         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2991                 SEQ_OPTS_PUTS("errors=panic");
2992         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2993                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2994         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2995                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2996         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2997                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2998         if (nodefs || sbi->s_stripe)
2999                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
3000         if (nodefs || EXT4_MOUNT_DATA_FLAGS &
3001                         (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
3002                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3003                         SEQ_OPTS_PUTS("data=journal");
3004                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3005                         SEQ_OPTS_PUTS("data=ordered");
3006                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
3007                         SEQ_OPTS_PUTS("data=writeback");
3008         }
3009         if (nodefs ||
3010             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
3011                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
3012                                sbi->s_inode_readahead_blks);
3013
3014         if (test_opt(sb, INIT_INODE_TABLE) && (nodefs ||
3015                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
3016                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
3017         if (nodefs || sbi->s_max_dir_size_kb)
3018                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
3019         if (test_opt(sb, DATA_ERR_ABORT))
3020                 SEQ_OPTS_PUTS("data_err=abort");
3021
3022         fscrypt_show_test_dummy_encryption(seq, sep, sb);
3023
3024         if (sb->s_flags & SB_INLINECRYPT)
3025                 SEQ_OPTS_PUTS("inlinecrypt");
3026
3027         if (test_opt(sb, DAX_ALWAYS)) {
3028                 if (IS_EXT2_SB(sb))
3029                         SEQ_OPTS_PUTS("dax");
3030                 else
3031                         SEQ_OPTS_PUTS("dax=always");
3032         } else if (test_opt2(sb, DAX_NEVER)) {
3033                 SEQ_OPTS_PUTS("dax=never");
3034         } else if (test_opt2(sb, DAX_INODE)) {
3035                 SEQ_OPTS_PUTS("dax=inode");
3036         }
3037
3038         if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD &&
3039                         !test_opt2(sb, MB_OPTIMIZE_SCAN)) {
3040                 SEQ_OPTS_PUTS("mb_optimize_scan=0");
3041         } else if (sbi->s_groups_count < MB_DEFAULT_LINEAR_SCAN_THRESHOLD &&
3042                         test_opt2(sb, MB_OPTIMIZE_SCAN)) {
3043                 SEQ_OPTS_PUTS("mb_optimize_scan=1");
3044         }
3045
3046         ext4_show_quota_options(seq, sb);
3047         return 0;
3048 }
3049
3050 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
3051 {
3052         return _ext4_show_options(seq, root->d_sb, 0);
3053 }
3054
3055 int ext4_seq_options_show(struct seq_file *seq, void *offset)
3056 {
3057         struct super_block *sb = seq->private;
3058         int rc;
3059
3060         seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
3061         rc = _ext4_show_options(seq, sb, 1);
3062         seq_puts(seq, "\n");
3063         return rc;
3064 }
3065
3066 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
3067                             int read_only)
3068 {
3069         struct ext4_sb_info *sbi = EXT4_SB(sb);
3070         int err = 0;
3071
3072         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
3073                 ext4_msg(sb, KERN_ERR, "revision level too high, "
3074                          "forcing read-only mode");
3075                 err = -EROFS;
3076                 goto done;
3077         }
3078         if (read_only)
3079                 goto done;
3080         if (!(sbi->s_mount_state & EXT4_VALID_FS))
3081                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
3082                          "running e2fsck is recommended");
3083         else if (sbi->s_mount_state & EXT4_ERROR_FS)
3084                 ext4_msg(sb, KERN_WARNING,
3085                          "warning: mounting fs with errors, "
3086                          "running e2fsck is recommended");
3087         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
3088                  le16_to_cpu(es->s_mnt_count) >=
3089                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
3090                 ext4_msg(sb, KERN_WARNING,
3091                          "warning: maximal mount count reached, "
3092                          "running e2fsck is recommended");
3093         else if (le32_to_cpu(es->s_checkinterval) &&
3094                  (ext4_get_tstamp(es, s_lastcheck) +
3095                   le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds()))
3096                 ext4_msg(sb, KERN_WARNING,
3097                          "warning: checktime reached, "
3098                          "running e2fsck is recommended");
3099         if (!sbi->s_journal)
3100                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
3101         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
3102                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
3103         le16_add_cpu(&es->s_mnt_count, 1);
3104         ext4_update_tstamp(es, s_mtime);
3105         if (sbi->s_journal) {
3106                 ext4_set_feature_journal_needs_recovery(sb);
3107                 if (ext4_has_feature_orphan_file(sb))
3108                         ext4_set_feature_orphan_present(sb);
3109         }
3110
3111         err = ext4_commit_super(sb);
3112 done:
3113         if (test_opt(sb, DEBUG))
3114                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
3115                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
3116                         sb->s_blocksize,
3117                         sbi->s_groups_count,
3118                         EXT4_BLOCKS_PER_GROUP(sb),
3119                         EXT4_INODES_PER_GROUP(sb),
3120                         sbi->s_mount_opt, sbi->s_mount_opt2);
3121         return err;
3122 }
3123
3124 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
3125 {
3126         struct ext4_sb_info *sbi = EXT4_SB(sb);
3127         struct flex_groups **old_groups, **new_groups;
3128         int size, i, j;
3129
3130         if (!sbi->s_log_groups_per_flex)
3131                 return 0;
3132
3133         size = ext4_flex_group(sbi, ngroup - 1) + 1;
3134         if (size <= sbi->s_flex_groups_allocated)
3135                 return 0;
3136
3137         new_groups = kvzalloc(roundup_pow_of_two(size *
3138                               sizeof(*sbi->s_flex_groups)), GFP_KERNEL);
3139         if (!new_groups) {
3140                 ext4_msg(sb, KERN_ERR,
3141                          "not enough memory for %d flex group pointers", size);
3142                 return -ENOMEM;
3143         }
3144         for (i = sbi->s_flex_groups_allocated; i < size; i++) {
3145                 new_groups[i] = kvzalloc(roundup_pow_of_two(
3146                                          sizeof(struct flex_groups)),
3147                                          GFP_KERNEL);
3148                 if (!new_groups[i]) {
3149                         for (j = sbi->s_flex_groups_allocated; j < i; j++)
3150                                 kvfree(new_groups[j]);
3151                         kvfree(new_groups);
3152                         ext4_msg(sb, KERN_ERR,
3153                                  "not enough memory for %d flex groups", size);
3154                         return -ENOMEM;
3155                 }
3156         }
3157         rcu_read_lock();
3158         old_groups = rcu_dereference(sbi->s_flex_groups);
3159         if (old_groups)
3160                 memcpy(new_groups, old_groups,
3161                        (sbi->s_flex_groups_allocated *
3162                         sizeof(struct flex_groups *)));
3163         rcu_read_unlock();
3164         rcu_assign_pointer(sbi->s_flex_groups, new_groups);
3165         sbi->s_flex_groups_allocated = size;
3166         if (old_groups)
3167                 ext4_kvfree_array_rcu(old_groups);
3168         return 0;
3169 }
3170
3171 static int ext4_fill_flex_info(struct super_block *sb)
3172 {
3173         struct ext4_sb_info *sbi = EXT4_SB(sb);
3174         struct ext4_group_desc *gdp = NULL;
3175         struct flex_groups *fg;
3176         ext4_group_t flex_group;
3177         int i, err;
3178
3179         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
3180         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
3181                 sbi->s_log_groups_per_flex = 0;
3182                 return 1;
3183         }
3184
3185         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
3186         if (err)
3187                 goto failed;
3188
3189         for (i = 0; i < sbi->s_groups_count; i++) {
3190                 gdp = ext4_get_group_desc(sb, i, NULL);
3191
3192                 flex_group = ext4_flex_group(sbi, i);
3193                 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
3194                 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes);
3195                 atomic64_add(ext4_free_group_clusters(sb, gdp),
3196                              &fg->free_clusters);
3197                 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs);
3198         }
3199
3200         return 1;
3201 failed:
3202         return 0;
3203 }
3204
3205 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
3206                                    struct ext4_group_desc *gdp)
3207 {
3208         int offset = offsetof(struct ext4_group_desc, bg_checksum);
3209         __u16 crc = 0;
3210         __le32 le_group = cpu_to_le32(block_group);
3211         struct ext4_sb_info *sbi = EXT4_SB(sb);
3212
3213         if (ext4_has_metadata_csum(sbi->s_sb)) {
3214                 /* Use new metadata_csum algorithm */
3215                 __u32 csum32;
3216                 __u16 dummy_csum = 0;
3217
3218                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
3219                                      sizeof(le_group));
3220                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
3221                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
3222                                      sizeof(dummy_csum));
3223                 offset += sizeof(dummy_csum);
3224                 if (offset < sbi->s_desc_size)
3225                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
3226                                              sbi->s_desc_size - offset);
3227
3228                 crc = csum32 & 0xFFFF;
3229                 goto out;
3230         }
3231
3232         /* old crc16 code */
3233         if (!ext4_has_feature_gdt_csum(sb))
3234                 return 0;
3235
3236         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
3237         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
3238         crc = crc16(crc, (__u8 *)gdp, offset);
3239         offset += sizeof(gdp->bg_checksum); /* skip checksum */
3240         /* for checksum of struct ext4_group_desc do the rest...*/
3241         if (ext4_has_feature_64bit(sb) &&
3242             offset < le16_to_cpu(sbi->s_es->s_desc_size))
3243                 crc = crc16(crc, (__u8 *)gdp + offset,
3244                             le16_to_cpu(sbi->s_es->s_desc_size) -
3245                                 offset);
3246
3247 out:
3248         return cpu_to_le16(crc);
3249 }
3250
3251 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
3252                                 struct ext4_group_desc *gdp)
3253 {
3254         if (ext4_has_group_desc_csum(sb) &&
3255             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
3256                 return 0;
3257
3258         return 1;
3259 }
3260
3261 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
3262                               struct ext4_group_desc *gdp)
3263 {
3264         if (!ext4_has_group_desc_csum(sb))
3265                 return;
3266         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
3267 }
3268
3269 /* Called at mount-time, super-block is locked */
3270 static int ext4_check_descriptors(struct super_block *sb,
3271                                   ext4_fsblk_t sb_block,
3272                                   ext4_group_t *first_not_zeroed)
3273 {
3274         struct ext4_sb_info *sbi = EXT4_SB(sb);
3275         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
3276         ext4_fsblk_t last_block;
3277         ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0);
3278         ext4_fsblk_t block_bitmap;
3279         ext4_fsblk_t inode_bitmap;
3280         ext4_fsblk_t inode_table;
3281         int flexbg_flag = 0;
3282         ext4_group_t i, grp = sbi->s_groups_count;
3283
3284         if (ext4_has_feature_flex_bg(sb))
3285                 flexbg_flag = 1;
3286
3287         ext4_debug("Checking group descriptors");
3288
3289         for (i = 0; i < sbi->s_groups_count; i++) {
3290                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
3291
3292                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
3293                         last_block = ext4_blocks_count(sbi->s_es) - 1;
3294                 else
3295                         last_block = first_block +
3296                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
3297
3298                 if ((grp == sbi->s_groups_count) &&
3299                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3300                         grp = i;
3301
3302                 block_bitmap = ext4_block_bitmap(sb, gdp);
3303                 if (block_bitmap == sb_block) {
3304                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3305                                  "Block bitmap for group %u overlaps "
3306                                  "superblock", i);
3307                         if (!sb_rdonly(sb))
3308                                 return 0;
3309                 }
3310                 if (block_bitmap >= sb_block + 1 &&
3311                     block_bitmap <= last_bg_block) {
3312                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3313                                  "Block bitmap for group %u overlaps "
3314                                  "block group descriptors", i);
3315                         if (!sb_rdonly(sb))
3316                                 return 0;
3317                 }
3318                 if (block_bitmap < first_block || block_bitmap > last_block) {
3319                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3320                                "Block bitmap for group %u not in group "
3321                                "(block %llu)!", i, block_bitmap);
3322                         return 0;
3323                 }
3324                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
3325                 if (inode_bitmap == sb_block) {
3326                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3327                                  "Inode bitmap for group %u overlaps "
3328                                  "superblock", i);
3329                         if (!sb_rdonly(sb))
3330                                 return 0;
3331                 }
3332                 if (inode_bitmap >= sb_block + 1 &&
3333                     inode_bitmap <= last_bg_block) {
3334                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3335                                  "Inode bitmap for group %u overlaps "
3336                                  "block group descriptors", i);
3337                         if (!sb_rdonly(sb))
3338                                 return 0;
3339                 }
3340                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
3341                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3342                                "Inode bitmap for group %u not in group "
3343                                "(block %llu)!", i, inode_bitmap);
3344                         return 0;
3345                 }
3346                 inode_table = ext4_inode_table(sb, gdp);
3347                 if (inode_table == sb_block) {
3348                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3349                                  "Inode table for group %u overlaps "
3350                                  "superblock", i);
3351                         if (!sb_rdonly(sb))
3352                                 return 0;
3353                 }
3354                 if (inode_table >= sb_block + 1 &&
3355                     inode_table <= last_bg_block) {
3356                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3357                                  "Inode table for group %u overlaps "
3358                                  "block group descriptors", i);
3359                         if (!sb_rdonly(sb))
3360                                 return 0;
3361                 }
3362                 if (inode_table < first_block ||
3363                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
3364                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3365                                "Inode table for group %u not in group "
3366                                "(block %llu)!", i, inode_table);
3367                         return 0;
3368                 }
3369                 ext4_lock_group(sb, i);
3370                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
3371                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3372                                  "Checksum for group %u failed (%u!=%u)",
3373                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
3374                                      gdp)), le16_to_cpu(gdp->bg_checksum));
3375                         if (!sb_rdonly(sb)) {
3376                                 ext4_unlock_group(sb, i);
3377                                 return 0;
3378                         }
3379                 }
3380                 ext4_unlock_group(sb, i);
3381                 if (!flexbg_flag)
3382                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
3383         }
3384         if (NULL != first_not_zeroed)
3385                 *first_not_zeroed = grp;
3386         return 1;
3387 }
3388
3389 /*
3390  * Maximal extent format file size.
3391  * Resulting logical blkno at s_maxbytes must fit in our on-disk
3392  * extent format containers, within a sector_t, and within i_blocks
3393  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
3394  * so that won't be a limiting factor.
3395  *
3396  * However there is other limiting factor. We do store extents in the form
3397  * of starting block and length, hence the resulting length of the extent
3398  * covering maximum file size must fit into on-disk format containers as
3399  * well. Given that length is always by 1 unit bigger than max unit (because
3400  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
3401  *
3402  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
3403  */
3404 static loff_t ext4_max_size(int blkbits, int has_huge_files)
3405 {
3406         loff_t res;
3407         loff_t upper_limit = MAX_LFS_FILESIZE;
3408
3409         BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64));
3410
3411         if (!has_huge_files) {
3412                 upper_limit = (1LL << 32) - 1;
3413
3414                 /* total blocks in file system block size */
3415                 upper_limit >>= (blkbits - 9);
3416                 upper_limit <<= blkbits;
3417         }
3418
3419         /*
3420          * 32-bit extent-start container, ee_block. We lower the maxbytes
3421          * by one fs block, so ee_len can cover the extent of maximum file
3422          * size
3423          */
3424         res = (1LL << 32) - 1;
3425         res <<= blkbits;
3426
3427         /* Sanity check against vm- & vfs- imposed limits */
3428         if (res > upper_limit)
3429                 res = upper_limit;
3430
3431         return res;
3432 }
3433
3434 /*
3435  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
3436  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
3437  * We need to be 1 filesystem block less than the 2^48 sector limit.
3438  */
3439 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
3440 {
3441         loff_t upper_limit, res = EXT4_NDIR_BLOCKS;
3442         int meta_blocks;
3443         unsigned int ppb = 1 << (bits - 2);
3444
3445         /*
3446          * This is calculated to be the largest file size for a dense, block
3447          * mapped file such that the file's total number of 512-byte sectors,
3448          * including data and all indirect blocks, does not exceed (2^48 - 1).
3449          *
3450          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
3451          * number of 512-byte sectors of the file.
3452          */
3453         if (!has_huge_files) {
3454                 /*
3455                  * !has_huge_files or implies that the inode i_block field
3456                  * represents total file blocks in 2^32 512-byte sectors ==
3457                  * size of vfs inode i_blocks * 8
3458                  */
3459                 upper_limit = (1LL << 32) - 1;
3460
3461                 /* total blocks in file system block size */
3462                 upper_limit >>= (bits - 9);
3463
3464         } else {
3465                 /*
3466                  * We use 48 bit ext4_inode i_blocks
3467                  * With EXT4_HUGE_FILE_FL set the i_blocks
3468                  * represent total number of blocks in
3469                  * file system block size
3470                  */
3471                 upper_limit = (1LL << 48) - 1;
3472
3473         }
3474
3475         /* Compute how many blocks we can address by block tree */
3476         res += ppb;
3477         res += ppb * ppb;
3478         res += ((loff_t)ppb) * ppb * ppb;
3479         /* Compute how many metadata blocks are needed */
3480         meta_blocks = 1;
3481         meta_blocks += 1 + ppb;
3482         meta_blocks += 1 + ppb + ppb * ppb;
3483         /* Does block tree limit file size? */
3484         if (res + meta_blocks <= upper_limit)
3485                 goto check_lfs;
3486
3487         res = upper_limit;
3488         /* How many metadata blocks are needed for addressing upper_limit? */
3489         upper_limit -= EXT4_NDIR_BLOCKS;
3490         /* indirect blocks */
3491         meta_blocks = 1;
3492         upper_limit -= ppb;
3493         /* double indirect blocks */
3494         if (upper_limit < ppb * ppb) {
3495                 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb);
3496                 res -= meta_blocks;
3497                 goto check_lfs;
3498         }
3499         meta_blocks += 1 + ppb;
3500         upper_limit -= ppb * ppb;
3501         /* tripple indirect blocks for the rest */
3502         meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb) +
3503                 DIV_ROUND_UP_ULL(upper_limit, ppb*ppb);
3504         res -= meta_blocks;
3505 check_lfs:
3506         res <<= bits;
3507         if (res > MAX_LFS_FILESIZE)
3508                 res = MAX_LFS_FILESIZE;
3509
3510         return res;
3511 }
3512
3513 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
3514                                    ext4_fsblk_t logical_sb_block, int nr)
3515 {
3516         struct ext4_sb_info *sbi = EXT4_SB(sb);
3517         ext4_group_t bg, first_meta_bg;
3518         int has_super = 0;
3519
3520         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
3521
3522         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
3523                 return logical_sb_block + nr + 1;
3524         bg = sbi->s_desc_per_block * nr;
3525         if (ext4_bg_has_super(sb, bg))
3526                 has_super = 1;
3527
3528         /*
3529          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
3530          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
3531          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
3532          * compensate.
