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 page *page,
 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 = page_buffers(page);
 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                         redirty_page_for_writepage(wbc, page);
 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 = jbd2_journal_submit_inode_data_buffers(jinode);
 544
 545         return ret;
 546 }
 547
 548 static int ext4_journal_finish_inode_data_buffers(struct jbd2_inode *jinode)
 549 {
 550         int ret = 0;
 551
 552         if (!ext4_should_journal_data(jinode->i_vfs_inode))
 553                 ret = jbd2_journal_finish_inode_data_buffers(jinode);
 554
 555         return ret;
 556 }
 557
 558 static bool system_going_down(void)
 559 {
 560         return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF
 561                 || system_state == SYSTEM_RESTART;
 562 }
 563
 564 struct ext4_err_translation {
 565         int code;
 566         int errno;
 567 };
 568
 569 #define EXT4_ERR_TRANSLATE(err) { .code = EXT4_ERR_##err, .errno = err }
 570
 571 static struct ext4_err_translation err_translation[] = {
 572         EXT4_ERR_TRANSLATE(EIO),
 573         EXT4_ERR_TRANSLATE(ENOMEM),
 574         EXT4_ERR_TRANSLATE(EFSBADCRC),
 575         EXT4_ERR_TRANSLATE(EFSCORRUPTED),
 576         EXT4_ERR_TRANSLATE(ENOSPC),
 577         EXT4_ERR_TRANSLATE(ENOKEY),
 578         EXT4_ERR_TRANSLATE(EROFS),
 579         EXT4_ERR_TRANSLATE(EFBIG),
 580         EXT4_ERR_TRANSLATE(EEXIST),
 581         EXT4_ERR_TRANSLATE(ERANGE),
 582         EXT4_ERR_TRANSLATE(EOVERFLOW),
 583         EXT4_ERR_TRANSLATE(EBUSY),
 584         EXT4_ERR_TRANSLATE(ENOTDIR),
 585         EXT4_ERR_TRANSLATE(ENOTEMPTY),
 586         EXT4_ERR_TRANSLATE(ESHUTDOWN),
 587         EXT4_ERR_TRANSLATE(EFAULT),
 588 };
 589
 590 static int ext4_errno_to_code(int errno)
 591 {
 592         int i;
 593
 594         for (i = 0; i < ARRAY_SIZE(err_translation); i++)
 595                 if (err_translation[i].errno == errno)
 596                         return err_translation[i].code;
 597         return EXT4_ERR_UNKNOWN;
 598 }
 599
 600 static void save_error_info(struct super_block *sb, int error,
 601                             __u32 ino, __u64 block,
 602                             const char *func, unsigned int line)
 603 {
 604         struct ext4_sb_info *sbi = EXT4_SB(sb);
 605
 606         /* We default to EFSCORRUPTED error... */
 607         if (error == 0)
 608                 error = EFSCORRUPTED;
 609
 610         spin_lock(&sbi->s_error_lock);
 611         sbi->s_add_error_count++;
 612         sbi->s_last_error_code = error;
 613         sbi->s_last_error_line = line;
 614         sbi->s_last_error_ino = ino;
 615         sbi->s_last_error_block = block;
 616         sbi->s_last_error_func = func;
 617         sbi->s_last_error_time = ktime_get_real_seconds();
 618         if (!sbi->s_first_error_time) {
 619                 sbi->s_first_error_code = error;
 620                 sbi->s_first_error_line = line;
 621                 sbi->s_first_error_ino = ino;
 622                 sbi->s_first_error_block = block;
 623                 sbi->s_first_error_func = func;
 624                 sbi->s_first_error_time = sbi->s_last_error_time;
 625         }
 626         spin_unlock(&sbi->s_error_lock);
 627 }
 628
 629 /* Deal with the reporting of failure conditions on a filesystem such as
 630  * inconsistencies detected or read IO failures.
 631  *
 632  * On ext2, we can store the error state of the filesystem in the
 633  * superblock.  That is not possible on ext4, because we may have other
 634  * write ordering constraints on the superblock which prevent us from
 635  * writing it out straight away; and given that the journal is about to
 636  * be aborted, we can't rely on the current, or future, transactions to
 637  * write out the superblock safely.
 638  *
 639  * We'll just use the jbd2_journal_abort() error code to record an error in
 640  * the journal instead.  On recovery, the journal will complain about
 641  * that error until we've noted it down and cleared it.
 642  *
 643  * If force_ro is set, we unconditionally force the filesystem into an
 644  * ABORT|READONLY state, unless the error response on the fs has been set to
 645  * panic in which case we take the easy way out and panic immediately. This is
 646  * used to deal with unrecoverable failures such as journal IO errors or ENOMEM
 647  * at a critical moment in log management.