3533          */
3534         if (sb->s_blocksize == 1024 && nr == 0 &&
3535             le32_to_cpu(sbi->s_es->s_first_data_block) == 0)
3536                 has_super++;
3537
3538         return (has_super + ext4_group_first_block_no(sb, bg));
3539 }
3540
3541 /**
3542  * ext4_get_stripe_size: Get the stripe size.
3543  * @sbi: In memory super block info
3544  *
3545  * If we have specified it via mount option, then
3546  * use the mount option value. If the value specified at mount time is
3547  * greater than the blocks per group use the super block value.
3548  * If the super block value is greater than blocks per group return 0.
3549  * Allocator needs it be less than blocks per group.
3550  *
3551  */
3552 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
3553 {
3554         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
3555         unsigned long stripe_width =
3556                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
3557         int ret;
3558
3559         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
3560                 ret = sbi->s_stripe;
3561         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
3562                 ret = stripe_width;
3563         else if (stride && stride <= sbi->s_blocks_per_group)
3564                 ret = stride;
3565         else
3566                 ret = 0;
3567
3568         /*
3569          * If the stripe width is 1, this makes no sense and
3570          * we set it to 0 to turn off stripe handling code.
3571          */
3572         if (ret <= 1)
3573                 ret = 0;
3574
3575         return ret;
3576 }
3577
3578 /*
3579  * Check whether this filesystem can be mounted based on
3580  * the features present and the RDONLY/RDWR mount requested.
3581  * Returns 1 if this filesystem can be mounted as requested,
3582  * 0 if it cannot be.
3583  */
3584 int ext4_feature_set_ok(struct super_block *sb, int readonly)
3585 {
3586         if (ext4_has_unknown_ext4_incompat_features(sb)) {
3587                 ext4_msg(sb, KERN_ERR,
3588                         "Couldn't mount because of "
3589                         "unsupported optional features (%x)",
3590                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
3591                         ~EXT4_FEATURE_INCOMPAT_SUPP));
3592                 return 0;
3593         }
3594
3595 #if !IS_ENABLED(CONFIG_UNICODE)
3596         if (ext4_has_feature_casefold(sb)) {
3597                 ext4_msg(sb, KERN_ERR,
3598                          "Filesystem with casefold feature cannot be "
3599                          "mounted without CONFIG_UNICODE");
3600                 return 0;
3601         }
3602 #endif
3603
3604         if (readonly)
3605                 return 1;
3606
3607         if (ext4_has_feature_readonly(sb)) {
3608                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
3609                 sb->s_flags |= SB_RDONLY;
3610                 return 1;
3611         }
3612
3613         /* Check that feature set is OK for a read-write mount */
3614         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
3615                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
3616                          "unsupported optional features (%x)",
3617                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
3618                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
3619                 return 0;
3620         }
3621         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
3622                 ext4_msg(sb, KERN_ERR,
3623                          "Can't support bigalloc feature without "
3624                          "extents feature\n");
3625                 return 0;
3626         }
3627
3628 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2)
3629         if (!readonly && (ext4_has_feature_quota(sb) ||
3630                           ext4_has_feature_project(sb))) {
3631                 ext4_msg(sb, KERN_ERR,
3632                          "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2");
3633                 return 0;
3634         }
3635 #endif  /* CONFIG_QUOTA */
3636         return 1;
3637 }
3638
3639 /*
3640  * This function is called once a day if we have errors logged
3641  * on the file system
3642  */
3643 static void print_daily_error_info(struct timer_list *t)
3644 {
3645         struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report);
3646         struct super_block *sb = sbi->s_sb;
3647         struct ext4_super_block *es = sbi->s_es;
3648
3649         if (es->s_error_count)
3650                 /* fsck newer than v1.41.13 is needed to clean this condition. */
3651                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
3652                          le32_to_cpu(es->s_error_count));
3653         if (es->s_first_error_time) {
3654                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d",
3655                        sb->s_id,
3656                        ext4_get_tstamp(es, s_first_error_time),
3657                        (int) sizeof(es->s_first_error_func),
3658                        es->s_first_error_func,
3659                        le32_to_cpu(es->s_first_error_line));
3660                 if (es->s_first_error_ino)
3661                         printk(KERN_CONT ": inode %u",
3662                                le32_to_cpu(es->s_first_error_ino));
3663                 if (es->s_first_error_block)
3664                         printk(KERN_CONT ": block %llu", (unsigned long long)
3665                                le64_to_cpu(es->s_first_error_block));
3666                 printk(KERN_CONT "\n");
3667         }
3668         if (es->s_last_error_time) {
3669                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d",
3670                        sb->s_id,
3671                        ext4_get_tstamp(es, s_last_error_time),
3672                        (int) sizeof(es->s_last_error_func),
3673                        es->s_last_error_func,
3674                        le32_to_cpu(es->s_last_error_line));
3675                 if (es->s_last_error_ino)
3676                         printk(KERN_CONT ": inode %u",
3677                                le32_to_cpu(es->s_last_error_ino));
3678                 if (es->s_last_error_block)
3679                         printk(KERN_CONT ": block %llu", (unsigned long long)
3680                                le64_to_cpu(es->s_last_error_block));
3681                 printk(KERN_CONT "\n");
3682         }
3683         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
3684 }
3685
3686 /* Find next suitable group and run ext4_init_inode_table */
3687 static int ext4_run_li_request(struct ext4_li_request *elr)
3688 {
3689         struct ext4_group_desc *gdp = NULL;
3690         struct super_block *sb = elr->lr_super;
3691         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3692         ext4_group_t group = elr->lr_next_group;
3693         unsigned int prefetch_ios = 0;
3694         int ret = 0;
3695         u64 start_time;
3696
3697         if (elr->lr_mode == EXT4_LI_MODE_PREFETCH_BBITMAP) {
3698                 elr->lr_next_group = ext4_mb_prefetch(sb, group,
3699                                 EXT4_SB(sb)->s_mb_prefetch, &prefetch_ios);
3700                 if (prefetch_ios)
3701                         ext4_mb_prefetch_fini(sb, elr->lr_next_group,
3702                                               prefetch_ios);
3703                 trace_ext4_prefetch_bitmaps(sb, group, elr->lr_next_group,
3704                                             prefetch_ios);
3705                 if (group >= elr->lr_next_group) {
3706                         ret = 1;
3707                         if (elr->lr_first_not_zeroed != ngroups &&
3708                             !sb_rdonly(sb) && test_opt(sb, INIT_INODE_TABLE)) {
3709                                 elr->lr_next_group = elr->lr_first_not_zeroed;
3710                                 elr->lr_mode = EXT4_LI_MODE_ITABLE;
3711                                 ret = 0;
3712                         }
3713                 }
3714                 return ret;
3715         }
3716
3717         for (; group < ngroups; group++) {
3718                 gdp = ext4_get_group_desc(sb, group, NULL);
3719                 if (!gdp) {
3720                         ret = 1;
3721                         break;
3722                 }
3723
3724                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3725                         break;
3726         }
3727
3728         if (group >= ngroups)
3729                 ret = 1;
3730
3731         if (!ret) {
3732                 start_time = ktime_get_real_ns();
3733                 ret = ext4_init_inode_table(sb, group,
3734                                             elr->lr_timeout ? 0 : 1);
3735                 trace_ext4_lazy_itable_init(sb, group);
3736                 if (elr->lr_timeout == 0) {
3737                         elr->lr_timeout = nsecs_to_jiffies((ktime_get_real_ns() - start_time) *
3738                                 EXT4_SB(elr->lr_super)->s_li_wait_mult);
3739                 }
3740                 elr->lr_next_sched = jiffies + elr->lr_timeout;
3741                 elr->lr_next_group = group + 1;
3742         }
3743         return ret;
3744 }
3745
3746 /*
3747  * Remove lr_request from the list_request and free the
3748  * request structure. Should be called with li_list_mtx held
3749  */
3750 static void ext4_remove_li_request(struct ext4_li_request *elr)
3751 {
3752         if (!elr)
3753                 return;
3754
3755         list_del(&elr->lr_request);
3756         EXT4_SB(elr->lr_super)->s_li_request = NULL;
3757         kfree(elr);
3758 }
3759
3760 static void ext4_unregister_li_request(struct super_block *sb)
3761 {
3762         mutex_lock(&ext4_li_mtx);
3763         if (!ext4_li_info) {
3764                 mutex_unlock(&ext4_li_mtx);
3765                 return;
3766         }
3767
3768         mutex_lock(&ext4_li_info->li_list_mtx);
3769         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
3770         mutex_unlock(&ext4_li_info->li_list_mtx);
3771         mutex_unlock(&ext4_li_mtx);
3772 }
3773
3774 static struct task_struct *ext4_lazyinit_task;
3775
3776 /*
3777  * This is the function where ext4lazyinit thread lives. It walks
3778  * through the request list searching for next scheduled filesystem.
3779  * When such a fs is found, run the lazy initialization request
3780  * (ext4_rn_li_request) and keep track of the time spend in this
3781  * function. Based on that time we compute next schedule time of
3782  * the request. When walking through the list is complete, compute
3783  * next waking time and put itself into sleep.
3784  */
3785 static int ext4_lazyinit_thread(void *arg)
3786 {
3787         struct ext4_lazy_init *eli = arg;
3788         struct list_head *pos, *n;
3789         struct ext4_li_request *elr;
3790         unsigned long next_wakeup, cur;
3791
3792         BUG_ON(NULL == eli);
3793         set_freezable();
3794
3795 cont_thread:
3796         while (true) {
3797                 next_wakeup = MAX_JIFFY_OFFSET;
3798
3799                 mutex_lock(&eli->li_list_mtx);
3800                 if (list_empty(&eli->li_request_list)) {
3801                         mutex_unlock(&eli->li_list_mtx);
3802                         goto exit_thread;
3803                 }
3804                 list_for_each_safe(pos, n, &eli->li_request_list) {
3805                         int err = 0;
3806                         int progress = 0;
3807                         elr = list_entry(pos, struct ext4_li_request,
3808                                          lr_request);
3809
3810                         if (time_before(jiffies, elr->lr_next_sched)) {
3811                                 if (time_before(elr->lr_next_sched, next_wakeup))
3812                                         next_wakeup = elr->lr_next_sched;
3813                                 continue;
3814                         }
3815                         if (down_read_trylock(&elr->lr_super->s_umount)) {
3816                                 if (sb_start_write_trylock(elr->lr_super)) {
3817                                         progress = 1;
3818                                         /*
3819                                          * We hold sb->s_umount, sb can not
3820                                          * be removed from the list, it is
3821                                          * now safe to drop li_list_mtx
3822                                          */
3823                                         mutex_unlock(&eli->li_list_mtx);
3824                                         err = ext4_run_li_request(elr);
3825                                         sb_end_write(elr->lr_super);
3826                                         mutex_lock(&eli->li_list_mtx);
3827                                         n = pos->next;
3828                                 }
3829                                 up_read((&elr->lr_super->s_umount));
3830                         }
3831                         /* error, remove the lazy_init job */
3832                         if (err) {
3833                                 ext4_remove_li_request(elr);
3834                                 continue;
3835                         }
3836                         if (!progress) {
3837                                 elr->lr_next_sched = jiffies +
3838                                         get_random_u32_below(EXT4_DEF_LI_MAX_START_DELAY * HZ);
3839                         }
3840                         if (time_before(elr->lr_next_sched, next_wakeup))
3841                                 next_wakeup = elr->lr_next_sched;
3842                 }
3843                 mutex_unlock(&eli->li_list_mtx);
3844
3845                 try_to_freeze();
3846
3847                 cur = jiffies;
3848                 if ((time_after_eq(cur, next_wakeup)) ||
3849                     (MAX_JIFFY_OFFSET == next_wakeup)) {
3850                         cond_resched();
3851                         continue;
3852                 }
3853
3854                 schedule_timeout_interruptible(next_wakeup - cur);
3855
3856                 if (kthread_should_stop()) {
3857                         ext4_clear_request_list();
3858                         goto exit_thread;
3859                 }
3860         }
3861
3862 exit_thread:
3863         /*
3864          * It looks like the request list is empty, but we need
3865          * to check it under the li_list_mtx lock, to prevent any
3866          * additions into it, and of course we should lock ext4_li_mtx
3867          * to atomically free the list and ext4_li_info, because at
3868          * this point another ext4 filesystem could be registering
3869          * new one.
3870          */
3871         mutex_lock(&ext4_li_mtx);
3872         mutex_lock(&eli->li_list_mtx);
3873         if (!list_empty(&eli->li_request_list)) {
3874                 mutex_unlock(&eli->li_list_mtx);
3875                 mutex_unlock(&ext4_li_mtx);
3876                 goto cont_thread;
3877         }
3878         mutex_unlock(&eli->li_list_mtx);
3879         kfree(ext4_li_info);
3880         ext4_li_info = NULL;
3881         mutex_unlock(&ext4_li_mtx);
3882
3883         return 0;
3884 }
3885
3886 static void ext4_clear_request_list(void)
3887 {
3888         struct list_head *pos, *n;
3889         struct ext4_li_request *elr;
3890
3891         mutex_lock(&ext4_li_info->li_list_mtx);
3892         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3893                 elr = list_entry(pos, struct ext4_li_request,
3894                                  lr_request);
3895                 ext4_remove_li_request(elr);
3896         }
3897         mutex_unlock(&ext4_li_info->li_list_mtx);
3898 }
3899
3900 static int ext4_run_lazyinit_thread(void)
3901 {
3902         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3903                                          ext4_li_info, "ext4lazyinit");
3904         if (IS_ERR(ext4_lazyinit_task)) {
3905                 int err = PTR_ERR(ext4_lazyinit_task);
3906                 ext4_clear_request_list();
3907                 kfree(ext4_li_info);
3908                 ext4_li_info = NULL;
3909                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3910                                  "initialization thread\n",
3911                                  err);
3912                 return err;
3913         }
3914         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3915         return 0;
3916 }
3917
3918 /*
3919  * Check whether it make sense to run itable init. thread or not.
3920  * If there is at least one uninitialized inode table, return
3921  * corresponding group number, else the loop goes through all
3922  * groups and return total number of groups.
3923  */
3924 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3925 {
3926         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3927         struct ext4_group_desc *gdp = NULL;
3928
3929         if (!ext4_has_group_desc_csum(sb))
3930                 return ngroups;
3931
3932         for (group = 0; group < ngroups; group++) {
3933                 gdp = ext4_get_group_desc(sb, group, NULL);
3934                 if (!gdp)
3935                         continue;
3936
3937                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3938                         break;
3939         }
3940
3941         return group;
3942 }
3943
3944 static int ext4_li_info_new(void)
3945 {
3946         struct ext4_lazy_init *eli = NULL;
3947
3948         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3949         if (!eli)
3950                 return -ENOMEM;
3951
3952         INIT_LIST_HEAD(&eli->li_request_list);
3953         mutex_init(&eli->li_list_mtx);
3954
3955         eli->li_state |= EXT4_LAZYINIT_QUIT;
3956
3957         ext4_li_info = eli;
3958
3959         return 0;
3960 }
3961
3962 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3963                                             ext4_group_t start)
3964 {
3965         struct ext4_li_request *elr;
3966
3967         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3968         if (!elr)
3969                 return NULL;
3970
3971         elr->lr_super = sb;
3972         elr->lr_first_not_zeroed = start;
3973         if (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS)) {
3974                 elr->lr_mode = EXT4_LI_MODE_ITABLE;
3975                 elr->lr_next_group = start;
3976         } else {
3977                 elr->lr_mode = EXT4_LI_MODE_PREFETCH_BBITMAP;
3978         }
3979
3980         /*
3981          * Randomize first schedule time of the request to
3982          * spread the inode table initialization requests
3983          * better.
3984          */
3985         elr->lr_next_sched = jiffies + get_random_u32_below(EXT4_DEF_LI_MAX_START_DELAY * HZ);
3986         return elr;
3987 }
3988
3989 int ext4_register_li_request(struct super_block *sb,
3990                              ext4_group_t first_not_zeroed)
3991 {
3992         struct ext4_sb_info *sbi = EXT4_SB(sb);
3993         struct ext4_li_request *elr = NULL;
3994         ext4_group_t ngroups = sbi->s_groups_count;
3995         int ret = 0;
3996
3997         mutex_lock(&ext4_li_mtx);
3998         if (sbi->s_li_request != NULL) {
3999                 /*
4000                  * Reset timeout so it can be computed again, because
4001                  * s_li_wait_mult might have changed.
4002                  */
4003                 sbi->s_li_request->lr_timeout = 0;
4004                 goto out;
4005         }
4006
4007         if (sb_rdonly(sb) ||
4008             (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS) &&
4009              (first_not_zeroed == ngroups || !test_opt(sb, INIT_INODE_TABLE))))
4010                 goto out;
4011
4012         elr = ext4_li_request_new(sb, first_not_zeroed);
4013         if (!elr) {
4014                 ret = -ENOMEM;
4015                 goto out;
4016         }
4017
4018         if (NULL == ext4_li_info) {
4019                 ret = ext4_li_info_new();
4020                 if (ret)
4021                         goto out;
4022         }
4023
4024         mutex_lock(&ext4_li_info->li_list_mtx);
4025         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
4026         mutex_unlock(&ext4_li_info->li_list_mtx);
4027
4028         sbi->s_li_request = elr;
4029         /*
4030          * set elr to NULL here since it has been inserted to
4031          * the request_list and the removal and free of it is
4032          * handled by ext4_clear_request_list from now on.
4033          */
4034         elr = NULL;
4035
4036         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
4037                 ret = ext4_run_lazyinit_thread();
4038                 if (ret)
4039                         goto out;
4040         }
4041 out:
4042         mutex_unlock(&ext4_li_mtx);
4043         if (ret)
4044                 kfree(elr);
4045         return ret;
4046 }
4047
4048 /*
4049  * We do not need to lock anything since this is called on
4050  * module unload.
4051  */
4052 static void ext4_destroy_lazyinit_thread(void)
4053 {
4054         /*
4055          * If thread exited earlier
4056          * there's nothing to be done.
4057          */
4058         if (!ext4_li_info || !ext4_lazyinit_task)
4059                 return;
4060
4061         kthread_stop(ext4_lazyinit_task);
4062 }
4063
4064 static int set_journal_csum_feature_set(struct super_block *sb)
4065 {
4066         int ret = 1;
4067         int compat, incompat;
4068         struct ext4_sb_info *sbi = EXT4_SB(sb);
4069
4070         if (ext4_has_metadata_csum(sb)) {
4071                 /* journal checksum v3 */
4072                 compat = 0;
4073                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
4074         } else {
4075                 /* journal checksum v1 */
4076                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
4077                 incompat = 0;
4078         }
4079
4080         jbd2_journal_clear_features(sbi->s_journal,
4081                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
4082                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
4083                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
4084         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4085                 ret = jbd2_journal_set_features(sbi->s_journal,
4086                                 compat, 0,
4087                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
4088                                 incompat);
4089         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
4090                 ret = jbd2_journal_set_features(sbi->s_journal,
4091                                 compat, 0,
4092                                 incompat);
4093                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
4094                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
4095         } else {
4096                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
4097                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
4098         }
4099
4100         return ret;
4101 }
4102
4103 /*
4104  * Note: calculating the overhead so we can be compatible with
4105  * historical BSD practice is quite difficult in the face of
4106  * clusters/bigalloc.  This is because multiple metadata blocks from
4107  * different block group can end up in the same allocation cluster.