 648  */
 649 static void ext4_handle_error(struct super_block *sb, bool force_ro, int error,
 650                               __u32 ino, __u64 block,
 651                               const char *func, unsigned int line)
 652 {
 653         journal_t *journal = EXT4_SB(sb)->s_journal;
 654         bool continue_fs = !force_ro && test_opt(sb, ERRORS_CONT);
 655
 656         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 657         if (test_opt(sb, WARN_ON_ERROR))
 658                 WARN_ON_ONCE(1);
 659
 660         if (!continue_fs && !sb_rdonly(sb)) {
 661                 ext4_set_mount_flag(sb, EXT4_MF_FS_ABORTED);
 662                 if (journal)
 663                         jbd2_journal_abort(journal, -EIO);
 664         }
 665
 666         if (!bdev_read_only(sb->s_bdev)) {
 667                 save_error_info(sb, error, ino, block, func, line);
 668                 /*
 669                  * In case the fs should keep running, we need to writeout
 670                  * superblock through the journal. Due to lock ordering
 671                  * constraints, it may not be safe to do it right here so we
 672                  * defer superblock flushing to a workqueue.
 673                  */
 674                 if (continue_fs && journal)
 675                         schedule_work(&EXT4_SB(sb)->s_error_work);
 676                 else
 677                         ext4_commit_super(sb);
 678         }
 679
 680         /*
 681          * We force ERRORS_RO behavior when system is rebooting. Otherwise we
 682          * could panic during 'reboot -f' as the underlying device got already
 683          * disabled.
 684          */
 685         if (test_opt(sb, ERRORS_PANIC) && !system_going_down()) {
 686                 panic("EXT4-fs (device %s): panic forced after error\n",
 687                         sb->s_id);
 688         }
 689
 690         if (sb_rdonly(sb) || continue_fs)
 691                 return;
 692
 693         ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 694         /*
 695          * Make sure updated value of ->s_mount_flags will be visible before
 696          * ->s_flags update
 697          */
 698         smp_wmb();
 699         sb->s_flags |= SB_RDONLY;
 700 }
 701
 702 static void flush_stashed_error_work(struct work_struct *work)
 703 {
 704         struct ext4_sb_info *sbi = container_of(work, struct ext4_sb_info,
 705                                                 s_error_work);
 706         journal_t *journal = sbi->s_journal;
 707         handle_t *handle;
 708
 709         /*
 710          * If the journal is still running, we have to write out superblock
 711          * through the journal to avoid collisions of other journalled sb
 712          * updates.
 713          *
 714          * We use directly jbd2 functions here to avoid recursing back into
 715          * ext4 error handling code during handling of previous errors.
 716          */
 717         if (!sb_rdonly(sbi->s_sb) && journal) {
 718                 struct buffer_head *sbh = sbi->s_sbh;
 719                 handle = jbd2_journal_start(journal, 1);
 720                 if (IS_ERR(handle))
 721                         goto write_directly;
 722                 if (jbd2_journal_get_write_access(handle, sbh)) {
 723                         jbd2_journal_stop(handle);
 724                         goto write_directly;
 725                 }
 726                 ext4_update_super(sbi->s_sb);
 727                 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
 728                         ext4_msg(sbi->s_sb, KERN_ERR, "previous I/O error to "
 729                                  "superblock detected");
 730                         clear_buffer_write_io_error(sbh);
 731                         set_buffer_uptodate(sbh);
 732                 }
 733
 734                 if (jbd2_journal_dirty_metadata(handle, sbh)) {
 735                         jbd2_journal_stop(handle);
 736                         goto write_directly;
 737                 }
 738                 jbd2_journal_stop(handle);
 739                 ext4_notify_error_sysfs(sbi);
 740                 return;
 741         }
 742 write_directly:
 743         /*
 744          * Write through journal failed. Write sb directly to get error info
 745          * out and hope for the best.
 746          */
 747         ext4_commit_super(sbi->s_sb);
 748         ext4_notify_error_sysfs(sbi);
 749 }
 750
 751 #define ext4_error_ratelimit(sb)                                        \
 752                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 753                              "EXT4-fs error")
 754
 755 void __ext4_error(struct super_block *sb, const char *function,
 756                   unsigned int line, bool force_ro, int error, __u64 block,
 757                   const char *fmt, ...)
 758 {
 759         struct va_format vaf;
 760         va_list args;
 761
 762         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 763                 return;
 764
 765         trace_ext4_error(sb, function, line);
 766         if (ext4_error_ratelimit(sb)) {
 767                 va_start(args, fmt);
 768                 vaf.fmt = fmt;
 769                 vaf.va = &args;
 770                 printk(KERN_CRIT
 771                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 772                        sb->s_id, function, line, current->comm, &vaf);
 773                 va_end(args);
 774         }
 775         fsnotify_sb_error(sb, NULL, error ? error : EFSCORRUPTED);
 776
 777         ext4_handle_error(sb, force_ro, error, 0, block, function, line);
 778 }
 779
 780 void __ext4_error_inode(struct inode *inode, const char *function,
 781                         unsigned int line, ext4_fsblk_t block, int error,
 782                         const char *fmt, ...)
 783 {
 784         va_list args;
 785         struct va_format vaf;
 786
 787         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 788                 return;
 789
 790         trace_ext4_error(inode->i_sb, function, line);
 791         if (ext4_error_ratelimit(inode->i_sb)) {
 792                 va_start(args, fmt);
 793                 vaf.fmt = fmt;
 794                 vaf.va = &args;
 795                 if (block)
 796                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 797                                "inode #%lu: block %llu: comm %s: %pV\n",
 798                                inode->i_sb->s_id, function, line, inode->i_ino,
 799                                block, current->comm, &vaf);
 800                 else
 801                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 802                                "inode #%lu: comm %s: %pV\n",
 803                                inode->i_sb->s_id, function, line, inode->i_ino,
 804                                current->comm, &vaf);
 805                 va_end(args);
 806         }
 807         fsnotify_sb_error(inode->i_sb, inode, error ? error : EFSCORRUPTED);
 808
 809         ext4_handle_error(inode->i_sb, false, error, inode->i_ino, block,
 810                           function, line);
 811 }
 812
 813 void __ext4_error_file(struct file *file, const char *function,
 814                        unsigned int line, ext4_fsblk_t block,
 815                        const char *fmt, ...)
 816 {
 817         va_list args;
 818         struct va_format vaf;
 819         struct inode *inode = file_inode(file);
 820         char pathname[80], *path;
 821
 822         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 823                 return;
 824
 825         trace_ext4_error(inode->i_sb, function, line);
 826         if (ext4_error_ratelimit(inode->i_sb)) {
 827                 path = file_path(file, pathname, sizeof(pathname));
 828                 if (IS_ERR(path))
 829                         path = "(unknown)";
 830                 va_start(args, fmt);
 831                 vaf.fmt = fmt;
 832                 vaf.va = &args;
 833                 if (block)
 834                         printk(KERN_CRIT
 835                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 836                                "block %llu: comm %s: path %s: %pV\n",
 837                                inode->i_sb->s_id, function, line, inode->i_ino,
 838                                block, current->comm, path, &vaf);
 839                 else
 840                         printk(KERN_CRIT
 841                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 842                                "comm %s: path %s: %pV\n",
 843                                inode->i_sb->s_id, function, line, inode->i_ino,
 844                                current->comm, path, &vaf);
 845                 va_end(args);
 846         }
 847         fsnotify_sb_error(inode->i_sb, inode, EFSCORRUPTED);
 848
 849         ext4_handle_error(inode->i_sb, false, EFSCORRUPTED, inode->i_ino, block,
 850                           function, line);
 851 }
 852
 853 const char *ext4_decode_error(struct super_block *sb, int errno,
 854                               char nbuf[16])
 855 {
 856         char *errstr = NULL;
 857
 858         switch (errno) {
 859         case -EFSCORRUPTED:
 860                 errstr = "Corrupt filesystem";
 861                 break;
 862         case -EFSBADCRC:
 863                 errstr = "Filesystem failed CRC";
 864                 break;
 865         case -EIO:
 866                 errstr = "IO failure";
 867                 break;
 868         case -ENOMEM:
 869                 errstr = "Out of memory";
 870                 break;
 871         case -EROFS:
 872                 if (!sb || (EXT4_SB(sb)->s_journal &&
 873                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 874                         errstr = "Journal has aborted";
 875                 else
 876                         errstr = "Readonly filesystem";
 877                 break;
 878         default:
 879                 /* If the caller passed in an extra buffer for unknown
 880                  * errors, textualise them now.  Else we just return
 881                  * NULL. */
 882                 if (nbuf) {
 883                         /* Check for truncated error codes... */
 884                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 885                                 errstr = nbuf;
 886                 }
 887                 break;
 888         }
 889
 890         return errstr;
 891 }
 892
 893 /* __ext4_std_error decodes expected errors from journaling functions
 894  * automatically and invokes the appropriate error response.  */
 895
 896 void __ext4_std_error(struct super_block *sb, const char *function,
 897                       unsigned int line, int errno)
 898 {
 899         char nbuf[16];
 900         const char *errstr;
 901
 902         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 903                 return;
 904
 905         /* Special case: if the error is EROFS, and we're not already
 906          * inside a transaction, then there's really no point in logging
 907          * an error. */
 908         if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb))
 909                 return;
 910
 911         if (ext4_error_ratelimit(sb)) {
 912                 errstr = ext4_decode_error(sb, errno, nbuf);
 913                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 914                        sb->s_id, function, line, errstr);
 915         }
 916         fsnotify_sb_error(sb, NULL, errno ? errno : EFSCORRUPTED);
 917
 918         ext4_handle_error(sb, false, -errno, 0, 0, function, line);
 919 }
 920
 921 void __ext4_msg(struct super_block *sb,
 922                 const char *prefix, const char *fmt, ...)
 923 {
 924         struct va_format vaf;
 925         va_list args;
 926
 927         if (sb) {
 928                 atomic_inc(&EXT4_SB(sb)->s_msg_count);
 929                 if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state),
 930                                   "EXT4-fs"))
 931                         return;
 932         }
 933
 934         va_start(args, fmt);
 935         vaf.fmt = fmt;
 936         vaf.va = &args;
 937         if (sb)
 938                 printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 939         else
 940                 printk("%sEXT4-fs: %pV\n", prefix, &vaf);
 941         va_end(args);
 942 }
 943
 944 static int ext4_warning_ratelimit(struct super_block *sb)
 945 {
 946         atomic_inc(&EXT4_SB(sb)->s_warning_count);
 947         return ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state),
 948                             "EXT4-fs warning");
 949 }
 950
 951 void __ext4_warning(struct super_block *sb, const char *function,
 952                     unsigned int line, const char *fmt, ...)