4108  * Calculating the exact overhead in the face of clustered allocation
4109  * requires either O(all block bitmaps) in memory or O(number of block
4110  * groups**2) in time.  We will still calculate the superblock for
4111  * older file systems --- and if we come across with a bigalloc file
4112  * system with zero in s_overhead_clusters the estimate will be close to
4113  * correct especially for very large cluster sizes --- but for newer
4114  * file systems, it's better to calculate this figure once at mkfs
4115  * time, and store it in the superblock.  If the superblock value is
4116  * present (even for non-bigalloc file systems), we will use it.
4117  */
4118 static int count_overhead(struct super_block *sb, ext4_group_t grp,
4119                           char *buf)
4120 {
4121         struct ext4_sb_info     *sbi = EXT4_SB(sb);
4122         struct ext4_group_desc  *gdp;
4123         ext4_fsblk_t            first_block, last_block, b;
4124         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
4125         int                     s, j, count = 0;
4126         int                     has_super = ext4_bg_has_super(sb, grp);
4127
4128         if (!ext4_has_feature_bigalloc(sb))
4129                 return (has_super + ext4_bg_num_gdb(sb, grp) +
4130                         (has_super ? le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0) +
4131                         sbi->s_itb_per_group + 2);
4132
4133         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
4134                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
4135         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
4136         for (i = 0; i < ngroups; i++) {
4137                 gdp = ext4_get_group_desc(sb, i, NULL);
4138                 b = ext4_block_bitmap(sb, gdp);
4139                 if (b >= first_block && b <= last_block) {
4140                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
4141                         count++;
4142                 }
4143                 b = ext4_inode_bitmap(sb, gdp);
4144                 if (b >= first_block && b <= last_block) {
4145                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
4146                         count++;
4147                 }
4148                 b = ext4_inode_table(sb, gdp);
4149                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
4150                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
4151                                 int c = EXT4_B2C(sbi, b - first_block);
4152                                 ext4_set_bit(c, buf);
4153                                 count++;
4154                         }
4155                 if (i != grp)
4156                         continue;
4157                 s = 0;
4158                 if (ext4_bg_has_super(sb, grp)) {
4159                         ext4_set_bit(s++, buf);
4160                         count++;
4161                 }
4162                 j = ext4_bg_num_gdb(sb, grp);
4163                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
4164                         ext4_error(sb, "Invalid number of block group "
4165                                    "descriptor blocks: %d", j);
4166                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
4167                 }
4168                 count += j;
4169                 for (; j > 0; j--)
4170                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
4171         }
4172         if (!count)
4173                 return 0;
4174         return EXT4_CLUSTERS_PER_GROUP(sb) -
4175                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
4176 }
4177
4178 /*
4179  * Compute the overhead and stash it in sbi->s_overhead
4180  */
4181 int ext4_calculate_overhead(struct super_block *sb)
4182 {
4183         struct ext4_sb_info *sbi = EXT4_SB(sb);
4184         struct ext4_super_block *es = sbi->s_es;
4185         struct inode *j_inode;
4186         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
4187         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4188         ext4_fsblk_t overhead = 0;
4189         char *buf = (char *) get_zeroed_page(GFP_NOFS);
4190
4191         if (!buf)
4192                 return -ENOMEM;
4193
4194         /*
4195          * Compute the overhead (FS structures).  This is constant
4196          * for a given filesystem unless the number of block groups
4197          * changes so we cache the previous value until it does.
4198          */
4199
4200         /*
4201          * All of the blocks before first_data_block are overhead
4202          */
4203         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
4204
4205         /*
4206          * Add the overhead found in each block group
4207          */
4208         for (i = 0; i < ngroups; i++) {
4209                 int blks;
4210
4211                 blks = count_overhead(sb, i, buf);
4212                 overhead += blks;
4213                 if (blks)
4214                         memset(buf, 0, PAGE_SIZE);
4215                 cond_resched();
4216         }
4217
4218         /*
4219          * Add the internal journal blocks whether the journal has been
4220          * loaded or not
4221          */
4222         if (sbi->s_journal && !sbi->s_journal_bdev)
4223                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_total_len);
4224         else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) {
4225                 /* j_inum for internal journal is non-zero */
4226                 j_inode = ext4_get_journal_inode(sb, j_inum);
4227                 if (j_inode) {
4228                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
4229                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
4230                         iput(j_inode);
4231                 } else {
4232                         ext4_msg(sb, KERN_ERR, "can't get journal size");
4233                 }
4234         }
4235         sbi->s_overhead = overhead;
4236         smp_wmb();
4237         free_page((unsigned long) buf);
4238         return 0;
4239 }
4240
4241 static void ext4_set_resv_clusters(struct super_block *sb)
4242 {
4243         ext4_fsblk_t resv_clusters;
4244         struct ext4_sb_info *sbi = EXT4_SB(sb);
4245
4246         /*
4247          * There's no need to reserve anything when we aren't using extents.
4248          * The space estimates are exact, there are no unwritten extents,
4249          * hole punching doesn't need new metadata... This is needed especially
4250          * to keep ext2/3 backward compatibility.
4251          */
4252         if (!ext4_has_feature_extents(sb))
4253                 return;
4254         /*
4255          * By default we reserve 2% or 4096 clusters, whichever is smaller.
4256          * This should cover the situations where we can not afford to run
4257          * out of space like for example punch hole, or converting
4258          * unwritten extents in delalloc path. In most cases such
4259          * allocation would require 1, or 2 blocks, higher numbers are
4260          * very rare.
4261          */
4262         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
4263                          sbi->s_cluster_bits);
4264
4265         do_div(resv_clusters, 50);
4266         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
4267
4268         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
4269 }
4270
4271 static const char *ext4_quota_mode(struct super_block *sb)
4272 {
4273 #ifdef CONFIG_QUOTA
4274         if (!ext4_quota_capable(sb))
4275                 return "none";
4276
4277         if (EXT4_SB(sb)->s_journal && ext4_is_quota_journalled(sb))
4278                 return "journalled";
4279         else
4280                 return "writeback";
4281 #else
4282         return "disabled";
4283 #endif
4284 }
4285
4286 static void ext4_setup_csum_trigger(struct super_block *sb,
4287                                     enum ext4_journal_trigger_type type,
4288                                     void (*trigger)(
4289                                         struct jbd2_buffer_trigger_type *type,
4290                                         struct buffer_head *bh,
4291                                         void *mapped_data,
4292                                         size_t size))
4293 {
4294         struct ext4_sb_info *sbi = EXT4_SB(sb);
4295
4296         sbi->s_journal_triggers[type].sb = sb;
4297         sbi->s_journal_triggers[type].tr_triggers.t_frozen = trigger;
4298 }
4299
4300 static void ext4_free_sbi(struct ext4_sb_info *sbi)
4301 {
4302         if (!sbi)
4303                 return;
4304
4305         kfree(sbi->s_blockgroup_lock);
4306         fs_put_dax(sbi->s_daxdev, NULL);
4307         kfree(sbi);
4308 }
4309
4310 static struct ext4_sb_info *ext4_alloc_sbi(struct super_block *sb)
4311 {
4312         struct ext4_sb_info *sbi;
4313
4314         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
4315         if (!sbi)
4316                 return NULL;
4317
4318         sbi->s_daxdev = fs_dax_get_by_bdev(sb->s_bdev, &sbi->s_dax_part_off,
4319                                            NULL, NULL);
4320
4321         sbi->s_blockgroup_lock =
4322                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
4323
4324         if (!sbi->s_blockgroup_lock)
4325                 goto err_out;
4326
4327         sb->s_fs_info = sbi;
4328         sbi->s_sb = sb;
4329         return sbi;
4330 err_out:
4331         fs_put_dax(sbi->s_daxdev, NULL);
4332         kfree(sbi);
4333         return NULL;
4334 }
4335
4336 static void ext4_set_def_opts(struct super_block *sb,
4337                               struct ext4_super_block *es)
4338 {
4339         unsigned long def_mount_opts;
4340
4341         /* Set defaults before we parse the mount options */
4342         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
4343         set_opt(sb, INIT_INODE_TABLE);
4344         if (def_mount_opts & EXT4_DEFM_DEBUG)
4345                 set_opt(sb, DEBUG);
4346         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
4347                 set_opt(sb, GRPID);
4348         if (def_mount_opts & EXT4_DEFM_UID16)
4349                 set_opt(sb, NO_UID32);
4350         /* xattr user namespace & acls are now defaulted on */
4351         set_opt(sb, XATTR_USER);
4352 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4353         set_opt(sb, POSIX_ACL);
4354 #endif
4355         if (ext4_has_feature_fast_commit(sb))
4356                 set_opt2(sb, JOURNAL_FAST_COMMIT);
4357         /* don't forget to enable journal_csum when metadata_csum is enabled. */
4358         if (ext4_has_metadata_csum(sb))
4359                 set_opt(sb, JOURNAL_CHECKSUM);
4360
4361         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
4362                 set_opt(sb, JOURNAL_DATA);
4363         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
4364                 set_opt(sb, ORDERED_DATA);
4365         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
4366                 set_opt(sb, WRITEBACK_DATA);
4367
4368         if (le16_to_cpu(es->s_errors) == EXT4_ERRORS_PANIC)
4369                 set_opt(sb, ERRORS_PANIC);
4370         else if (le16_to_cpu(es->s_errors) == EXT4_ERRORS_CONTINUE)
4371                 set_opt(sb, ERRORS_CONT);
4372         else
4373                 set_opt(sb, ERRORS_RO);
4374         /* block_validity enabled by default; disable with noblock_validity */
4375         set_opt(sb, BLOCK_VALIDITY);
4376         if (def_mount_opts & EXT4_DEFM_DISCARD)
4377                 set_opt(sb, DISCARD);
4378
4379         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
4380                 set_opt(sb, BARRIER);
4381
4382         /*
4383          * enable delayed allocation by default
4384          * Use -o nodelalloc to turn it off
4385          */
4386         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
4387             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
4388                 set_opt(sb, DELALLOC);
4389
4390         if (sb->s_blocksize == PAGE_SIZE)
4391                 set_opt(sb, DIOREAD_NOLOCK);
4392 }
4393
4394 static int ext4_handle_clustersize(struct super_block *sb)
4395 {
4396         struct ext4_sb_info *sbi = EXT4_SB(sb);
4397         struct ext4_super_block *es = sbi->s_es;
4398         int clustersize;
4399
4400         /* Handle clustersize */
4401         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
4402         if (ext4_has_feature_bigalloc(sb)) {
4403                 if (clustersize < sb->s_blocksize) {
4404                         ext4_msg(sb, KERN_ERR,
4405                                  "cluster size (%d) smaller than "
4406                                  "block size (%lu)", clustersize, sb->s_blocksize);
4407                         return -EINVAL;
4408                 }
4409                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
4410                         le32_to_cpu(es->s_log_block_size);
4411                 sbi->s_clusters_per_group =
4412                         le32_to_cpu(es->s_clusters_per_group);
4413                 if (sbi->s_clusters_per_group > sb->s_blocksize * 8) {
4414                         ext4_msg(sb, KERN_ERR,
4415                                  "#clusters per group too big: %lu",
4416                                  sbi->s_clusters_per_group);
4417                         return -EINVAL;
4418                 }
4419                 if (sbi->s_blocks_per_group !=
4420                     (sbi->s_clusters_per_group * (clustersize / sb->s_blocksize))) {
4421                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
4422                                  "clusters per group (%lu) inconsistent",
4423                                  sbi->s_blocks_per_group,
4424                                  sbi->s_clusters_per_group);
4425                         return -EINVAL;
4426                 }
4427         } else {
4428                 if (clustersize != sb->s_blocksize) {
4429                         ext4_msg(sb, KERN_ERR,
4430                                  "fragment/cluster size (%d) != "
4431                                  "block size (%lu)", clustersize, sb->s_blocksize);
4432                         return -EINVAL;
4433                 }
4434                 if (sbi->s_blocks_per_group > sb->s_blocksize * 8) {
4435                         ext4_msg(sb, KERN_ERR,
4436                                  "#blocks per group too big: %lu",
4437                                  sbi->s_blocks_per_group);
4438                         return -EINVAL;
4439                 }
4440                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
4441                 sbi->s_cluster_bits = 0;
4442         }
4443         sbi->s_cluster_ratio = clustersize / sb->s_blocksize;
4444
4445         /* Do we have standard group size of clustersize * 8 blocks ? */
4446         if (sbi->s_blocks_per_group == clustersize << 3)
4447                 set_opt2(sb, STD_GROUP_SIZE);
4448
4449         return 0;
4450 }
4451
4452 static void ext4_fast_commit_init(struct super_block *sb)
4453 {
4454         struct ext4_sb_info *sbi = EXT4_SB(sb);
4455
4456         /* Initialize fast commit stuff */
4457         atomic_set(&sbi->s_fc_subtid, 0);
4458         INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_MAIN]);
4459         INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_STAGING]);
4460         INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_MAIN]);
4461         INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_STAGING]);
4462         sbi->s_fc_bytes = 0;
4463         ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE);
4464         sbi->s_fc_ineligible_tid = 0;
4465         spin_lock_init(&sbi->s_fc_lock);
4466         memset(&sbi->s_fc_stats, 0, sizeof(sbi->s_fc_stats));
4467         sbi->s_fc_replay_state.fc_regions = NULL;
4468         sbi->s_fc_replay_state.fc_regions_size = 0;
4469         sbi->s_fc_replay_state.fc_regions_used = 0;
4470         sbi->s_fc_replay_state.fc_regions_valid = 0;
4471         sbi->s_fc_replay_state.fc_modified_inodes = NULL;
4472         sbi->s_fc_replay_state.fc_modified_inodes_size = 0;
4473         sbi->s_fc_replay_state.fc_modified_inodes_used = 0;
4474 }
4475
4476 static int ext4_inode_info_init(struct super_block *sb,
4477                                 struct ext4_super_block *es)
4478 {
4479         struct ext4_sb_info *sbi = EXT4_SB(sb);
4480
4481         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
4482                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
4483                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
4484         } else {
4485                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
4486                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
4487                 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) {
4488                         ext4_msg(sb, KERN_ERR, "invalid first ino: %u",
4489                                  sbi->s_first_ino);
4490                         return -EINVAL;
4491                 }
4492                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
4493                     (!is_power_of_2(sbi->s_inode_size)) ||
4494                     (sbi->s_inode_size > sb->s_blocksize)) {
4495                         ext4_msg(sb, KERN_ERR,
4496                                "unsupported inode size: %d",
4497                                sbi->s_inode_size);
4498                         ext4_msg(sb, KERN_ERR, "blocksize: %lu", sb->s_blocksize);
4499                         return -EINVAL;
4500                 }
4501                 /*
4502                  * i_atime_extra is the last extra field available for
4503                  * [acm]times in struct ext4_inode. Checking for that
4504                  * field should suffice to ensure we have extra space
4505                  * for all three.
4506                  */
4507                 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) +
4508                         sizeof(((struct ext4_inode *)0)->i_atime_extra)) {
4509                         sb->s_time_gran = 1;
4510                         sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX;
4511                 } else {
4512                         sb->s_time_gran = NSEC_PER_SEC;
4513                         sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX;
4514                 }
4515                 sb->s_time_min = EXT4_TIMESTAMP_MIN;
4516         }
4517
4518         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
4519                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4520                         EXT4_GOOD_OLD_INODE_SIZE;
4521                 if (ext4_has_feature_extra_isize(sb)) {
4522                         unsigned v, max = (sbi->s_inode_size -
4523                                            EXT4_GOOD_OLD_INODE_SIZE);
4524
4525                         v = le16_to_cpu(es->s_want_extra_isize);
4526                         if (v > max) {
4527                                 ext4_msg(sb, KERN_ERR,
4528                                          "bad s_want_extra_isize: %d", v);
4529                                 return -EINVAL;
4530                         }
4531                         if (sbi->s_want_extra_isize < v)
4532                                 sbi->s_want_extra_isize = v;
4533
4534                         v = le16_to_cpu(es->s_min_extra_isize);
4535                         if (v > max) {
4536                                 ext4_msg(sb, KERN_ERR,
4537                                          "bad s_min_extra_isize: %d", v);
4538                                 return -EINVAL;
4539                         }
4540                         if (sbi->s_want_extra_isize < v)
4541                                 sbi->s_want_extra_isize = v;
4542                 }
4543         }
4544
4545         return 0;
4546 }
4547
4548 #if IS_ENABLED(CONFIG_UNICODE)
4549 static int ext4_encoding_init(struct super_block *sb, struct ext4_super_block *es)
4550 {
4551         const struct ext4_sb_encodings *encoding_info;
4552         struct unicode_map *encoding;
4553         __u16 encoding_flags = le16_to_cpu(es->s_encoding_flags);
4554
4555         if (!ext4_has_feature_casefold(sb) || sb->s_encoding)
4556                 return 0;
4557
4558         encoding_info = ext4_sb_read_encoding(es);
4559         if (!encoding_info) {
4560                 ext4_msg(sb, KERN_ERR,
4561                         "Encoding requested by superblock is unknown");
4562                 return -EINVAL;
4563         }
4564
4565         encoding = utf8_load(encoding_info->version);
4566         if (IS_ERR(encoding)) {
4567                 ext4_msg(sb, KERN_ERR,
4568                         "can't mount with superblock charset: %s-%u.%u.%u "
4569                         "not supported by the kernel. flags: 0x%x.",
4570                         encoding_info->name,
4571                         unicode_major(encoding_info->version),
4572                         unicode_minor(encoding_info->version),
4573                         unicode_rev(encoding_info->version),
4574                         encoding_flags);
4575                 return -EINVAL;
4576         }
4577         ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: "
4578                 "%s-%u.%u.%u with flags 0x%hx", encoding_info->name,
4579                 unicode_major(encoding_info->version),
4580                 unicode_minor(encoding_info->version),
4581                 unicode_rev(encoding_info->version),
4582                 encoding_flags);
4583
4584         sb->s_encoding = encoding;
4585         sb->s_encoding_flags = encoding_flags;
4586
4587         return 0;
4588 }
4589 #else
4590 static inline int ext4_encoding_init(struct super_block *sb, struct ext4_super_block *es)
4591 {
4592         return 0;
4593 }
4594 #endif
4595
4596 static int ext4_init_metadata_csum(struct super_block *sb, struct ext4_super_block *es)
4597 {
4598         struct ext4_sb_info *sbi = EXT4_SB(sb);
4599
4600         /* Warn if metadata_csum and gdt_csum are both set. */
4601         if (ext4_has_feature_metadata_csum(sb) &&
4602             ext4_has_feature_gdt_csum(sb))
4603                 ext4_warning(sb, "metadata_csum and uninit_bg are "
4604                              "redundant flags; please run fsck.");
4605
4606         /* Check for a known checksum algorithm */
4607         if (!ext4_verify_csum_type(sb, es)) {
4608                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
4609                          "unknown checksum algorithm.");
4610                 return -EINVAL;
4611         }
4612         ext4_setup_csum_trigger(sb, EXT4_JTR_ORPHAN_FILE,
4613                                 ext4_orphan_file_block_trigger);
4614
4615         /* Load the checksum driver */
4616         sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
4617         if (IS_ERR(sbi->s_chksum_driver)) {
4618                 int ret = PTR_ERR(sbi->s_chksum_driver);
4619                 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
4620                 sbi->s_chksum_driver = NULL;
4621                 return ret;
4622         }
4623
4624         /* Check superblock checksum */
4625         if (!ext4_superblock_csum_verify(sb, es)) {
4626                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
4627                          "invalid superblock checksum.  Run e2fsck?");
4628                 return -EFSBADCRC;
4629         }
4630
4631         /* Precompute checksum seed for all metadata */
4632         if (ext4_has_feature_csum_seed(sb))
4633                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
4634         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
4635                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
4636                                                sizeof(es->s_uuid));
4637         return 0;
4638 }
4639
4640 static int ext4_check_feature_compatibility(struct super_block *sb,
4641                                             struct ext4_super_block *es,
4642                                             int silent)
4643 {
4644         struct ext4_sb_info *sbi = EXT4_SB(sb);
4645
4646         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
4647             (ext4_has_compat_features(sb) ||
4648              ext4_has_ro_compat_features(sb) ||
4649              ext4_has_incompat_features(sb)))
4650                 ext4_msg(sb, KERN_WARNING,
4651                        "feature flags set on rev 0 fs, "
4652                        "running e2fsck is recommended");
4653
4654         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
4655                 set_opt2(sb, HURD_COMPAT);
4656                 if (ext4_has_feature_64bit(sb)) {
4657                         ext4_msg(sb, KERN_ERR,
4658                                  "The Hurd can't support 64-bit file systems");
4659                         return -EINVAL;
4660                 }
4661
4662                 /*
4663                  * ea_inode feature uses l_i_version field which is not
4664                  * available in HURD_COMPAT mode.