 953 {
 954         struct va_format vaf;
 955         va_list args;
 956
 957         if (!ext4_warning_ratelimit(sb))
 958                 return;
 959
 960         va_start(args, fmt);
 961         vaf.fmt = fmt;
 962         vaf.va = &args;
 963         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 964                sb->s_id, function, line, &vaf);
 965         va_end(args);
 966 }
 967
 968 void __ext4_warning_inode(const struct inode *inode, const char *function,
 969                           unsigned int line, const char *fmt, ...)
 970 {
 971         struct va_format vaf;
 972         va_list args;
 973
 974         if (!ext4_warning_ratelimit(inode->i_sb))
 975                 return;
 976
 977         va_start(args, fmt);
 978         vaf.fmt = fmt;
 979         vaf.va = &args;
 980         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 981                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 982                function, line, inode->i_ino, current->comm, &vaf);
 983         va_end(args);
 984 }
 985
 986 void __ext4_grp_locked_error(const char *function, unsigned int line,
 987                              struct super_block *sb, ext4_group_t grp,
 988                              unsigned long ino, ext4_fsblk_t block,
 989                              const char *fmt, ...)
 990 __releases(bitlock)
 991 __acquires(bitlock)
 992 {
 993         struct va_format vaf;
 994         va_list args;
 995
 996         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 997                 return;
 998
 999         trace_ext4_error(sb, function, line);
1000         if (ext4_error_ratelimit(sb)) {
1001                 va_start(args, fmt);
1002                 vaf.fmt = fmt;
1003                 vaf.va = &args;
1004                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
1005                        sb->s_id, function, line, grp);
1006                 if (ino)
1007                         printk(KERN_CONT "inode %lu: ", ino);
1008                 if (block)
1009                         printk(KERN_CONT "block %llu:",
1010                                (unsigned long long) block);
1011                 printk(KERN_CONT "%pV\n", &vaf);
1012                 va_end(args);
1013         }
1014
1015         if (test_opt(sb, ERRORS_CONT)) {
1016                 if (test_opt(sb, WARN_ON_ERROR))
1017                         WARN_ON_ONCE(1);
1018                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
1019                 if (!bdev_read_only(sb->s_bdev)) {
1020                         save_error_info(sb, EFSCORRUPTED, ino, block, function,
1021                                         line);
1022                         schedule_work(&EXT4_SB(sb)->s_error_work);
1023                 }
1024                 return;
1025         }
1026         ext4_unlock_group(sb, grp);
1027         ext4_handle_error(sb, false, EFSCORRUPTED, ino, block, function, line);
1028         /*
1029          * We only get here in the ERRORS_RO case; relocking the group
1030          * may be dangerous, but nothing bad will happen since the
1031          * filesystem will have already been marked read/only and the
1032          * journal has been aborted.  We return 1 as a hint to callers
1033          * who might what to use the return value from
1034          * ext4_grp_locked_error() to distinguish between the
1035          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
1036          * aggressively from the ext4 function in question, with a
1037          * more appropriate error code.
1038          */
1039         ext4_lock_group(sb, grp);
1040         return;
1041 }
1042
1043 void ext4_mark_group_bitmap_corrupted(struct super_block *sb,
1044                                      ext4_group_t group,
1045                                      unsigned int flags)
1046 {
1047         struct ext4_sb_info *sbi = EXT4_SB(sb);
1048         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1049         struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
1050         int ret;
1051
1052         if (!grp || !gdp)
1053                 return;
1054         if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) {
1055                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT,
1056                                             &grp->bb_state);
1057                 if (!ret)
1058                         percpu_counter_sub(&sbi->s_freeclusters_counter,
1059                                            grp->bb_free);
1060         }
1061
1062         if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) {
1063                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT,
1064                                             &grp->bb_state);
1065                 if (!ret && gdp) {
1066                         int count;
1067
1068                         count = ext4_free_inodes_count(sb, gdp);
1069                         percpu_counter_sub(&sbi->s_freeinodes_counter,
1070                                            count);
1071                 }
1072         }
1073 }
1074
1075 void ext4_update_dynamic_rev(struct super_block *sb)
1076 {
1077         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1078
1079         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
1080                 return;
1081
1082         ext4_warning(sb,
1083                      "updating to rev %d because of new feature flag, "
1084                      "running e2fsck is recommended",
1085                      EXT4_DYNAMIC_REV);
1086
1087         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
1088         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
1089         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
1090         /* leave es->s_feature_*compat flags alone */
1091         /* es->s_uuid will be set by e2fsck if empty */
1092
1093         /*
1094          * The rest of the superblock fields should be zero, and if not it
1095          * means they are likely already in use, so leave them alone.  We
1096          * can leave it up to e2fsck to clean up any inconsistencies there.
1097          */
1098 }
1099
1100 /*
1101  * Open the external journal device
1102  */
1103 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
1104 {
1105         struct block_device *bdev;
1106
1107         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
1108         if (IS_ERR(bdev))
1109                 goto fail;
1110         return bdev;
1111
1112 fail:
1113         ext4_msg(sb, KERN_ERR,
1114                  "failed to open journal device unknown-block(%u,%u) %ld",
1115                  MAJOR(dev), MINOR(dev), PTR_ERR(bdev));
1116         return NULL;
1117 }
1118
1119 /*
1120  * Release the journal device
1121  */
1122 static void ext4_blkdev_put(struct block_device *bdev)
1123 {
1124         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1125 }
1126
1127 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
1128 {
1129         struct block_device *bdev;
1130         bdev = sbi->s_journal_bdev;
1131         if (bdev) {
1132                 /*
1133                  * Invalidate the journal device's buffers.  We don't want them
1134                  * floating about in memory - the physical journal device may
1135                  * hotswapped, and it breaks the `ro-after' testing code.