4665                  */
4666                 if (ext4_has_feature_ea_inode(sb)) {
4667                         ext4_msg(sb, KERN_ERR,
4668                                  "ea_inode feature is not supported for Hurd");
4669                         return -EINVAL;
4670                 }
4671         }
4672
4673         if (IS_EXT2_SB(sb)) {
4674                 if (ext2_feature_set_ok(sb))
4675                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
4676                                  "using the ext4 subsystem");
4677                 else {
4678                         /*
4679                          * If we're probing be silent, if this looks like
4680                          * it's actually an ext[34] filesystem.
4681                          */
4682                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4683                                 return -EINVAL;
4684                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
4685                                  "to feature incompatibilities");
4686                         return -EINVAL;
4687                 }
4688         }
4689
4690         if (IS_EXT3_SB(sb)) {
4691                 if (ext3_feature_set_ok(sb))
4692                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
4693                                  "using the ext4 subsystem");
4694                 else {
4695                         /*
4696                          * If we're probing be silent, if this looks like
4697                          * it's actually an ext4 filesystem.
4698                          */
4699                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4700                                 return -EINVAL;
4701                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
4702                                  "to feature incompatibilities");
4703                         return -EINVAL;
4704                 }
4705         }
4706
4707         /*
4708          * Check feature flags regardless of the revision level, since we
4709          * previously didn't change the revision level when setting the flags,
4710          * so there is a chance incompat flags are set on a rev 0 filesystem.
4711          */
4712         if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
4713                 return -EINVAL;
4714
4715         if (sbi->s_daxdev) {
4716                 if (sb->s_blocksize == PAGE_SIZE)
4717                         set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags);
4718                 else
4719                         ext4_msg(sb, KERN_ERR, "unsupported blocksize for DAX\n");
4720         }
4721
4722         if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) {
4723                 if (ext4_has_feature_inline_data(sb)) {
4724                         ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
4725                                         " that may contain inline data");
4726                         return -EINVAL;
4727                 }
4728                 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) {
4729                         ext4_msg(sb, KERN_ERR,
4730                                 "DAX unsupported by block device.");
4731                         return -EINVAL;
4732                 }
4733         }
4734
4735         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
4736                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
4737                          es->s_encryption_level);
4738                 return -EINVAL;
4739         }
4740
4741         return 0;
4742 }
4743
4744 static int ext4_check_geometry(struct super_block *sb,
4745                                struct ext4_super_block *es)
4746 {
4747         struct ext4_sb_info *sbi = EXT4_SB(sb);
4748         __u64 blocks_count;
4749         int err;
4750
4751         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (sb->s_blocksize / 4)) {
4752                 ext4_msg(sb, KERN_ERR,
4753                          "Number of reserved GDT blocks insanely large: %d",
4754                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
4755                 return -EINVAL;
4756         }
4757         /*
4758          * Test whether we have more sectors than will fit in sector_t,
4759          * and whether the max offset is addressable by the page cache.
4760          */
4761         err = generic_check_addressable(sb->s_blocksize_bits,
4762                                         ext4_blocks_count(es));
4763         if (err) {
4764                 ext4_msg(sb, KERN_ERR, "filesystem"
4765                          " too large to mount safely on this system");
4766                 return err;
4767         }
4768
4769         /* check blocks count against device size */
4770         blocks_count = sb_bdev_nr_blocks(sb);
4771         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
4772                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
4773                        "exceeds size of device (%llu blocks)",
4774                        ext4_blocks_count(es), blocks_count);
4775                 return -EINVAL;
4776         }
4777
4778         /*
4779          * It makes no sense for the first data block to be beyond the end
4780          * of the filesystem.
4781          */
4782         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
4783                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4784                          "block %u is beyond end of filesystem (%llu)",
4785                          le32_to_cpu(es->s_first_data_block),
4786                          ext4_blocks_count(es));
4787                 return -EINVAL;
4788         }
4789         if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) &&
4790             (sbi->s_cluster_ratio == 1)) {
4791                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4792                          "block is 0 with a 1k block and cluster size");
4793                 return -EINVAL;
4794         }
4795
4796         blocks_count = (ext4_blocks_count(es) -
4797                         le32_to_cpu(es->s_first_data_block) +
4798                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
4799         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
4800         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
4801                 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu "
4802                        "(block count %llu, first data block %u, "
4803                        "blocks per group %lu)", blocks_count,
4804                        ext4_blocks_count(es),
4805                        le32_to_cpu(es->s_first_data_block),
4806                        EXT4_BLOCKS_PER_GROUP(sb));
4807                 return -EINVAL;
4808         }
4809         sbi->s_groups_count = blocks_count;
4810         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
4811                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
4812         if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) !=
4813             le32_to_cpu(es->s_inodes_count)) {
4814                 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu",
4815                          le32_to_cpu(es->s_inodes_count),
4816                          ((u64)sbi->s_groups_count * sbi->s_inodes_per_group));
4817                 return -EINVAL;
4818         }
4819
4820         return 0;
4821 }
4822
4823 static int ext4_group_desc_init(struct super_block *sb,
4824                                 struct ext4_super_block *es,
4825                                 ext4_fsblk_t logical_sb_block,
4826                                 ext4_group_t *first_not_zeroed)
4827 {
4828         struct ext4_sb_info *sbi = EXT4_SB(sb);
4829         unsigned int db_count;
4830         ext4_fsblk_t block;
4831         int i;
4832
4833         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
4834                    EXT4_DESC_PER_BLOCK(sb);
4835         if (ext4_has_feature_meta_bg(sb)) {
4836                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
4837                         ext4_msg(sb, KERN_WARNING,
4838                                  "first meta block group too large: %u "
4839                                  "(group descriptor block count %u)",
4840                                  le32_to_cpu(es->s_first_meta_bg), db_count);
4841                         return -EINVAL;
4842                 }
4843         }
4844         rcu_assign_pointer(sbi->s_group_desc,
4845                            kvmalloc_array(db_count,
4846                                           sizeof(struct buffer_head *),
4847                                           GFP_KERNEL));
4848         if (sbi->s_group_desc == NULL) {
4849                 ext4_msg(sb, KERN_ERR, "not enough memory");
4850                 return -ENOMEM;
4851         }
4852
4853         bgl_lock_init(sbi->s_blockgroup_lock);
4854
4855         /* Pre-read the descriptors into the buffer cache */
4856         for (i = 0; i < db_count; i++) {
4857                 block = descriptor_loc(sb, logical_sb_block, i);
4858                 ext4_sb_breadahead_unmovable(sb, block);
4859         }
4860
4861         for (i = 0; i < db_count; i++) {
4862                 struct buffer_head *bh;
4863
4864                 block = descriptor_loc(sb, logical_sb_block, i);
4865                 bh = ext4_sb_bread_unmovable(sb, block);
4866                 if (IS_ERR(bh)) {
4867                         ext4_msg(sb, KERN_ERR,
4868                                "can't read group descriptor %d", i);
4869                         sbi->s_gdb_count = i;
4870                         return PTR_ERR(bh);
4871                 }
4872                 rcu_read_lock();
4873                 rcu_dereference(sbi->s_group_desc)[i] = bh;
4874                 rcu_read_unlock();
4875         }
4876         sbi->s_gdb_count = db_count;
4877         if (!ext4_check_descriptors(sb, logical_sb_block, first_not_zeroed)) {
4878                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
4879                 return -EFSCORRUPTED;
4880         }
4881
4882         return 0;
4883 }
4884
4885 static int ext4_load_and_init_journal(struct super_block *sb,
4886                                       struct ext4_super_block *es,
4887                                       struct ext4_fs_context *ctx)
4888 {
4889         struct ext4_sb_info *sbi = EXT4_SB(sb);
4890         int err;
4891
4892         err = ext4_load_journal(sb, es, ctx->journal_devnum);
4893         if (err)
4894                 return err;
4895
4896         if (ext4_has_feature_64bit(sb) &&
4897             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4898                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4899                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4900                 goto out;
4901         }
4902
4903         if (!set_journal_csum_feature_set(sb)) {
4904                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4905                          "feature set");
4906                 goto out;
4907         }
4908
4909         if (test_opt2(sb, JOURNAL_FAST_COMMIT) &&
4910                 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4911                                           JBD2_FEATURE_INCOMPAT_FAST_COMMIT)) {
4912                 ext4_msg(sb, KERN_ERR,
4913                         "Failed to set fast commit journal feature");
4914                 goto out;
4915         }
4916
4917         /* We have now updated the journal if required, so we can
4918          * validate the data journaling mode. */
4919         switch (test_opt(sb, DATA_FLAGS)) {
4920         case 0:
4921                 /* No mode set, assume a default based on the journal
4922                  * capabilities: ORDERED_DATA if the journal can
4923                  * cope, else JOURNAL_DATA
4924                  */
4925                 if (jbd2_journal_check_available_features
4926                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4927                         set_opt(sb, ORDERED_DATA);
4928                         sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
4929                 } else {
4930                         set_opt(sb, JOURNAL_DATA);
4931                         sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
4932                 }
4933                 break;
4934
4935         case EXT4_MOUNT_ORDERED_DATA:
4936         case EXT4_MOUNT_WRITEBACK_DATA:
4937                 if (!jbd2_journal_check_available_features
4938                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4939                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4940                                "requested data journaling mode");
4941                         goto out;
4942                 }
4943                 break;
4944         default:
4945                 break;
4946         }
4947
4948         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4949             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4950                 ext4_msg(sb, KERN_ERR, "can't mount with "
4951                         "journal_async_commit in data=ordered mode");
4952                 goto out;
4953         }
4954
4955         set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio);
4956
4957         sbi->s_journal->j_submit_inode_data_buffers =
4958                 ext4_journal_submit_inode_data_buffers;
4959         sbi->s_journal->j_finish_inode_data_buffers =
4960                 ext4_journal_finish_inode_data_buffers;
4961
4962         return 0;
4963
4964 out:
4965         /* flush s_error_work before journal destroy. */
4966         flush_work(&sbi->s_error_work);
4967         jbd2_journal_destroy(sbi->s_journal);
4968         sbi->s_journal = NULL;
4969         return -EINVAL;
4970 }
4971
4972 static int ext4_check_journal_data_mode(struct super_block *sb)
4973 {
4974         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4975                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with "
4976                             "data=journal disables delayed allocation, "
4977                             "dioread_nolock, O_DIRECT and fast_commit support!\n");
4978                 /* can't mount with both data=journal and dioread_nolock. */
4979                 clear_opt(sb, DIOREAD_NOLOCK);
4980                 clear_opt2(sb, JOURNAL_FAST_COMMIT);
4981                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4982                         ext4_msg(sb, KERN_ERR, "can't mount with "
4983                                  "both data=journal and delalloc");
4984                         return -EINVAL;
4985                 }
4986                 if (test_opt(sb, DAX_ALWAYS)) {
4987                         ext4_msg(sb, KERN_ERR, "can't mount with "
4988                                  "both data=journal and dax");
4989                         return -EINVAL;
4990                 }
4991                 if (ext4_has_feature_encrypt(sb)) {
4992                         ext4_msg(sb, KERN_WARNING,
4993                                  "encrypted files will use data=ordered "
4994                                  "instead of data journaling mode");
4995                 }
4996                 if (test_opt(sb, DELALLOC))
4997                         clear_opt(sb, DELALLOC);
4998         } else {
4999                 sb->s_iflags |= SB_I_CGROUPWB;
5000         }
5001
5002         return 0;
5003 }
5004
5005 static int ext4_load_super(struct super_block *sb, ext4_fsblk_t *lsb,
5006                            int silent)
5007 {
5008         struct ext4_sb_info *sbi = EXT4_SB(sb);
5009         struct ext4_super_block *es;
5010         ext4_fsblk_t logical_sb_block;
5011         unsigned long offset = 0;
5012         struct buffer_head *bh;
5013         int ret = -EINVAL;
5014         int blocksize;
5015
5016         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
5017         if (!blocksize) {
5018                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
5019                 return -EINVAL;
5020         }
5021
5022         /*
5023          * The ext4 superblock will not be buffer aligned for other than 1kB
5024          * block sizes.  We need to calculate the offset from buffer start.
5025          */
5026         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
5027                 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE;
5028                 offset = do_div(logical_sb_block, blocksize);
5029         } else {
5030                 logical_sb_block = sbi->s_sb_block;
5031         }
5032
5033         bh = ext4_sb_bread_unmovable(sb, logical_sb_block);
5034         if (IS_ERR(bh)) {
5035                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
5036                 return PTR_ERR(bh);
5037         }
5038         /*
5039          * Note: s_es must be initialized as soon as possible because
5040          *       some ext4 macro-instructions depend on its value
5041          */
5042         es = (struct ext4_super_block *) (bh->b_data + offset);
5043         sbi->s_es = es;
5044         sb->s_magic = le16_to_cpu(es->s_magic);
5045         if (sb->s_magic != EXT4_SUPER_MAGIC) {
5046                 if (!silent)
5047                         ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
5048                 goto out;
5049         }
5050
5051         if (le32_to_cpu(es->s_log_block_size) >
5052             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
5053                 ext4_msg(sb, KERN_ERR,
5054                          "Invalid log block size: %u",
5055                          le32_to_cpu(es->s_log_block_size));
5056                 goto out;
5057         }
5058         if (le32_to_cpu(es->s_log_cluster_size) >
5059             (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
5060                 ext4_msg(sb, KERN_ERR,
5061                          "Invalid log cluster size: %u",
5062                          le32_to_cpu(es->s_log_cluster_size));
5063                 goto out;
5064         }
5065
5066         blocksize = EXT4_MIN_BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
5067
5068         /*
5069          * If the default block size is not the same as the real block size,
5070          * we need to reload it.
5071          */
5072         if (sb->s_blocksize == blocksize) {
5073                 *lsb = logical_sb_block;
5074                 sbi->s_sbh = bh;
5075                 return 0;
5076         }
5077
5078         /*
5079          * bh must be released before kill_bdev(), otherwise
5080          * it won't be freed and its page also. kill_bdev()
5081          * is called by sb_set_blocksize().
5082          */
5083         brelse(bh);
5084         /* Validate the filesystem blocksize */
5085         if (!sb_set_blocksize(sb, blocksize)) {
5086                 ext4_msg(sb, KERN_ERR, "bad block size %d",
5087                                 blocksize);
5088                 bh = NULL;
5089                 goto out;
5090         }
5091
5092         logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE;
5093         offset = do_div(logical_sb_block, blocksize);
5094         bh = ext4_sb_bread_unmovable(sb, logical_sb_block);
5095         if (IS_ERR(bh)) {
5096                 ext4_msg(sb, KERN_ERR, "Can't read superblock on 2nd try");
5097                 ret = PTR_ERR(bh);
5098                 bh = NULL;
5099                 goto out;
5100         }
5101         es = (struct ext4_super_block *)(bh->b_data + offset);
5102         sbi->s_es = es;
5103         if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
5104                 ext4_msg(sb, KERN_ERR, "Magic mismatch, very weird!");
5105                 goto out;
5106         }
5107         *lsb = logical_sb_block;
5108         sbi->s_sbh = bh;
5109         return 0;
5110 out:
5111         brelse(bh);
5112         return ret;
5113 }
5114
5115 static void ext4_hash_info_init(struct super_block *sb)
5116 {
5117         struct ext4_sb_info *sbi = EXT4_SB(sb);
5118         struct ext4_super_block *es = sbi->s_es;
5119         unsigned int i;
5120
5121         for (i = 0; i < 4; i++)
5122                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
5123
5124         sbi->s_def_hash_version = es->s_def_hash_version;
5125         if (ext4_has_feature_dir_index(sb)) {
5126                 i = le32_to_cpu(es->s_flags);
5127                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
5128                         sbi->s_hash_unsigned = 3;
5129                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
5130 #ifdef __CHAR_UNSIGNED__
5131                         if (!sb_rdonly(sb))
5132                                 es->s_flags |=
5133                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
5134                         sbi->s_hash_unsigned = 3;
5135 #else
5136                         if (!sb_rdonly(sb))
5137                                 es->s_flags |=
5138                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
5139 #endif
5140                 }
5141         }
5142 }
5143
5144 static int ext4_block_group_meta_init(struct super_block *sb, int silent)
5145 {
5146         struct ext4_sb_info *sbi = EXT4_SB(sb);
5147         struct ext4_super_block *es = sbi->s_es;
5148         int has_huge_files;
5149
5150         has_huge_files = ext4_has_feature_huge_file(sb);
5151         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
5152                                                       has_huge_files);
5153         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
5154
5155         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
5156         if (ext4_has_feature_64bit(sb)) {
5157                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
5158                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
5159                     !is_power_of_2(sbi->s_desc_size)) {
5160                         ext4_msg(sb, KERN_ERR,
5161                                "unsupported descriptor size %lu",
5162                                sbi->s_desc_size);
5163                         return -EINVAL;
5164                 }
5165         } else
5166                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
5167
5168         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
5169         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
5170
5171         sbi->s_inodes_per_block = sb->s_blocksize / EXT4_INODE_SIZE(sb);
5172         if (sbi->s_inodes_per_block == 0 || sbi->s_blocks_per_group == 0) {
5173                 if (!silent)
5174                         ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
5175                 return -EINVAL;
5176         }
5177         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
5178             sbi->s_inodes_per_group > sb->s_blocksize * 8) {
5179                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
5180                          sbi->s_inodes_per_group);
5181                 return -EINVAL;
5182         }
5183         sbi->s_itb_per_group = sbi->s_inodes_per_group /
5184                                         sbi->s_inodes_per_block;
5185         sbi->s_desc_per_block = sb->s_blocksize / EXT4_DESC_SIZE(sb);
5186         sbi->s_mount_state = le16_to_cpu(es->s_state) & ~EXT4_FC_REPLAY;
5187         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
5188         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
5189
5190         return 0;
5191 }
5192
5193 static int __ext4_fill_super(struct fs_context *fc, struct super_block *sb)
5194 {
5195         struct ext4_super_block *es = NULL;
5196         struct ext4_sb_info *sbi = EXT4_SB(sb);
5197         ext4_fsblk_t logical_sb_block;
5198         struct inode *root;
5199         int ret = -ENOMEM;
5200         unsigned int i;
5201         int needs_recovery;
5202         int err = 0;
5203         ext4_group_t first_not_zeroed;
5204         struct ext4_fs_context *ctx = fc->fs_private;
5205         int silent = fc->sb_flags & SB_SILENT;
5206
5207         /* Set defaults for the variables that will be set during parsing */
5208         if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO))
5209                 ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
5210
5211         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
5212         sbi->s_sectors_written_start =
5213                 part_stat_read(sb->s_bdev, sectors[STAT_WRITE]);
5214
5215         /* -EINVAL is default */
5216         ret = -EINVAL;
5217         err = ext4_load_super(sb, &logical_sb_block, silent);
5218         if (err)
5219                 goto out_fail;
5220
5221         es = sbi->s_es;
5222         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
5223
5224         err = ext4_init_metadata_csum(sb, es);
5225         if (err)
5226                 goto failed_mount;
5227
5228         ext4_set_def_opts(sb, es);
5229
5230         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
5231         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
5232         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
5233         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
5234         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
5235
5236         /*
5237          * set default s_li_wait_mult for lazyinit, for the case there is
5238          * no mount option specified.
5239          */
5240         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
5241
5242         if (ext4_inode_info_init(sb, es))
5243                 goto failed_mount;
5244
5245         err = parse_apply_sb_mount_options(sb, ctx);
5246         if (err < 0)
5247                 goto failed_mount;
5248
5249         sbi->s_def_mount_opt = sbi->s_mount_opt;
5250         sbi->s_def_mount_opt2 = sbi->s_mount_opt2;
5251
5252         err = ext4_check_opt_consistency(fc, sb);
5253         if (err < 0)
5254                 goto failed_mount;
5255
5256         ext4_apply_options(fc, sb);
5257
5258         if (ext4_encoding_init(sb, es))
5259                 goto failed_mount;
5260
5261         if (ext4_check_journal_data_mode(sb))
5262                 goto failed_mount;
5263
5264         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
5265                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
5266
5267         /* i_version is always enabled now */
5268         sb->s_flags |= SB_I_VERSION;
5269
5270         if (ext4_check_feature_compatibility(sb, es, silent))
5271                 goto failed_mount;
5272
5273         if (ext4_block_group_meta_init(sb, silent))
5274                 goto failed_mount;
5275
5276         ext4_hash_info_init(sb);
5277
5278         if (ext4_handle_clustersize(sb))
5279                 goto failed_mount;
5280
5281         if (ext4_check_geometry(sb, es))
5282                 goto failed_mount;
5283
5284         timer_setup(&sbi->s_err_report, print_daily_error_info, 0);
5285         spin_lock_init(&sbi->s_error_lock);
5286         INIT_WORK(&sbi->s_error_work, flush_stashed_error_work);
5287
5288         err = ext4_group_desc_init(sb, es, logical_sb_block, &first_not_zeroed);
5289         if (err)
5290                 goto failed_mount3;
5291
5292         /* Register extent status tree shrinker */
5293         if (ext4_es_register_shrinker(sbi))
5294                 goto failed_mount3;
5295
5296         sbi->s_stripe = ext4_get_stripe_size(sbi);
5297         sbi->s_extent_max_zeroout_kb = 32;
5298
5299         /*
5300          * set up enough so that it can read an inode
5301          */
5302         sb->s_op = &ext4_sops;
5303         sb->s_export_op = &ext4_export_ops;
5304         sb->s_xattr = ext4_xattr_handlers;
5305 #ifdef CONFIG_FS_ENCRYPTION
5306         sb->s_cop = &ext4_cryptops;
5307 #endif
5308 #ifdef CONFIG_FS_VERITY
5309         sb->s_vop = &ext4_verityops;
5310 #endif
5311 #ifdef CONFIG_QUOTA
5312         sb->dq_op = &ext4_quota_operations;
5313         if (ext4_has_feature_quota(sb))
5314                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
5315         else
5316                 sb->s_qcop = &ext4_qctl_operations;
5317         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
5318 #endif
5319         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
5320
5321         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
5322         mutex_init(&sbi->s_orphan_lock);
5323
5324         ext4_fast_commit_init(sb);
5325
5326         sb->s_root = NULL;
5327
5328         needs_recovery = (es->s_last_orphan != 0 ||
5329                           ext4_has_feature_orphan_present(sb) ||
5330                           ext4_has_feature_journal_needs_recovery(sb));
5331
5332         if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb))
5333                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
5334                         goto failed_mount3a;
5335
5336         /*
5337          * The first inode we look at is the journal inode.  Don't try
5338          * root first: it may be modified in the journal!
5339          */
5340         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
5341                 err = ext4_load_and_init_journal(sb, es, ctx);
5342                 if (err)
5343                         goto failed_mount3a;
5344         } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
5345                    ext4_has_feature_journal_needs_recovery(sb)) {
5346                 ext4_msg(sb, KERN_ERR, "required journal recovery "
5347                        "suppressed and not mounted read-only");
5348                 goto failed_mount3a;
5349         } else {
5350                 /* Nojournal mode, all journal mount options are illegal */
5351                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5352                         ext4_msg(sb, KERN_ERR, "can't mount with "
5353                                  "journal_async_commit, fs mounted w/o journal");
5354                         goto failed_mount3a;
5355                 }
5356
5357                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
5358                         ext4_msg(sb, KERN_ERR, "can't mount with "
5359                                  "journal_checksum, fs mounted w/o journal");
5360                         goto failed_mount3a;
5361                 }
5362                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
5363                         ext4_msg(sb, KERN_ERR, "can't mount with "
5364                                  "commit=%lu, fs mounted w/o journal",
5365                                  sbi->s_commit_interval / HZ);
5366                         goto failed_mount3a;
5367                 }
5368                 if (EXT4_MOUNT_DATA_FLAGS &
5369                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
5370                         ext4_msg(sb, KERN_ERR, "can't mount with "
5371                                  "data=, fs mounted w/o journal");
5372                         goto failed_mount3a;
5373                 }
5374                 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM;
5375                 clear_opt(sb, JOURNAL_CHECKSUM);
5376                 clear_opt(sb, DATA_FLAGS);
5377                 clear_opt2(sb, JOURNAL_FAST_COMMIT);
5378                 sbi->s_journal = NULL;
5379                 needs_recovery = 0;
5380         }
5381
5382         if (!test_opt(sb, NO_MBCACHE)) {
5383                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
5384                 if (!sbi->s_ea_block_cache) {
5385                         ext4_msg(sb, KERN_ERR,
5386                                  "Failed to create ea_block_cache");
5387                         goto failed_mount_wq;
5388                 }
5389
5390                 if (ext4_has_feature_ea_inode(sb)) {
5391                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
5392                         if (!sbi->s_ea_inode_cache) {
5393                                 ext4_msg(sb, KERN_ERR,
5394                                          "Failed to create ea_inode_cache");
5395                                 goto failed_mount_wq;
5396                         }
5397                 }
5398         }
5399
5400         /*
5401          * Get the # of file system overhead blocks from the
5402          * superblock if present.
5403          */
5404         sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
5405         /* ignore the precalculated value if it is ridiculous */
5406         if (sbi->s_overhead > ext4_blocks_count(es))
5407                 sbi->s_overhead = 0;
5408         /*
5409          * If the bigalloc feature is not enabled recalculating the
5410          * overhead doesn't take long, so we might as well just redo
5411          * it to make sure we are using the correct value.
5412          */
5413         if (!ext4_has_feature_bigalloc(sb))
5414                 sbi->s_overhead = 0;
5415         if (sbi->s_overhead == 0) {
5416                 err = ext4_calculate_overhead(sb);
5417                 if (err)
5418                         goto failed_mount_wq;
5419         }
5420
5421         /*
5422          * The maximum number of concurrent works can be high and
5423          * concurrency isn't really necessary.  Limit it to 1.
5424          */
5425         EXT4_SB(sb)->rsv_conversion_wq =
5426                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
5427         if (!EXT4_SB(sb)->rsv_conversion_wq) {
5428                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
5429                 ret = -ENOMEM;
5430                 goto failed_mount4;
5431         }
5432
5433         /*
5434          * The jbd2_journal_load will have done any necessary log recovery,
5435          * so we can safely mount the rest of the filesystem now.
5436          */
5437
5438         root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL);
5439         if (IS_ERR(root)) {
5440                 ext4_msg(sb, KERN_ERR, "get root inode failed");
5441                 ret = PTR_ERR(root);
5442                 root = NULL;
5443                 goto failed_mount4;
5444         }
5445         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
5446                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
5447                 iput(root);
5448                 goto failed_mount4;
5449         }
5450
5451         sb->s_root = d_make_root(root);
5452         if (!sb->s_root) {
5453                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
5454                 ret = -ENOMEM;
5455                 goto failed_mount4;
5456         }
5457
5458         ret = ext4_setup_super(sb, es, sb_rdonly(sb));
5459         if (ret == -EROFS) {
5460                 sb->s_flags |= SB_RDONLY;
5461                 ret = 0;
5462         } else if (ret)
5463                 goto failed_mount4a;
5464
5465         ext4_set_resv_clusters(sb);
5466
5467         if (test_opt(sb, BLOCK_VALIDITY)) {
5468                 err = ext4_setup_system_zone(sb);
5469                 if (err) {
5470                         ext4_msg(sb, KERN_ERR, "failed to initialize system "
5471                                  "zone (%d)", err);
5472                         goto failed_mount4a;
5473                 }
5474         }
5475         ext4_fc_replay_cleanup(sb);
5476
5477         ext4_ext_init(sb);
5478
5479         /*
5480          * Enable optimize_scan if number of groups is > threshold. This can be
5481          * turned off by passing "mb_optimize_scan=0". This can also be
5482          * turned on forcefully by passing "mb_optimize_scan=1".
5483          */
5484         if (!(ctx->spec & EXT4_SPEC_mb_optimize_scan)) {
5485                 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD)
5486                         set_opt2(sb, MB_OPTIMIZE_SCAN);
5487                 else
5488                         clear_opt2(sb, MB_OPTIMIZE_SCAN);
5489         }
5490
5491         err = ext4_mb_init(sb);
5492         if (err) {
5493                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
5494                          err);
5495                 goto failed_mount5;
5496         }
5497
5498         /*
5499          * We can only set up the journal commit callback once
5500          * mballoc is initialized
5501          */
5502         if (sbi->s_journal)
5503                 sbi->s_journal->j_commit_callback =
5504                         ext4_journal_commit_callback;
5505
5506         if (ext4_percpu_param_init(sbi))
5507                 goto failed_mount6;
5508
5509         if (ext4_has_feature_flex_bg(sb))
5510                 if (!ext4_fill_flex_info(sb)) {
5511                         ext4_msg(sb, KERN_ERR,
5512                                "unable to initialize "
5513                                "flex_bg meta info!");
5514                         ret = -ENOMEM;
5515                         goto failed_mount6;
5516                 }
5517
5518         err = ext4_register_li_request(sb, first_not_zeroed);
5519         if (err)
5520                 goto failed_mount6;
5521
5522         err = ext4_register_sysfs(sb);
5523         if (err)
5524                 goto failed_mount7;
5525
5526         err = ext4_init_orphan_info(sb);
5527         if (err)
5528                 goto failed_mount8;
5529 #ifdef CONFIG_QUOTA
5530         /* Enable quota usage during mount. */
5531         if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
5532                 err = ext4_enable_quotas(sb);
5533                 if (err)
5534                         goto failed_mount9;
5535         }
5536 #endif  /* CONFIG_QUOTA */
5537
5538         /*
5539          * Save the original bdev mapping's wb_err value which could be
5540          * used to detect the metadata async write error.
5541          */
5542         spin_lock_init(&sbi->s_bdev_wb_lock);
5543         errseq_check_and_advance(&sb->s_bdev->bd_inode->i_mapping->wb_err,
5544                                  &sbi->s_bdev_wb_err);
5545         sb->s_bdev->bd_super = sb;
5546         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
5547         ext4_orphan_cleanup(sb, es);
5548         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
5549         /*
5550          * Update the checksum after updating free space/inode counters and
5551          * ext4_orphan_cleanup. Otherwise the superblock can have an incorrect
5552          * checksum in the buffer cache until it is written out and
5553          * e2fsprogs programs trying to open a file system immediately
5554          * after it is mounted can fail.
5555          */
5556         ext4_superblock_csum_set(sb);
5557         if (needs_recovery) {
5558                 ext4_msg(sb, KERN_INFO, "recovery complete");
5559                 err = ext4_mark_recovery_complete(sb, es);
5560                 if (err)
5561                         goto failed_mount9;
5562         }
5563
5564         if (test_opt(sb, DISCARD) && !bdev_max_discard_sectors(sb->s_bdev))
5565                 ext4_msg(sb, KERN_WARNING,
5566                          "mounting with \"discard\" option, but the device does not support discard");
5567
5568         if (es->s_error_count)
5569                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
5570
5571         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
5572         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
5573         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
5574         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
5575         atomic_set(&sbi->s_warning_count, 0);
5576         atomic_set(&sbi->s_msg_count, 0);
5577
5578         return 0;
5579
5580 failed_mount9:
5581         ext4_release_orphan_info(sb);
5582 failed_mount8:
5583         ext4_unregister_sysfs(sb);
5584         kobject_put(&sbi->s_kobj);
5585 failed_mount7:
5586         ext4_unregister_li_request(sb);
5587 failed_mount6:
5588         ext4_mb_release(sb);
5589         ext4_flex_groups_free(sbi);
5590         ext4_percpu_param_destroy(sbi);
5591 failed_mount5:
5592         ext4_ext_release(sb);
5593         ext4_release_system_zone(sb);
5594 failed_mount4a:
5595         dput(sb->s_root);
5596         sb->s_root = NULL;
5597 failed_mount4:
5598         ext4_msg(sb, KERN_ERR, "mount failed");
5599         if (EXT4_SB(sb)->rsv_conversion_wq)
5600                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
5601 failed_mount_wq:
5602         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
5603         sbi->s_ea_inode_cache = NULL;
5604
5605         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
5606         sbi->s_ea_block_cache = NULL;
5607
5608         if (sbi->s_journal) {
5609                 /* flush s_error_work before journal destroy. */
5610                 flush_work(&sbi->s_error_work);
5611                 jbd2_journal_destroy(sbi->s_journal);
5612                 sbi->s_journal = NULL;
5613         }
5614 failed_mount3a:
5615         ext4_es_unregister_shrinker(sbi);
5616 failed_mount3:
5617         /* flush s_error_work before sbi destroy */
5618         flush_work(&sbi->s_error_work);
5619         del_timer_sync(&sbi->s_err_report);
5620         ext4_stop_mmpd(sbi);
5621         ext4_group_desc_free(sbi);
5622 failed_mount:
5623         if (sbi->s_chksum_driver)
5624                 crypto_free_shash(sbi->s_chksum_driver);
5625
5626 #if IS_ENABLED(CONFIG_UNICODE)
5627         utf8_unload(sb->s_encoding);
5628 #endif
5629
5630 #ifdef CONFIG_QUOTA
5631         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5632                 kfree(get_qf_name(sb, sbi, i));
5633 #endif
5634         fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy);
5635         /* ext4_blkdev_remove() calls kill_bdev(), release bh before it. */
5636         brelse(sbi->s_sbh);
5637         ext4_blkdev_remove(sbi);
5638 out_fail:
5639         sb->s_fs_info = NULL;
5640         return err ? err : ret;
5641 }
5642
5643 static int ext4_fill_super(struct super_block *sb, struct fs_context *fc)
5644 {
5645         struct ext4_fs_context *ctx = fc->fs_private;
5646         struct ext4_sb_info *sbi;
5647         const char *descr;
5648         int ret;
5649
5650         sbi = ext4_alloc_sbi(sb);
5651         if (!sbi)
5652                 return -ENOMEM;
5653
5654         fc->s_fs_info = sbi;
5655
5656         /* Cleanup superblock name */
5657         strreplace(sb->s_id, '/', '!');
5658
5659         sbi->s_sb_block = 1;    /* Default super block location */
5660         if (ctx->spec & EXT4_SPEC_s_sb_block)
5661                 sbi->s_sb_block = ctx->s_sb_block;
5662
5663         ret = __ext4_fill_super(fc, sb);
5664         if (ret < 0)
5665                 goto free_sbi;
5666
5667         if (sbi->s_journal) {
5668                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
5669                         descr = " journalled data mode";
5670                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
5671                         descr = " ordered data mode";
5672                 else
5673                         descr = " writeback data mode";
5674         } else
5675                 descr = "out journal";
5676
5677         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
5678                 ext4_msg(sb, KERN_INFO, "mounted filesystem %pU with%s. "
5679                          "Quota mode: %s.", &sb->s_uuid, descr,
5680                          ext4_quota_mode(sb));
5681
5682         /* Update the s_overhead_clusters if necessary */
5683         ext4_update_overhead(sb, false);
5684         return 0;
5685
5686 free_sbi:
5687         ext4_free_sbi(sbi);
5688         fc->s_fs_info = NULL;
5689         return ret;
5690 }
5691
5692 static int ext4_get_tree(struct fs_context *fc)
5693 {
5694         return get_tree_bdev(fc, ext4_fill_super);
5695 }
5696
5697 /*
5698  * Setup any per-fs journal parameters now.  We'll do this both on
5699  * initial mount, once the journal has been initialised but before we've
5700  * done any recovery; and again on any subsequent remount.