1136                  */
1137                 invalidate_bdev(bdev);
1138                 ext4_blkdev_put(bdev);
1139                 sbi->s_journal_bdev = NULL;
1140         }
1141 }
1142
1143 static inline struct inode *orphan_list_entry(struct list_head *l)
1144 {
1145         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
1146 }
1147
1148 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
1149 {
1150         struct list_head *l;
1151
1152         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
1153                  le32_to_cpu(sbi->s_es->s_last_orphan));
1154
1155         printk(KERN_ERR "sb_info orphan list:\n");
1156         list_for_each(l, &sbi->s_orphan) {
1157                 struct inode *inode = orphan_list_entry(l);
1158                 printk(KERN_ERR "  "
1159                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
1160                        inode->i_sb->s_id, inode->i_ino, inode,
1161                        inode->i_mode, inode->i_nlink,
1162                        NEXT_ORPHAN(inode));
1163         }
1164 }
1165
1166 #ifdef CONFIG_QUOTA
1167 static int ext4_quota_off(struct super_block *sb, int type);
1168
1169 static inline void ext4_quota_off_umount(struct super_block *sb)
1170 {
1171         int type;
1172
1173         /* Use our quota_off function to clear inode flags etc. */
1174         for (type = 0; type < EXT4_MAXQUOTAS; type++)
1175                 ext4_quota_off(sb, type);
1176 }
1177
1178 /*
1179  * This is a helper function which is used in the mount/remount
1180  * codepaths (which holds s_umount) to fetch the quota file name.
1181  */
1182 static inline char *get_qf_name(struct super_block *sb,
1183                                 struct ext4_sb_info *sbi,
1184                                 int type)
1185 {
1186         return rcu_dereference_protected(sbi->s_qf_names[type],
1187                                          lockdep_is_held(&sb->s_umount));
1188 }
1189 #else
1190 static inline void ext4_quota_off_umount(struct super_block *sb)
1191 {
1192 }
1193 #endif
1194
1195 static void ext4_put_super(struct super_block *sb)
1196 {
1197         struct ext4_sb_info *sbi = EXT4_SB(sb);
1198         struct ext4_super_block *es = sbi->s_es;
1199         struct buffer_head **group_desc;
1200         struct flex_groups **flex_groups;
1201         int aborted = 0;
1202         int i, err;
1203
1204         /*
1205          * Unregister sysfs before destroying jbd2 journal.
1206          * Since we could still access attr_journal_task attribute via sysfs
1207          * path which could have sbi->s_journal->j_task as NULL
1208          * Unregister sysfs before flush sbi->s_error_work.
1209          * Since user may read /proc/fs/ext4/xx/mb_groups during umount, If
1210          * read metadata verify failed then will queue error work.
1211          * flush_stashed_error_work will call start_this_handle may trigger
1212          * BUG_ON.
1213          */
1214         ext4_unregister_sysfs(sb);
1215
1216         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs unmount"))
1217                 ext4_msg(sb, KERN_INFO, "unmounting filesystem.");
1218
1219         ext4_unregister_li_request(sb);
1220         ext4_quota_off_umount(sb);
1221
1222         flush_work(&sbi->s_error_work);
1223         destroy_workqueue(sbi->rsv_conversion_wq);
1224         ext4_release_orphan_info(sb);
1225
1226         if (sbi->s_journal) {
1227                 aborted = is_journal_aborted(sbi->s_journal);
1228                 err = jbd2_journal_destroy(sbi->s_journal);
1229                 sbi->s_journal = NULL;
1230                 if ((err < 0) && !aborted) {
1231                         ext4_abort(sb, -err, "Couldn't clean up the journal");
1232                 }
1233         }
1234
1235         ext4_es_unregister_shrinker(sbi);
1236         del_timer_sync(&sbi->s_err_report);
1237         ext4_release_system_zone(sb);
1238         ext4_mb_release(sb);
1239         ext4_ext_release(sb);
1240
1241         if (!sb_rdonly(sb) && !aborted) {
1242                 ext4_clear_feature_journal_needs_recovery(sb);
1243                 ext4_clear_feature_orphan_present(sb);
1244                 es->s_state = cpu_to_le16(sbi->s_mount_state);
1245         }
1246         if (!sb_rdonly(sb))
1247                 ext4_commit_super(sb);
1248
1249         rcu_read_lock();
1250         group_desc = rcu_dereference(sbi->s_group_desc);
1251         for (i = 0; i < sbi->s_gdb_count; i++)
1252                 brelse(group_desc[i]);
1253         kvfree(group_desc);
1254         flex_groups = rcu_dereference(sbi->s_flex_groups);
1255         if (flex_groups) {
1256                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
1257                         kvfree(flex_groups[i]);
1258                 kvfree(flex_groups);
1259         }
1260         rcu_read_unlock();
1261         percpu_counter_destroy(&sbi->s_freeclusters_counter);
1262         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1263         percpu_counter_destroy(&sbi->s_dirs_counter);
1264         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
1265         percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit);
1266         percpu_free_rwsem(&sbi->s_writepages_rwsem);
1267 #ifdef CONFIG_QUOTA
1268         for (i = 0; i < EXT4_MAXQUOTAS; i++)
1269                 kfree(get_qf_name(sb, sbi, i));
1270 #endif
1271
1272         /* Debugging code just in case the in-memory inode orphan list
1273          * isn't empty.  The on-disk one can be non-empty if we've
1274          * detected an error and taken the fs readonly, but the
1275          * in-memory list had better be clean by this point. */
1276         if (!list_empty(&sbi->s_orphan))
1277                 dump_orphan_list(sb, sbi);