5701  */
5702 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
5703 {
5704         struct ext4_sb_info *sbi = EXT4_SB(sb);
5705
5706         journal->j_commit_interval = sbi->s_commit_interval;
5707         journal->j_min_batch_time = sbi->s_min_batch_time;
5708         journal->j_max_batch_time = sbi->s_max_batch_time;
5709         ext4_fc_init(sb, journal);
5710
5711         write_lock(&journal->j_state_lock);
5712         if (test_opt(sb, BARRIER))
5713                 journal->j_flags |= JBD2_BARRIER;
5714         else
5715                 journal->j_flags &= ~JBD2_BARRIER;
5716         if (test_opt(sb, DATA_ERR_ABORT))
5717                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
5718         else
5719                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
5720         write_unlock(&journal->j_state_lock);
5721 }
5722
5723 static struct inode *ext4_get_journal_inode(struct super_block *sb,
5724                                              unsigned int journal_inum)
5725 {
5726         struct inode *journal_inode;
5727
5728         /*
5729          * Test for the existence of a valid inode on disk.  Bad things
5730          * happen if we iget() an unused inode, as the subsequent iput()
5731          * will try to delete it.
5732          */
5733         journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL);
5734         if (IS_ERR(journal_inode)) {
5735                 ext4_msg(sb, KERN_ERR, "no journal found");
5736                 return NULL;
5737         }
5738         if (!journal_inode->i_nlink) {
5739                 make_bad_inode(journal_inode);
5740                 iput(journal_inode);
5741                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
5742                 return NULL;
5743         }
5744
5745         ext4_debug("Journal inode found at %p: %lld bytes\n",
5746                   journal_inode, journal_inode->i_size);
5747         if (!S_ISREG(journal_inode->i_mode) || IS_ENCRYPTED(journal_inode)) {
5748                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
5749                 iput(journal_inode);
5750                 return NULL;
5751         }
5752         return journal_inode;
5753 }
5754
5755 static int ext4_journal_bmap(journal_t *journal, sector_t *block)
5756 {
5757         struct ext4_map_blocks map;
5758         int ret;
5759
5760         if (journal->j_inode == NULL)
5761                 return 0;
5762
5763         map.m_lblk = *block;
5764         map.m_len = 1;
5765         ret = ext4_map_blocks(NULL, journal->j_inode, &map, 0);
5766         if (ret <= 0) {
5767                 ext4_msg(journal->j_inode->i_sb, KERN_CRIT,
5768                          "journal bmap failed: block %llu ret %d\n",
5769                          *block, ret);
5770                 jbd2_journal_abort(journal, ret ? ret : -EIO);
5771                 return ret;
5772         }
5773         *block = map.m_pblk;
5774         return 0;
5775 }
5776
5777 static journal_t *ext4_get_journal(struct super_block *sb,
5778                                    unsigned int journal_inum)
5779 {
5780         struct inode *journal_inode;
5781         journal_t *journal;
5782
5783         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5784                 return NULL;
5785
5786         journal_inode = ext4_get_journal_inode(sb, journal_inum);
5787         if (!journal_inode)
5788                 return NULL;
5789
5790         journal = jbd2_journal_init_inode(journal_inode);
5791         if (!journal) {
5792                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
5793                 iput(journal_inode);
5794                 return NULL;
5795         }
5796         journal->j_private = sb;
5797         journal->j_bmap = ext4_journal_bmap;
5798         ext4_init_journal_params(sb, journal);
5799         return journal;
5800 }
5801
5802 static journal_t *ext4_get_dev_journal(struct super_block *sb,
5803                                        dev_t j_dev)
5804 {
5805         struct buffer_head *bh;
5806         journal_t *journal;
5807         ext4_fsblk_t start;
5808         ext4_fsblk_t len;
5809         int hblock, blocksize;
5810         ext4_fsblk_t sb_block;
5811         unsigned long offset;
5812         struct ext4_super_block *es;
5813         struct block_device *bdev;
5814
5815         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5816                 return NULL;
5817
5818         bdev = ext4_blkdev_get(j_dev, sb);
5819         if (bdev == NULL)
5820                 return NULL;
5821
5822         blocksize = sb->s_blocksize;
5823         hblock = bdev_logical_block_size(bdev);
5824         if (blocksize < hblock) {
5825                 ext4_msg(sb, KERN_ERR,
5826                         "blocksize too small for journal device");
5827                 goto out_bdev;
5828         }
5829
5830         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
5831         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
5832         set_blocksize(bdev, blocksize);
5833         if (!(bh = __bread(bdev, sb_block, blocksize))) {
5834                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
5835                        "external journal");
5836                 goto out_bdev;
5837         }
5838
5839         es = (struct ext4_super_block *) (bh->b_data + offset);
5840         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
5841             !(le32_to_cpu(es->s_feature_incompat) &
5842               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
5843                 ext4_msg(sb, KERN_ERR, "external journal has "
5844                                         "bad superblock");
5845                 brelse(bh);
5846                 goto out_bdev;
5847         }
5848
5849         if ((le32_to_cpu(es->s_feature_ro_compat) &
5850              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
5851             es->s_checksum != ext4_superblock_csum(sb, es)) {
5852                 ext4_msg(sb, KERN_ERR, "external journal has "
5853                                        "corrupt superblock");
5854                 brelse(bh);
5855                 goto out_bdev;
5856         }
5857
5858         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
5859                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
5860                 brelse(bh);
5861                 goto out_bdev;
5862         }
5863
5864         len = ext4_blocks_count(es);
5865         start = sb_block + 1;
5866         brelse(bh);     /* we're done with the superblock */
5867
5868         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
5869                                         start, len, blocksize);
5870         if (!journal) {
5871                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
5872                 goto out_bdev;
5873         }
5874         journal->j_private = sb;
5875         if (ext4_read_bh_lock(journal->j_sb_buffer, REQ_META | REQ_PRIO, true)) {
5876                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
5877                 goto out_journal;
5878         }
5879         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
5880                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
5881                                         "user (unsupported) - %d",
5882                         be32_to_cpu(journal->j_superblock->s_nr_users));
5883                 goto out_journal;
5884         }
5885         EXT4_SB(sb)->s_journal_bdev = bdev;
5886         ext4_init_journal_params(sb, journal);
5887         return journal;
5888
5889 out_journal:
5890         jbd2_journal_destroy(journal);
5891 out_bdev:
5892         ext4_blkdev_put(bdev);
5893         return NULL;
5894 }
5895
5896 static int ext4_load_journal(struct super_block *sb,
5897                              struct ext4_super_block *es,
5898                              unsigned long journal_devnum)
5899 {
5900         journal_t *journal;
5901         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
5902         dev_t journal_dev;
5903         int err = 0;
5904         int really_read_only;
5905         int journal_dev_ro;
5906
5907         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5908                 return -EFSCORRUPTED;
5909
5910         if (journal_devnum &&
5911             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5912                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
5913                         "numbers have changed");
5914                 journal_dev = new_decode_dev(journal_devnum);
5915         } else
5916                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
5917
5918         if (journal_inum && journal_dev) {
5919                 ext4_msg(sb, KERN_ERR,
5920                          "filesystem has both journal inode and journal device!");
5921                 return -EINVAL;
5922         }
5923
5924         if (journal_inum) {
5925                 journal = ext4_get_journal(sb, journal_inum);
5926                 if (!journal)
5927                         return -EINVAL;
5928         } else {
5929                 journal = ext4_get_dev_journal(sb, journal_dev);
5930                 if (!journal)
5931                         return -EINVAL;
5932         }
5933
5934         journal_dev_ro = bdev_read_only(journal->j_dev);
5935         really_read_only = bdev_read_only(sb->s_bdev) | journal_dev_ro;
5936
5937         if (journal_dev_ro && !sb_rdonly(sb)) {
5938                 ext4_msg(sb, KERN_ERR,
5939                          "journal device read-only, try mounting with '-o ro'");
5940                 err = -EROFS;
5941                 goto err_out;
5942         }
5943
5944         /*
5945          * Are we loading a blank journal or performing recovery after a
5946          * crash?  For recovery, we need to check in advance whether we
5947          * can get read-write access to the device.
5948          */
5949         if (ext4_has_feature_journal_needs_recovery(sb)) {
5950                 if (sb_rdonly(sb)) {
5951                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
5952                                         "required on readonly filesystem");
5953                         if (really_read_only) {
5954                                 ext4_msg(sb, KERN_ERR, "write access "
5955                                         "unavailable, cannot proceed "
5956                                         "(try mounting with noload)");
5957                                 err = -EROFS;
5958                                 goto err_out;
5959                         }
5960                         ext4_msg(sb, KERN_INFO, "write access will "
5961                                "be enabled during recovery");
5962                 }
5963         }
5964
5965         if (!(journal->j_flags & JBD2_BARRIER))
5966                 ext4_msg(sb, KERN_INFO, "barriers disabled");
5967
5968         if (!ext4_has_feature_journal_needs_recovery(sb))
5969                 err = jbd2_journal_wipe(journal, !really_read_only);
5970         if (!err) {
5971                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
5972
5973                 if (save)
5974                         memcpy(save, ((char *) es) +
5975                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
5976                 err = jbd2_journal_load(journal);
5977                 if (save)
5978                         memcpy(((char *) es) + EXT4_S_ERR_START,
5979                                save, EXT4_S_ERR_LEN);
5980                 kfree(save);
5981                 es->s_state |= cpu_to_le16(EXT4_SB(sb)->s_mount_state &
5982                                            EXT4_ERROR_FS);
5983                 /* Write out restored error information to the superblock */
5984                 if (!bdev_read_only(sb->s_bdev)) {
5985                         int err2;
5986                         err2 = ext4_commit_super(sb);
5987                         err = err ? : err2;
5988                 }
5989         }
5990
5991         if (err) {
5992                 ext4_msg(sb, KERN_ERR, "error loading journal");
5993                 goto err_out;
5994         }
5995
5996         EXT4_SB(sb)->s_journal = journal;
5997         err = ext4_clear_journal_err(sb, es);
5998         if (err) {
5999                 EXT4_SB(sb)->s_journal = NULL;
6000                 jbd2_journal_destroy(journal);
6001                 return err;
6002         }
6003
6004         if (!really_read_only && journal_devnum &&
6005             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
6006                 es->s_journal_dev = cpu_to_le32(journal_devnum);
6007                 ext4_commit_super(sb);
6008         }
6009         if (!really_read_only && journal_inum &&
6010             journal_inum != le32_to_cpu(es->s_journal_inum)) {
6011                 es->s_journal_inum = cpu_to_le32(journal_inum);
6012                 ext4_commit_super(sb);
6013         }
6014
6015         return 0;
6016
6017 err_out:
6018         jbd2_journal_destroy(journal);
6019         return err;
6020 }
6021
6022 /* Copy state of EXT4_SB(sb) into buffer for on-disk superblock */
6023 static void ext4_update_super(struct super_block *sb)
6024 {
6025         struct ext4_sb_info *sbi = EXT4_SB(sb);
6026         struct ext4_super_block *es = sbi->s_es;
6027         struct buffer_head *sbh = sbi->s_sbh;
6028
6029         lock_buffer(sbh);
6030         /*
6031          * If the file system is mounted read-only, don't update the
6032          * superblock write time.  This avoids updating the superblock
6033          * write time when we are mounting the root file system
6034          * read/only but we need to replay the journal; at that point,
6035          * for people who are east of GMT and who make their clock
6036          * tick in localtime for Windows bug-for-bug compatibility,
6037          * the clock is set in the future, and this will cause e2fsck
6038          * to complain and force a full file system check.
6039          */
6040         if (!(sb->s_flags & SB_RDONLY))
6041                 ext4_update_tstamp(es, s_wtime);
6042         es->s_kbytes_written =
6043                 cpu_to_le64(sbi->s_kbytes_written +
6044                     ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) -
6045                       sbi->s_sectors_written_start) >> 1));
6046         if (percpu_counter_initialized(&sbi->s_freeclusters_counter))
6047                 ext4_free_blocks_count_set(es,
6048                         EXT4_C2B(sbi, percpu_counter_sum_positive(
6049                                 &sbi->s_freeclusters_counter)));
6050         if (percpu_counter_initialized(&sbi->s_freeinodes_counter))
6051                 es->s_free_inodes_count =
6052                         cpu_to_le32(percpu_counter_sum_positive(
6053                                 &sbi->s_freeinodes_counter));
6054         /* Copy error information to the on-disk superblock */
6055         spin_lock(&sbi->s_error_lock);
6056         if (sbi->s_add_error_count > 0) {
6057                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
6058                 if (!es->s_first_error_time && !es->s_first_error_time_hi) {
6059                         __ext4_update_tstamp(&es->s_first_error_time,
6060                                              &es->s_first_error_time_hi,
6061                                              sbi->s_first_error_time);
6062                         strncpy(es->s_first_error_func, sbi->s_first_error_func,
6063                                 sizeof(es->s_first_error_func));
6064                         es->s_first_error_line =
6065                                 cpu_to_le32(sbi->s_first_error_line);
6066                         es->s_first_error_ino =
6067                                 cpu_to_le32(sbi->s_first_error_ino);
6068                         es->s_first_error_block =
6069                                 cpu_to_le64(sbi->s_first_error_block);
6070                         es->s_first_error_errcode =
6071                                 ext4_errno_to_code(sbi->s_first_error_code);
6072                 }
6073                 __ext4_update_tstamp(&es->s_last_error_time,
6074                                      &es->s_last_error_time_hi,
6075                                      sbi->s_last_error_time);
6076                 strncpy(es->s_last_error_func, sbi->s_last_error_func,
6077                         sizeof(es->s_last_error_func));
6078                 es->s_last_error_line = cpu_to_le32(sbi->s_last_error_line);
6079                 es->s_last_error_ino = cpu_to_le32(sbi->s_last_error_ino);
6080                 es->s_last_error_block = cpu_to_le64(sbi->s_last_error_block);
6081                 es->s_last_error_errcode =
6082                                 ext4_errno_to_code(sbi->s_last_error_code);
6083                 /*
6084                  * Start the daily error reporting function if it hasn't been
6085                  * started already
6086                  */
6087                 if (!es->s_error_count)
6088                         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);
6089                 le32_add_cpu(&es->s_error_count, sbi->s_add_error_count);
6090                 sbi->s_add_error_count = 0;
6091         }
6092         spin_unlock(&sbi->s_error_lock);
6093
6094         ext4_superblock_csum_set(sb);
6095         unlock_buffer(sbh);
6096 }
6097
6098 static int ext4_commit_super(struct super_block *sb)
6099 {
6100         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
6101
6102         if (!sbh)
6103                 return -EINVAL;
6104         if (block_device_ejected(sb))
6105                 return -ENODEV;
6106
6107         ext4_update_super(sb);
6108
6109         lock_buffer(sbh);
6110         /* Buffer got discarded which means block device got invalidated */
6111         if (!buffer_mapped(sbh)) {
6112                 unlock_buffer(sbh);
6113                 return -EIO;
6114         }
6115
6116         if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
6117                 /*
6118                  * Oh, dear.  A previous attempt to write the
6119                  * superblock failed.  This could happen because the
6120                  * USB device was yanked out.  Or it could happen to
6121                  * be a transient write error and maybe the block will
6122                  * be remapped.  Nothing we can do but to retry the
6123                  * write and hope for the best.
6124                  */
6125                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
6126                        "superblock detected");
6127                 clear_buffer_write_io_error(sbh);
6128                 set_buffer_uptodate(sbh);
6129         }
6130         get_bh(sbh);
6131         /* Clear potential dirty bit if it was journalled update */
6132         clear_buffer_dirty(sbh);
6133         sbh->b_end_io = end_buffer_write_sync;
6134         submit_bh(REQ_OP_WRITE | REQ_SYNC |
6135                   (test_opt(sb, BARRIER) ? REQ_FUA : 0), sbh);
6136         wait_on_buffer(sbh);
6137         if (buffer_write_io_error(sbh)) {
6138                 ext4_msg(sb, KERN_ERR, "I/O error while writing "
6139                        "superblock");
6140                 clear_buffer_write_io_error(sbh);
6141                 set_buffer_uptodate(sbh);
6142                 return -EIO;
6143         }
6144         return 0;
6145 }
6146
6147 /*
6148  * Have we just finished recovery?  If so, and if we are mounting (or
6149  * remounting) the filesystem readonly, then we will end up with a
6150  * consistent fs on disk.  Record that fact.
6151  */
6152 static int ext4_mark_recovery_complete(struct super_block *sb,
6153                                        struct ext4_super_block *es)
6154 {
6155         int err;
6156         journal_t *journal = EXT4_SB(sb)->s_journal;
6157
6158         if (!ext4_has_feature_journal(sb)) {
6159                 if (journal != NULL) {
6160                         ext4_error(sb, "Journal got removed while the fs was "
6161                                    "mounted!");
6162                         return -EFSCORRUPTED;
6163                 }
6164                 return 0;
6165         }
6166         jbd2_journal_lock_updates(journal);
6167         err = jbd2_journal_flush(journal, 0);
6168         if (err < 0)
6169                 goto out;
6170
6171         if (sb_rdonly(sb) && (ext4_has_feature_journal_needs_recovery(sb) ||
6172             ext4_has_feature_orphan_present(sb))) {
6173                 if (!ext4_orphan_file_empty(sb)) {
6174                         ext4_error(sb, "Orphan file not empty on read-only fs.");
6175                         err = -EFSCORRUPTED;
6176                         goto out;
6177                 }
6178                 ext4_clear_feature_journal_needs_recovery(sb);
6179                 ext4_clear_feature_orphan_present(sb);
6180                 ext4_commit_super(sb);
6181         }
6182 out:
6183         jbd2_journal_unlock_updates(journal);
6184         return err;
6185 }
6186
6187 /*
6188  * If we are mounting (or read-write remounting) a filesystem whose journal
6189  * has recorded an error from a previous lifetime, move that error to the
6190  * main filesystem now.
6191  */
6192 static int ext4_clear_journal_err(struct super_block *sb,
6193                                    struct ext4_super_block *es)
6194 {
6195         journal_t *journal;
6196         int j_errno;
6197         const char *errstr;
6198
6199         if (!ext4_has_feature_journal(sb)) {
6200                 ext4_error(sb, "Journal got removed while the fs was mounted!");
6201                 return -EFSCORRUPTED;
6202         }
6203
6204         journal = EXT4_SB(sb)->s_journal;
6205
6206         /*
6207          * Now check for any error status which may have been recorded in the
6208          * journal by a prior ext4_error() or ext4_abort()
6209          */
6210
6211         j_errno = jbd2_journal_errno(journal);
6212         if (j_errno) {
6213                 char nbuf[16];
6214
6215                 errstr = ext4_decode_error(sb, j_errno, nbuf);
6216                 ext4_warning(sb, "Filesystem error recorded "
6217                              "from previous mount: %s", errstr);
6218
6219                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
6220                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
6221                 j_errno = ext4_commit_super(sb);
6222                 if (j_errno)
6223                         return j_errno;
6224                 ext4_warning(sb, "Marked fs in need of filesystem check.");
6225
6226                 jbd2_journal_clear_err(journal);
6227                 jbd2_journal_update_sb_errno(journal);
6228         }
6229         return 0;
6230 }
6231
6232 /*
6233  * Force the running and committing transactions to commit,
6234  * and wait on the commit.
6235  */
6236 int ext4_force_commit(struct super_block *sb)
6237 {
6238         journal_t *journal;
6239
6240         if (sb_rdonly(sb))
6241                 return 0;
6242
6243         journal = EXT4_SB(sb)->s_journal;
6244         return ext4_journal_force_commit(journal);
6245 }
6246
6247 static int ext4_sync_fs(struct super_block *sb, int wait)
6248 {
6249         int ret = 0;
6250         tid_t target;
6251         bool needs_barrier = false;
6252         struct ext4_sb_info *sbi = EXT4_SB(sb);
6253
6254         if (unlikely(ext4_forced_shutdown(sbi)))
6255                 return 0;
6256
6257         trace_ext4_sync_fs(sb, wait);
6258         flush_workqueue(sbi->rsv_conversion_wq);
6259         /*
6260          * Writeback quota in non-journalled quota case - journalled quota has
6261          * no dirty dquots
6262          */
6263         dquot_writeback_dquots(sb, -1);
6264         /*
6265          * Data writeback is possible w/o journal transaction, so barrier must
6266          * being sent at the end of the function. But we can skip it if
6267          * transaction_commit will do it for us.
6268          */
6269         if (sbi->s_journal) {
6270                 target = jbd2_get_latest_transaction(sbi->s_journal);
6271                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
6272                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
6273                         needs_barrier = true;
6274
6275                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
6276                         if (wait)
6277                                 ret = jbd2_log_wait_commit(sbi->s_journal,
6278                                                            target);
6279                 }
6280         } else if (wait && test_opt(sb, BARRIER))
6281                 needs_barrier = true;
6282         if (needs_barrier) {
6283                 int err;
6284                 err = blkdev_issue_flush(sb->s_bdev);
6285                 if (!ret)
6286                         ret = err;
6287         }
6288
6289         return ret;
6290 }
6291
6292 /*
6293  * LVM calls this function before a (read-only) snapshot is created.  This
6294  * gives us a chance to flush the journal completely and mark the fs clean.
6295  *
6296  * Note that only this function cannot bring a filesystem to be in a clean
6297  * state independently. It relies on upper layer to stop all data & metadata
6298  * modifications.
6299  */
6300 static int ext4_freeze(struct super_block *sb)
6301 {
6302         int error = 0;
6303         journal_t *journal;
6304
6305         if (sb_rdonly(sb))
6306                 return 0;
6307
6308         journal = EXT4_SB(sb)->s_journal;
6309
6310         if (journal) {
6311                 /* Now we set up the journal barrier. */
6312                 jbd2_journal_lock_updates(journal);
6313
6314                 /*
6315                  * Don't clear the needs_recovery flag if we failed to
6316                  * flush the journal.
6317                  */
6318                 error = jbd2_journal_flush(journal, 0);
6319                 if (error < 0)
6320                         goto out;
6321
6322                 /* Journal blocked and flushed, clear needs_recovery flag. */
6323                 ext4_clear_feature_journal_needs_recovery(sb);
6324                 if (ext4_orphan_file_empty(sb))
6325                         ext4_clear_feature_orphan_present(sb);
6326         }
6327
6328         error = ext4_commit_super(sb);
6329 out:
6330         if (journal)
6331                 /* we rely on upper layer to stop further updates */
6332                 jbd2_journal_unlock_updates(journal);
6333         return error;
6334 }
6335
6336 /*
6337  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
6338  * flag here, even though the filesystem is not technically dirty yet.
6339  */
6340 static int ext4_unfreeze(struct super_block *sb)
6341 {
6342         if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
6343                 return 0;
6344
6345         if (EXT4_SB(sb)->s_journal) {
6346                 /* Reset the needs_recovery flag before the fs is unlocked. */
6347                 ext4_set_feature_journal_needs_recovery(sb);
6348                 if (ext4_has_feature_orphan_file(sb))
6349                         ext4_set_feature_orphan_present(sb);
6350         }
6351
6352         ext4_commit_super(sb);
6353         return 0;
6354 }
6355
6356 /*
6357  * Structure to save mount options for ext4_remount's benefit
6358  */
6359 struct ext4_mount_options {
6360         unsigned long s_mount_opt;
6361         unsigned long s_mount_opt2;
6362         kuid_t s_resuid;
6363         kgid_t s_resgid;
6364         unsigned long s_commit_interval;
6365         u32 s_min_batch_time, s_max_batch_time;
6366 #ifdef CONFIG_QUOTA
6367         int s_jquota_fmt;
6368         char *s_qf_names[EXT4_MAXQUOTAS];
6369 #endif
6370 };
6371
6372 static int __ext4_remount(struct fs_context *fc, struct super_block *sb)
6373 {
6374         struct ext4_fs_context *ctx = fc->fs_private;
6375         struct ext4_super_block *es;
6376         struct ext4_sb_info *sbi = EXT4_SB(sb);
6377         unsigned long old_sb_flags;
6378         struct ext4_mount_options old_opts;
6379         ext4_group_t g;
6380         int err = 0;
6381 #ifdef CONFIG_QUOTA
6382         int enable_quota = 0;
6383         int i, j;
6384         char *to_free[EXT4_MAXQUOTAS];
6385 #endif
6386
6387
6388         /* Store the original options */
6389         old_sb_flags = sb->s_flags;
6390         old_opts.s_mount_opt = sbi->s_mount_opt;
6391         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
6392         old_opts.s_resuid = sbi->s_resuid;
6393         old_opts.s_resgid = sbi->s_resgid;
6394         old_opts.s_commit_interval = sbi->s_commit_interval;
6395         old_opts.s_min_batch_time = sbi->s_min_batch_time;
6396         old_opts.s_max_batch_time = sbi->s_max_batch_time;
6397 #ifdef CONFIG_QUOTA
6398         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
6399         for (i = 0; i < EXT4_MAXQUOTAS; i++)
6400                 if (sbi->s_qf_names[i]) {
6401                         char *qf_name = get_qf_name(sb, sbi, i);
6402
6403                         old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL);
6404                         if (!old_opts.s_qf_names[i]) {
6405                                 for (j = 0; j < i; j++)
6406                                         kfree(old_opts.s_qf_names[j]);
6407                                 return -ENOMEM;
6408                         }
6409                 } else
6410                         old_opts.s_qf_names[i] = NULL;
6411 #endif
6412         if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)) {
6413                 if (sbi->s_journal && sbi->s_journal->j_task->io_context)
6414                         ctx->journal_ioprio =
6415                                 sbi->s_journal->j_task->io_context->ioprio;
6416                 else
6417                         ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
6418
6419         }
6420
6421         ext4_apply_options(fc, sb);
6422
6423         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
6424             test_opt(sb, JOURNAL_CHECKSUM)) {
6425                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
6426                          "during remount not supported; ignoring");
6427                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
6428         }
6429
6430         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
6431                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
6432                         ext4_msg(sb, KERN_ERR, "can't mount with "
6433                                  "both data=journal and delalloc");
6434                         err = -EINVAL;
6435                         goto restore_opts;
6436                 }
6437                 if (test_opt(sb, DIOREAD_NOLOCK)) {
6438                         ext4_msg(sb, KERN_ERR, "can't mount with "
6439                                  "both data=journal and dioread_nolock");
6440                         err = -EINVAL;
6441                         goto restore_opts;
6442                 }
6443         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
6444                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
6445                         ext4_msg(sb, KERN_ERR, "can't mount with "
6446                                 "journal_async_commit in data=ordered mode");
6447                         err = -EINVAL;
6448                         goto restore_opts;
6449                 }
6450         }
6451
6452         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
6453                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
6454                 err = -EINVAL;
6455                 goto restore_opts;
6456         }
6457
6458         if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED))
6459                 ext4_abort(sb, ESHUTDOWN, "Abort forced by user");
6460
6461         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
6462                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
6463
6464         es = sbi->s_es;
6465
6466         if (sbi->s_journal) {
6467                 ext4_init_journal_params(sb, sbi->s_journal);
6468                 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio);
6469         }
6470
6471         /* Flush outstanding errors before changing fs state */
6472         flush_work(&sbi->s_error_work);
6473
6474         if ((bool)(fc->sb_flags & SB_RDONLY) != sb_rdonly(sb)) {
6475                 if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED)) {
6476                         err = -EROFS;
6477                         goto restore_opts;
6478                 }
6479
6480                 if (fc->sb_flags & SB_RDONLY) {
6481                         err = sync_filesystem(sb);
6482                         if (err < 0)
6483                                 goto restore_opts;
6484                         err = dquot_suspend(sb, -1);
6485                         if (err < 0)
6486                                 goto restore_opts;
6487
6488                         /*
6489                          * First of all, the unconditional stuff we have to do
6490                          * to disable replay of the journal when we next remount
6491                          */
6492                         sb->s_flags |= SB_RDONLY;
6493
6494                         /*
6495                          * OK, test if we are remounting a valid rw partition
6496                          * readonly, and if so set the rdonly flag and then
6497                          * mark the partition as valid again.
6498                          */
6499                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
6500                             (sbi->s_mount_state & EXT4_VALID_FS))
6501                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
6502
6503                         if (sbi->s_journal) {
6504                                 /*
6505                                  * We let remount-ro finish even if marking fs
6506                                  * as clean failed...
6507                                  */
6508                                 ext4_mark_recovery_complete(sb, es);
6509                         }
6510                 } else {
6511                         /* Make sure we can mount this feature set readwrite */
6512                         if (ext4_has_feature_readonly(sb) ||
6513                             !ext4_feature_set_ok(sb, 0)) {
6514                                 err = -EROFS;
6515                                 goto restore_opts;
6516                         }
6517                         /*
6518                          * Make sure the group descriptor checksums
6519                          * are sane.  If they aren't, refuse to remount r/w.
6520                          */
6521                         for (g = 0; g < sbi->s_groups_count; g++) {
6522                                 struct ext4_group_desc *gdp =
6523                                         ext4_get_group_desc(sb, g, NULL);
6524
6525                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
6526                                         ext4_msg(sb, KERN_ERR,
6527                "ext4_remount: Checksum for group %u failed (%u!=%u)",
6528                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
6529                                                le16_to_cpu(gdp->bg_checksum));
6530                                         err = -EFSBADCRC;
6531                                         goto restore_opts;
6532                                 }
6533                         }
6534
6535                         /*
6536                          * If we have an unprocessed orphan list hanging
6537                          * around from a previously readonly bdev mount,
6538                          * require a full umount/remount for now.
6539                          */
6540                         if (es->s_last_orphan || !ext4_orphan_file_empty(sb)) {
6541                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
6542                                        "remount RDWR because of unprocessed "
6543                                        "orphan inode list.  Please "
6544                                        "umount/remount instead");
6545                                 err = -EINVAL;
6546                                 goto restore_opts;
6547                         }
6548
6549                         /*
6550                          * Mounting a RDONLY partition read-write, so reread
6551                          * and store the current valid flag.  (It may have
6552                          * been changed by e2fsck since we originally mounted
6553                          * the partition.)
6554                          */
6555                         if (sbi->s_journal) {
6556                                 err = ext4_clear_journal_err(sb, es);
6557                                 if (err)
6558                                         goto restore_opts;
6559                         }
6560                         sbi->s_mount_state = (le16_to_cpu(es->s_state) &
6561                                               ~EXT4_FC_REPLAY);
6562
6563                         err = ext4_setup_super(sb, es, 0);
6564                         if (err)
6565                                 goto restore_opts;
6566
6567                         sb->s_flags &= ~SB_RDONLY;
6568                         if (ext4_has_feature_mmp(sb))
6569                                 if (ext4_multi_mount_protect(sb,
6570                                                 le64_to_cpu(es->s_mmp_block))) {
6571                                         err = -EROFS;
6572                                         goto restore_opts;
6573                                 }
6574 #ifdef CONFIG_QUOTA
6575                         enable_quota = 1;
6576 #endif
6577                 }
6578         }
6579
6580         /*
6581          * Reinitialize lazy itable initialization thread based on
6582          * current settings
6583          */
6584         if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
6585                 ext4_unregister_li_request(sb);
6586         else {
6587                 ext4_group_t first_not_zeroed;
6588                 first_not_zeroed = ext4_has_uninit_itable(sb);
6589                 ext4_register_li_request(sb, first_not_zeroed);
6590         }
6591
6592         /*
6593          * Handle creation of system zone data early because it can fail.
6594          * Releasing of existing data is done when we are sure remount will
6595          * succeed.
6596          */
6597         if (test_opt(sb, BLOCK_VALIDITY) && !sbi->s_system_blks) {
6598                 err = ext4_setup_system_zone(sb);
6599                 if (err)
6600                         goto restore_opts;
6601         }
6602
6603         if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) {
6604                 err = ext4_commit_super(sb);
6605                 if (err)
6606                         goto restore_opts;
6607         }
6608
6609 #ifdef CONFIG_QUOTA
6610         /* Release old quota file names */
6611         for (i = 0; i < EXT4_MAXQUOTAS; i++)
6612                 kfree(old_opts.s_qf_names[i]);
6613         if (enable_quota) {
6614                 if (sb_any_quota_suspended(sb))
6615                         dquot_resume(sb, -1);
6616                 else if (ext4_has_feature_quota(sb)) {
6617                         err = ext4_enable_quotas(sb);
6618                         if (err)
6619                                 goto restore_opts;
6620                 }
6621         }
6622 #endif
6623         if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks)
6624                 ext4_release_system_zone(sb);
6625
6626         if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb))
6627                 ext4_stop_mmpd(sbi);
6628
6629         return 0;
6630
6631 restore_opts:
6632         sb->s_flags = old_sb_flags;
6633         sbi->s_mount_opt = old_opts.s_mount_opt;
6634         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
6635         sbi->s_resuid = old_opts.s_resuid;
6636         sbi->s_resgid = old_opts.s_resgid;
6637         sbi->s_commit_interval = old_opts.s_commit_interval;
6638         sbi->s_min_batch_time = old_opts.s_min_batch_time;
6639         sbi->s_max_batch_time = old_opts.s_max_batch_time;
6640         if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks)
6641                 ext4_release_system_zone(sb);
6642 #ifdef CONFIG_QUOTA
6643         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
6644         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
6645                 to_free[i] = get_qf_name(sb, sbi, i);
6646                 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]);
6647         }
6648         synchronize_rcu();
6649         for (i = 0; i < EXT4_MAXQUOTAS; i++)
6650                 kfree(to_free[i]);
6651 #endif
6652         if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb))
6653                 ext4_stop_mmpd(sbi);
6654         return err;
6655 }
6656
6657 static int ext4_reconfigure(struct fs_context *fc)
6658 {
6659         struct super_block *sb = fc->root->d_sb;
6660         int ret;
6661
6662         fc->s_fs_info = EXT4_SB(sb);
6663
6664         ret = ext4_check_opt_consistency(fc, sb);
6665         if (ret < 0)
6666                 return ret;
6667
6668         ret = __ext4_remount(fc, sb);
6669         if (ret < 0)
6670                 return ret;
6671
6672         ext4_msg(sb, KERN_INFO, "re-mounted %pU. Quota mode: %s.",
6673                  &sb->s_uuid, ext4_quota_mode(sb));
6674
6675         return 0;
6676 }
6677
6678 #ifdef CONFIG_QUOTA
6679 static int ext4_statfs_project(struct super_block *sb,
6680                                kprojid_t projid, struct kstatfs *buf)
6681 {
6682         struct kqid qid;
6683         struct dquot *dquot;
6684         u64 limit;
6685         u64 curblock;
6686
6687         qid = make_kqid_projid(projid);
6688         dquot = dqget(sb, qid);
6689         if (IS_ERR(dquot))
6690                 return PTR_ERR(dquot);
6691         spin_lock(&dquot->dq_dqb_lock);
6692
6693         limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit,
6694                              dquot->dq_dqb.dqb_bhardlimit);
6695         limit >>= sb->s_blocksize_bits;
6696
6697         if (limit && buf->f_blocks > limit) {
6698                 curblock = (dquot->dq_dqb.dqb_curspace +
6699                             dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits;
6700                 buf->f_blocks = limit;
6701                 buf->f_bfree = buf->f_bavail =
6702                         (buf->f_blocks > curblock) ?