1278         ASSERT(list_empty(&sbi->s_orphan));
1279
1280         sync_blockdev(sb->s_bdev);
1281         invalidate_bdev(sb->s_bdev);
1282         if (sbi->s_journal_bdev && sbi->s_journal_bdev != sb->s_bdev) {
1283                 sync_blockdev(sbi->s_journal_bdev);
1284                 ext4_blkdev_remove(sbi);
1285         }
1286
1287         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
1288         sbi->s_ea_inode_cache = NULL;
1289
1290         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
1291         sbi->s_ea_block_cache = NULL;
1292
1293         ext4_stop_mmpd(sbi);
1294
1295         brelse(sbi->s_sbh);
1296         sb->s_fs_info = NULL;
1297         /*
1298          * Now that we are completely done shutting down the
1299          * superblock, we need to actually destroy the kobject.
1300          */
1301         kobject_put(&sbi->s_kobj);
1302         wait_for_completion(&sbi->s_kobj_unregister);
1303         if (sbi->s_chksum_driver)
1304                 crypto_free_shash(sbi->s_chksum_driver);
1305         kfree(sbi->s_blockgroup_lock);
1306         fs_put_dax(sbi->s_daxdev, NULL);
1307         fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy);
1308 #if IS_ENABLED(CONFIG_UNICODE)
1309         utf8_unload(sb->s_encoding);
1310 #endif
1311         kfree(sbi);
1312 }
1313
1314 static struct kmem_cache *ext4_inode_cachep;
1315
1316 /*
1317  * Called inside transaction, so use GFP_NOFS
1318  */
1319 static struct inode *ext4_alloc_inode(struct super_block *sb)
1320 {
1321         struct ext4_inode_info *ei;
1322
1323         ei = alloc_inode_sb(sb, ext4_inode_cachep, GFP_NOFS);
1324         if (!ei)
1325                 return NULL;
1326
1327         inode_set_iversion(&ei->vfs_inode, 1);
1328         ei->i_flags = 0;
1329         spin_lock_init(&ei->i_raw_lock);
1330         INIT_LIST_HEAD(&ei->i_prealloc_list);
1331         atomic_set(&ei->i_prealloc_active, 0);
1332         spin_lock_init(&ei->i_prealloc_lock);
1333         ext4_es_init_tree(&ei->i_es_tree);
1334         rwlock_init(&ei->i_es_lock);
1335         INIT_LIST_HEAD(&ei->i_es_list);
1336         ei->i_es_all_nr = 0;
1337         ei->i_es_shk_nr = 0;
1338         ei->i_es_shrink_lblk = 0;
1339         ei->i_reserved_data_blocks = 0;
1340         spin_lock_init(&(ei->i_block_reservation_lock));
1341         ext4_init_pending_tree(&ei->i_pending_tree);
1342 #ifdef CONFIG_QUOTA
1343         ei->i_reserved_quota = 0;
1344         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
1345 #endif
1346         ei->jinode = NULL;
1347         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
1348         spin_lock_init(&ei->i_completed_io_lock);
1349         ei->i_sync_tid = 0;
1350         ei->i_datasync_tid = 0;
1351         atomic_set(&ei->i_unwritten, 0);
1352         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
1353         ext4_fc_init_inode(&ei->vfs_inode);
1354         mutex_init(&ei->i_fc_lock);
1355         return &ei->vfs_inode;
1356 }
1357
1358 static int ext4_drop_inode(struct inode *inode)
1359 {
1360         int drop = generic_drop_inode(inode);
1361
1362         if (!drop)
1363                 drop = fscrypt_drop_inode(inode);
1364
1365         trace_ext4_drop_inode(inode, drop);
1366         return drop;
1367 }
1368
1369 static void ext4_free_in_core_inode(struct inode *inode)
1370 {
1371         fscrypt_free_inode(inode);
1372         if (!list_empty(&(EXT4_I(inode)->i_fc_list))) {
1373                 pr_warn("%s: inode %ld still in fc list",
1374                         __func__, inode->i_ino);
1375         }
1376         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1377 }
1378
1379 static void ext4_destroy_inode(struct inode *inode)
1380 {
1381         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1382                 ext4_msg(inode->i_sb, KERN_ERR,
1383                          "Inode %lu (%p): orphan list check failed!",
1384                          inode->i_ino, EXT4_I(inode));
1385                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1386                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1387                                 true);
1388                 dump_stack();
1389         }
1390
1391         if (EXT4_I(inode)->i_reserved_data_blocks)
1392                 ext4_msg(inode->i_sb, KERN_ERR,
1393                          "Inode %lu (%p): i_reserved_data_blocks (%u) not cleared!",
1394                          inode->i_ino, EXT4_I(inode),
1395                          EXT4_I(inode)->i_reserved_data_blocks);
1396 }
1397
1398 static void init_once(void *foo)
1399 {
1400         struct ext4_inode_info *ei = foo;
1401
1402         INIT_LIST_HEAD(&ei->i_orphan);
1403         init_rwsem(&ei->xattr_sem);
1404         init_rwsem(&ei->i_data_sem);
1405         inode_init_once(&ei->vfs_inode);
1406         ext4_fc_init_inode(&ei->vfs_inode);
1407 }
1408
1409 static int __init init_inodecache(void)
1410 {
1411         ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache",
1412                                 sizeof(struct ext4_inode_info), 0,
1413                                 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
1414                                         SLAB_ACCOUNT),
1415                                 offsetof(struct ext4_inode_info, i_data),
1416                                 sizeof_field(struct ext4_inode_info, i_data),
1417                                 init_once);
1418         if (ext4_inode_cachep == NULL)
1419                 return -ENOMEM;
1420         return 0;
1421 }
1422
1423 static void destroy_inodecache(void)
1424 {
1425         /*
1426          * Make sure all delayed rcu free inodes are flushed before we
1427          * destroy cache.