6703                          (buf->f_blocks - curblock) : 0;
6704         }
6705
6706         limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit,
6707                              dquot->dq_dqb.dqb_ihardlimit);
6708         if (limit && buf->f_files > limit) {
6709                 buf->f_files = limit;
6710                 buf->f_ffree =
6711                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
6712                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
6713         }
6714
6715         spin_unlock(&dquot->dq_dqb_lock);
6716         dqput(dquot);
6717         return 0;
6718 }
6719 #endif
6720
6721 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
6722 {
6723         struct super_block *sb = dentry->d_sb;
6724         struct ext4_sb_info *sbi = EXT4_SB(sb);
6725         struct ext4_super_block *es = sbi->s_es;
6726         ext4_fsblk_t overhead = 0, resv_blocks;
6727         s64 bfree;
6728         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
6729
6730         if (!test_opt(sb, MINIX_DF))
6731                 overhead = sbi->s_overhead;
6732
6733         buf->f_type = EXT4_SUPER_MAGIC;
6734         buf->f_bsize = sb->s_blocksize;
6735         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
6736         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
6737                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
6738         /* prevent underflow in case that few free space is available */
6739         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
6740         buf->f_bavail = buf->f_bfree -
6741                         (ext4_r_blocks_count(es) + resv_blocks);
6742         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
6743                 buf->f_bavail = 0;
6744         buf->f_files = le32_to_cpu(es->s_inodes_count);
6745         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
6746         buf->f_namelen = EXT4_NAME_LEN;
6747         buf->f_fsid = uuid_to_fsid(es->s_uuid);
6748
6749 #ifdef CONFIG_QUOTA
6750         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
6751             sb_has_quota_limits_enabled(sb, PRJQUOTA))
6752                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
6753 #endif
6754         return 0;
6755 }
6756
6757
6758 #ifdef CONFIG_QUOTA
6759
6760 /*
6761  * Helper functions so that transaction is started before we acquire dqio_sem
6762  * to keep correct lock ordering of transaction > dqio_sem
6763  */
6764 static inline struct inode *dquot_to_inode(struct dquot *dquot)
6765 {
6766         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
6767 }
6768
6769 static int ext4_write_dquot(struct dquot *dquot)
6770 {
6771         int ret, err;
6772         handle_t *handle;
6773         struct inode *inode;
6774
6775         inode = dquot_to_inode(dquot);
6776         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
6777                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
6778         if (IS_ERR(handle))
6779                 return PTR_ERR(handle);
6780         ret = dquot_commit(dquot);
6781         err = ext4_journal_stop(handle);
6782         if (!ret)
6783                 ret = err;
6784         return ret;
6785 }
6786
6787 static int ext4_acquire_dquot(struct dquot *dquot)
6788 {
6789         int ret, err;
6790         handle_t *handle;
6791
6792         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
6793                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
6794         if (IS_ERR(handle))
6795                 return PTR_ERR(handle);
6796         ret = dquot_acquire(dquot);
6797         err = ext4_journal_stop(handle);
6798         if (!ret)
6799                 ret = err;
6800         return ret;
6801 }
6802
6803 static int ext4_release_dquot(struct dquot *dquot)
6804 {
6805         int ret, err;
6806         handle_t *handle;
6807
6808         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
6809                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
6810         if (IS_ERR(handle)) {
6811                 /* Release dquot anyway to avoid endless cycle in dqput() */
6812                 dquot_release(dquot);
6813                 return PTR_ERR(handle);
6814         }
6815         ret = dquot_release(dquot);
6816         err = ext4_journal_stop(handle);
6817         if (!ret)
6818                 ret = err;
6819         return ret;
6820 }
6821
6822 static int ext4_mark_dquot_dirty(struct dquot *dquot)
6823 {
6824         struct super_block *sb = dquot->dq_sb;
6825
6826         if (ext4_is_quota_journalled(sb)) {
6827                 dquot_mark_dquot_dirty(dquot);
6828                 return ext4_write_dquot(dquot);
6829         } else {
6830                 return dquot_mark_dquot_dirty(dquot);
6831         }
6832 }
6833
6834 static int ext4_write_info(struct super_block *sb, int type)
6835 {
6836         int ret, err;
6837         handle_t *handle;
6838
6839         /* Data block + inode block */
6840         handle = ext4_journal_start_sb(sb, EXT4_HT_QUOTA, 2);
6841         if (IS_ERR(handle))
6842                 return PTR_ERR(handle);
6843         ret = dquot_commit_info(sb, type);
6844         err = ext4_journal_stop(handle);
6845         if (!ret)
6846                 ret = err;
6847         return ret;
6848 }
6849
6850 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
6851 {
6852         struct ext4_inode_info *ei = EXT4_I(inode);
6853
6854         /* The first argument of lockdep_set_subclass has to be
6855          * *exactly* the same as the argument to init_rwsem() --- in
6856          * this case, in init_once() --- or lockdep gets unhappy
6857          * because the name of the lock is set using the
6858          * stringification of the argument to init_rwsem().
6859          */
6860         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
6861         lockdep_set_subclass(&ei->i_data_sem, subclass);
6862 }
6863
6864 /*
6865  * Standard function to be called on quota_on
6866  */
6867 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
6868                          const struct path *path)
6869 {
6870         int err;
6871
6872         if (!test_opt(sb, QUOTA))
6873                 return -EINVAL;
6874
6875         /* Quotafile not on the same filesystem? */
6876         if (path->dentry->d_sb != sb)
6877                 return -EXDEV;
6878
6879         /* Quota already enabled for this file? */
6880         if (IS_NOQUOTA(d_inode(path->dentry)))
6881                 return -EBUSY;
6882
6883         /* Journaling quota? */
6884         if (EXT4_SB(sb)->s_qf_names[type]) {
6885                 /* Quotafile not in fs root? */
6886                 if (path->dentry->d_parent != sb->s_root)
6887                         ext4_msg(sb, KERN_WARNING,
6888                                 "Quota file not on filesystem root. "
6889                                 "Journaled quota will not work");
6890                 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
6891         } else {
6892                 /*
6893                  * Clear the flag just in case mount options changed since
6894                  * last time.
6895                  */
6896                 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
6897         }
6898
6899         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
6900         err = dquot_quota_on(sb, type, format_id, path);
6901         if (!err) {
6902                 struct inode *inode = d_inode(path->dentry);
6903                 handle_t *handle;
6904
6905                 /*
6906                  * Set inode flags to prevent userspace from messing with quota
6907                  * files. If this fails, we return success anyway since quotas
6908                  * are already enabled and this is not a hard failure.
6909                  */
6910                 inode_lock(inode);
6911                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
6912                 if (IS_ERR(handle))
6913                         goto unlock_inode;
6914                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
6915                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
6916                                 S_NOATIME | S_IMMUTABLE);
6917                 err = ext4_mark_inode_dirty(handle, inode);
6918                 ext4_journal_stop(handle);
6919         unlock_inode:
6920                 inode_unlock(inode);
6921                 if (err)
6922                         dquot_quota_off(sb, type);
6923         }
6924         if (err)
6925                 lockdep_set_quota_inode(path->dentry->d_inode,
6926                                              I_DATA_SEM_NORMAL);
6927         return err;
6928 }
6929
6930 static inline bool ext4_check_quota_inum(int type, unsigned long qf_inum)
6931 {
6932         switch (type) {
6933         case USRQUOTA:
6934                 return qf_inum == EXT4_USR_QUOTA_INO;
6935         case GRPQUOTA:
6936                 return qf_inum == EXT4_GRP_QUOTA_INO;
6937         case PRJQUOTA:
6938                 return qf_inum >= EXT4_GOOD_OLD_FIRST_INO;
6939         default:
6940                 BUG();
6941         }
6942 }
6943
6944 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
6945                              unsigned int flags)
6946 {
6947         int err;
6948         struct inode *qf_inode;
6949         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
6950                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
6951                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
6952                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
6953         };
6954
6955         BUG_ON(!ext4_has_feature_quota(sb));
6956
6957         if (!qf_inums[type])
6958                 return -EPERM;
6959
6960         if (!ext4_check_quota_inum(type, qf_inums[type])) {
6961                 ext4_error(sb, "Bad quota inum: %lu, type: %d",
6962                                 qf_inums[type], type);
6963                 return -EUCLEAN;
6964         }
6965
6966         qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL);
6967         if (IS_ERR(qf_inode)) {
6968                 ext4_error(sb, "Bad quota inode: %lu, type: %d",
6969                                 qf_inums[type], type);
6970                 return PTR_ERR(qf_inode);
6971         }
6972
6973         /* Don't account quota for quota files to avoid recursion */
6974         qf_inode->i_flags |= S_NOQUOTA;
6975         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
6976         err = dquot_load_quota_inode(qf_inode, type, format_id, flags);
6977         if (err)
6978                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
6979         iput(qf_inode);
6980
6981         return err;
6982 }
6983
6984 /* Enable usage tracking for all quota types. */
6985 int ext4_enable_quotas(struct super_block *sb)
6986 {
6987         int type, err = 0;
6988         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
6989                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
6990                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
6991                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
6992         };
6993         bool quota_mopt[EXT4_MAXQUOTAS] = {
6994                 test_opt(sb, USRQUOTA),
6995                 test_opt(sb, GRPQUOTA),
6996                 test_opt(sb, PRJQUOTA),
6997         };
6998
6999         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
7000         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
7001                 if (qf_inums[type]) {
7002                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
7003                                 DQUOT_USAGE_ENABLED |
7004                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
7005                         if (err) {
7006                                 ext4_warning(sb,
7007                                         "Failed to enable quota tracking "
7008                                         "(type=%d, err=%d, ino=%lu). "
7009                                         "Please run e2fsck to fix.", type,
7010                                         err, qf_inums[type]);
7011                                 for (type--; type >= 0; type--) {
7012                                         struct inode *inode;
7013
7014                                         inode = sb_dqopt(sb)->files[type];
7015                                         if (inode)
7016                                                 inode = igrab(inode);
7017                                         dquot_quota_off(sb, type);
7018                                         if (inode) {
7019                                                 lockdep_set_quota_inode(inode,
7020                                                         I_DATA_SEM_NORMAL);
7021                                                 iput(inode);
7022                                         }
7023                                 }
7024
7025                                 return err;
7026                         }
7027                 }
7028         }
7029         return 0;
7030 }
7031
7032 static int ext4_quota_off(struct super_block *sb, int type)
7033 {
7034         struct inode *inode = sb_dqopt(sb)->files[type];
7035         handle_t *handle;
7036         int err;
7037
7038         /* Force all delayed allocation blocks to be allocated.
7039          * Caller already holds s_umount sem */
7040         if (test_opt(sb, DELALLOC))
7041                 sync_filesystem(sb);
7042
7043         if (!inode || !igrab(inode))
7044                 goto out;
7045
7046         err = dquot_quota_off(sb, type);
7047         if (err || ext4_has_feature_quota(sb))
7048                 goto out_put;
7049
7050         inode_lock(inode);
7051         /*
7052          * Update modification times of quota files when userspace can
7053          * start looking at them. If we fail, we return success anyway since
7054          * this is not a hard failure and quotas are already disabled.
7055          */
7056         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
7057         if (IS_ERR(handle)) {
7058                 err = PTR_ERR(handle);
7059                 goto out_unlock;
7060         }
7061         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
7062         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
7063         inode->i_mtime = inode->i_ctime = current_time(inode);
7064         err = ext4_mark_inode_dirty(handle, inode);
7065         ext4_journal_stop(handle);
7066 out_unlock:
7067         inode_unlock(inode);
7068 out_put:
7069         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
7070         iput(inode);
7071         return err;
7072 out:
7073         return dquot_quota_off(sb, type);
7074 }
7075
7076 /* Read data from quotafile - avoid pagecache and such because we cannot afford
7077  * acquiring the locks... As quota files are never truncated and quota code
7078  * itself serializes the operations (and no one else should touch the files)
7079  * we don't have to be afraid of races */
7080 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
7081                                size_t len, loff_t off)
7082 {
7083         struct inode *inode = sb_dqopt(sb)->files[type];
7084         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
7085         int offset = off & (sb->s_blocksize - 1);
7086         int tocopy;
7087         size_t toread;
7088         struct buffer_head *bh;
7089         loff_t i_size = i_size_read(inode);
7090
7091         if (off > i_size)
7092                 return 0;
7093         if (off+len > i_size)
7094                 len = i_size-off;
7095         toread = len;
7096         while (toread > 0) {
7097                 tocopy = min_t(unsigned long, sb->s_blocksize - offset, toread);
7098                 bh = ext4_bread(NULL, inode, blk, 0);
7099                 if (IS_ERR(bh))
7100                         return PTR_ERR(bh);
7101                 if (!bh)        /* A hole? */
7102                         memset(data, 0, tocopy);
7103                 else
7104                         memcpy(data, bh->b_data+offset, tocopy);
7105                 brelse(bh);
7106                 offset = 0;
7107                 toread -= tocopy;
7108                 data += tocopy;
7109                 blk++;
7110         }
7111         return len;
7112 }
7113
7114 /* Write to quotafile (we know the transaction is already started and has
7115  * enough credits) */
7116 static ssize_t ext4_quota_write(struct super_block *sb, int type,
7117                                 const char *data, size_t len, loff_t off)
7118 {
7119         struct inode *inode = sb_dqopt(sb)->files[type];
7120         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
7121         int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1);
7122         int retries = 0;
7123         struct buffer_head *bh;
7124         handle_t *handle = journal_current_handle();
7125
7126         if (!handle) {
7127                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
7128                         " cancelled because transaction is not started",
7129                         (unsigned long long)off, (unsigned long long)len);
7130                 return -EIO;
7131         }
7132         /*
7133          * Since we account only one data block in transaction credits,
7134          * then it is impossible to cross a block boundary.
7135          */
7136         if (sb->s_blocksize - offset < len) {
7137                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
7138                         " cancelled because not block aligned",
7139                         (unsigned long long)off, (unsigned long long)len);
7140                 return -EIO;
7141         }
7142
7143         do {
7144                 bh = ext4_bread(handle, inode, blk,
7145                                 EXT4_GET_BLOCKS_CREATE |
7146                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
7147         } while (PTR_ERR(bh) == -ENOSPC &&
7148                  ext4_should_retry_alloc(inode->i_sb, &retries));
7149         if (IS_ERR(bh))
7150                 return PTR_ERR(bh);
7151         if (!bh)
7152                 goto out;
7153         BUFFER_TRACE(bh, "get write access");
7154         err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
7155         if (err) {
7156                 brelse(bh);
7157                 return err;
7158         }
7159         lock_buffer(bh);
7160         memcpy(bh->b_data+offset, data, len);
7161         flush_dcache_page(bh->b_page);
7162         unlock_buffer(bh);
7163         err = ext4_handle_dirty_metadata(handle, NULL, bh);
7164         brelse(bh);
7165 out:
7166         if (inode->i_size < off + len) {
7167                 i_size_write(inode, off + len);
7168                 EXT4_I(inode)->i_disksize = inode->i_size;
7169                 err2 = ext4_mark_inode_dirty(handle, inode);
7170                 if (unlikely(err2 && !err))
7171                         err = err2;
7172         }
7173         return err ? err : len;
7174 }
7175 #endif
7176
7177 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
7178 static inline void register_as_ext2(void)
7179 {
7180         int err = register_filesystem(&ext2_fs_type);
7181         if (err)
7182                 printk(KERN_WARNING
7183                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
7184 }
7185
7186 static inline void unregister_as_ext2(void)
7187 {
7188         unregister_filesystem(&ext2_fs_type);
7189 }
7190
7191 static inline int ext2_feature_set_ok(struct super_block *sb)
7192 {
7193         if (ext4_has_unknown_ext2_incompat_features(sb))
7194                 return 0;
7195         if (sb_rdonly(sb))
7196                 return 1;
7197         if (ext4_has_unknown_ext2_ro_compat_features(sb))
7198                 return 0;
7199         return 1;
7200 }
7201 #else
7202 static inline void register_as_ext2(void) { }
7203 static inline void unregister_as_ext2(void) { }
7204 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
7205 #endif
7206
7207 static inline void register_as_ext3(void)
7208 {
7209         int err = register_filesystem(&ext3_fs_type);
7210         if (err)
7211                 printk(KERN_WARNING
7212                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
7213 }
7214
7215 static inline void unregister_as_ext3(void)
7216 {
7217         unregister_filesystem(&ext3_fs_type);
7218 }
7219
7220 static inline int ext3_feature_set_ok(struct super_block *sb)
7221 {
7222         if (ext4_has_unknown_ext3_incompat_features(sb))
7223                 return 0;
7224         if (!ext4_has_feature_journal(sb))
7225                 return 0;
7226         if (sb_rdonly(sb))
7227                 return 1;
7228         if (ext4_has_unknown_ext3_ro_compat_features(sb))
7229                 return 0;
7230         return 1;
7231 }
7232
7233 static struct file_system_type ext4_fs_type = {
7234         .owner                  = THIS_MODULE,
7235         .name                   = "ext4",
7236         .init_fs_context        = ext4_init_fs_context,
7237         .parameters             = ext4_param_specs,
7238         .kill_sb                = kill_block_super,
7239         .fs_flags               = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
7240 };
7241 MODULE_ALIAS_FS("ext4");
7242
7243 /* Shared across all ext4 file systems */
7244 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
7245
7246 static int __init ext4_init_fs(void)
7247 {
7248         int i, err;
7249
7250         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
7251         ext4_li_info = NULL;
7252
7253         /* Build-time check for flags consistency */
7254         ext4_check_flag_values();
7255
7256         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
7257                 init_waitqueue_head(&ext4__ioend_wq[i]);
7258
7259         err = ext4_init_es();
7260         if (err)
7261                 return err;
7262
7263         err = ext4_init_pending();
7264         if (err)
7265                 goto out7;
7266
7267         err = ext4_init_post_read_processing();
7268         if (err)
7269                 goto out6;
7270
7271         err = ext4_init_pageio();
7272         if (err)
7273                 goto out5;
7274
7275         err = ext4_init_system_zone();
7276         if (err)
7277                 goto out4;
7278
7279         err = ext4_init_sysfs();
7280         if (err)
7281                 goto out3;
7282
7283         err = ext4_init_mballoc();
7284         if (err)
7285                 goto out2;
7286         err = init_inodecache();
7287         if (err)
7288                 goto out1;
7289
7290         err = ext4_fc_init_dentry_cache();
7291         if (err)
7292                 goto out05;
7293
7294         register_as_ext3();
7295         register_as_ext2();
7296         err = register_filesystem(&ext4_fs_type);
7297         if (err)
7298                 goto out;
7299
7300         return 0;
7301 out:
7302         unregister_as_ext2();
7303         unregister_as_ext3();
7304         ext4_fc_destroy_dentry_cache();
7305 out05:
7306         destroy_inodecache();
7307 out1:
7308         ext4_exit_mballoc();
7309 out2:
7310         ext4_exit_sysfs();
7311 out3:
7312         ext4_exit_system_zone();
7313 out4:
7314         ext4_exit_pageio();
7315 out5:
7316         ext4_exit_post_read_processing();
7317 out6:
7318         ext4_exit_pending();
7319 out7:
7320         ext4_exit_es();
7321
7322         return err;
7323 }
7324
7325 static void __exit ext4_exit_fs(void)
7326 {
7327         ext4_destroy_lazyinit_thread();
7328         unregister_as_ext2();
7329         unregister_as_ext3();
7330         unregister_filesystem(&ext4_fs_type);
7331         ext4_fc_destroy_dentry_cache();
7332         destroy_inodecache();
7333         ext4_exit_mballoc();
7334         ext4_exit_sysfs();
7335         ext4_exit_system_zone();
7336         ext4_exit_pageio();
7337         ext4_exit_post_read_processing();
7338         ext4_exit_es();
7339         ext4_exit_pending();
7340 }
7341
7342 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
7343 MODULE_DESCRIPTION("Fourth Extended Filesystem");
7344 MODULE_LICENSE("GPL");
7345 MODULE_SOFTDEP("pre: crc32c");
7346 module_init(ext4_init_fs)
7347 module_exit(ext4_exit_fs)