1428          */
1429         rcu_barrier();
1430         kmem_cache_destroy(ext4_inode_cachep);
1431 }
1432
1433 void ext4_clear_inode(struct inode *inode)
1434 {
1435         ext4_fc_del(inode);
1436         invalidate_inode_buffers(inode);
1437         clear_inode(inode);
1438         ext4_discard_preallocations(inode, 0);
1439         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1440         dquot_drop(inode);
1441         if (EXT4_I(inode)->jinode) {
1442                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1443                                                EXT4_I(inode)->jinode);
1444                 jbd2_free_inode(EXT4_I(inode)->jinode);
1445                 EXT4_I(inode)->jinode = NULL;
1446         }
1447         fscrypt_put_encryption_info(inode);
1448         fsverity_cleanup_inode(inode);
1449 }
1450
1451 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1452                                         u64 ino, u32 generation)
1453 {
1454         struct inode *inode;
1455
1456         /*
1457          * Currently we don't know the generation for parent directory, so
1458          * a generation of 0 means "accept any"
1459          */
1460         inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE);
1461         if (IS_ERR(inode))
1462                 return ERR_CAST(inode);
1463         if (generation && inode->i_generation != generation) {
1464                 iput(inode);
1465                 return ERR_PTR(-ESTALE);
1466         }
1467
1468         return inode;
1469 }
1470
1471 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1472                                         int fh_len, int fh_type)
1473 {
1474         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1475                                     ext4_nfs_get_inode);
1476 }
1477
1478 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1479                                         int fh_len, int fh_type)
1480 {
1481         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1482                                     ext4_nfs_get_inode);
1483 }
1484
1485 static int ext4_nfs_commit_metadata(struct inode *inode)
1486 {
1487         struct writeback_control wbc = {
1488                 .sync_mode = WB_SYNC_ALL
1489         };
1490
1491         trace_ext4_nfs_commit_metadata(inode);
1492         return ext4_write_inode(inode, &wbc);
1493 }
1494
1495 #ifdef CONFIG_QUOTA
1496 static const char * const quotatypes[] = INITQFNAMES;
1497 #define QTYPE2NAME(t) (quotatypes[t])
1498
1499 static int ext4_write_dquot(struct dquot *dquot);
1500 static int ext4_acquire_dquot(struct dquot *dquot);
1501 static int ext4_release_dquot(struct dquot *dquot);
1502 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1503 static int ext4_write_info(struct super_block *sb, int type);
1504 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1505                          const struct path *path);
1506 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1507                                size_t len, loff_t off);
1508 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1509                                 const char *data, size_t len, loff_t off);
1510 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1511                              unsigned int flags);
1512
1513 static struct dquot **ext4_get_dquots(struct inode *inode)
1514 {
1515         return EXT4_I(inode)->i_dquot;
1516 }
1517
1518 static const struct dquot_operations ext4_quota_operations = {
1519         .get_reserved_space     = ext4_get_reserved_space,
1520         .write_dquot            = ext4_write_dquot,
1521         .acquire_dquot          = ext4_acquire_dquot,
1522         .release_dquot          = ext4_release_dquot,
1523         .mark_dirty             = ext4_mark_dquot_dirty,
1524         .write_info             = ext4_write_info,
1525         .alloc_dquot            = dquot_alloc,
1526         .destroy_dquot          = dquot_destroy,
1527         .get_projid             = ext4_get_projid,
1528         .get_inode_usage        = ext4_get_inode_usage,
1529         .get_next_id            = dquot_get_next_id,
1530 };
1531
1532 static const struct quotactl_ops ext4_qctl_operations = {
1533         .quota_on       = ext4_quota_on,
1534         .quota_off      = ext4_quota_off,
1535         .quota_sync     = dquot_quota_sync,
1536         .get_state      = dquot_get_state,
1537         .set_info       = dquot_set_dqinfo,
1538         .get_dqblk      = dquot_get_dqblk,
1539         .set_dqblk      = dquot_set_dqblk,
1540         .get_nextdqblk  = dquot_get_next_dqblk,
1541 };
1542 #endif
1543
1544 static const struct super_operations ext4_sops = {
1545         .alloc_inode    = ext4_alloc_inode,
1546         .free_inode     = ext4_free_in_core_inode,
1547         .destroy_inode  = ext4_destroy_inode,
1548         .write_inode    = ext4_write_inode,
1549         .dirty_inode    = ext4_dirty_inode,
1550         .drop_inode     = ext4_drop_inode,
1551         .evict_inode    = ext4_evict_inode,
1552         .put_super      = ext4_put_super,
1553         .sync_fs        = ext4_sync_fs,
1554         .freeze_fs      = ext4_freeze,
1555         .unfreeze_fs    = ext4_unfreeze,
1556         .statfs         = ext4_statfs,
1557         .show_options   = ext4_show_options,
1558 #ifdef CONFIG_QUOTA
1559         .quota_read     = ext4_quota_read,
1560         .quota_write    = ext4_quota_write,
1561         .get_dquots     = ext4_get_dquots,
1562 #endif
1563 };
1564
1565 static const struct export_operations ext4_export_ops = {
1566         .fh_to_dentry = ext4_fh_to_dentry,
1567         .fh_to_parent = ext4_fh_to_parent,
1568         .get_parent = ext4_get_parent,
1569         .commit_metadata = ext4_nfs_commit_metadata,
1570 };
1571
1572 enum {
1573         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1574         Opt_resgid, Opt_resuid, Opt_sb,
1575         Opt_nouid32, Opt_debug, Opt_removed,
1576         Opt_user_xattr, Opt_acl,
1577         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1578         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1579         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1580         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1581         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1582         Opt_inlinecrypt,
1583         Opt_usrjquota, Opt_grpjquota, Opt_quota,
1584         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1585         Opt_usrquota, Opt_grpquota, Opt_prjquota,
1586         Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never,
1587         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
1588         Opt_nowarn_on_error, Opt_mblk_io_submit, Opt_debug_want_extra_isize,
1589         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1590         Opt_inode_readahead_blks, Opt_journal_ioprio,
1591         Opt_dioread_nolock, Opt_dioread_lock,
1592         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1593         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1594         Opt_no_prefetch_block_bitmaps, Opt_mb_optimize_scan,
1595         Opt_errors, Opt_data, Opt_data_err, Opt_jqfmt, Opt_dax_type,
1596 #ifdef CONFIG_EXT4_DEBUG
1597         Opt_fc_debug_max_replay, Opt_fc_debug_force
1598 #endif
1599 };
1600
1601 static const struct constant_table ext4_param_errors[] = {
1602         {"continue",    EXT4_MOUNT_ERRORS_CONT},
1603         {"panic",       EXT4_MOUNT_ERRORS_PANIC},
1604         {"remount-ro",  EXT4_MOUNT_ERRORS_RO},
1605         {}
1606 };
1607
1608 static const struct constant_table ext4_param_data[] = {
1609         {"journal",     EXT4_MOUNT_JOURNAL_DATA},
1610         {"ordered",     EXT4_MOUNT_ORDERED_DATA},
1611         {"writeback",   EXT4_MOUNT_WRITEBACK_DATA},
1612         {}
1613 };
1614
1615 static const struct constant_table ext4_param_data_err[] = {
1616         {"abort",       Opt_data_err_abort},
1617         {"ignore",      Opt_data_err_ignore},
1618         {}
1619 };
1620
1621 static const struct constant_table ext4_param_jqfmt[] = {
1622         {"vfsold",      QFMT_VFS_OLD},
1623         {"vfsv0",       QFMT_VFS_V0},
1624         {"vfsv1",       QFMT_VFS_V1},
1625         {}
1626 };
1627
1628 static const struct constant_table ext4_param_dax[] = {
1629         {"always",      Opt_dax_always},
1630         {"inode",       Opt_dax_inode},
1631         {"never",       Opt_dax_never},
1632         {}
1633 };
1634
1635 /* String parameter that allows empty argument */
1636 #define fsparam_string_empty(NAME, OPT) \
1637         __fsparam(fs_param_is_string, NAME, OPT, fs_param_can_be_empty, NULL)
1638
1639 /*
1640  * Mount option specification
1641  * We don't use fsparam_flag_no because of the way we set the
1642  * options and the way we show them in _ext4_show_options(). To
1643  * keep the changes to a minimum, let's keep the negative options
1644  * separate for now.
1645  */
1646 static const struct fs_parameter_spec ext4_param_specs[] = {
1647         fsparam_flag    ("bsddf",               Opt_bsd_df),
1648         fsparam_flag    ("minixdf",             Opt_minix_df),
1649         fsparam_flag    ("grpid",               Opt_grpid),
1650         fsparam_flag    ("bsdgroups",           Opt_grpid),
1651         fsparam_flag    ("nogrpid",             Opt_nogrpid),
1652         fsparam_flag    ("sysvgroups",          Opt_nogrpid),
1653         fsparam_u32     ("resgid",              Opt_resgid),
1654         fsparam_u32     ("resuid",              Opt_resuid),
1655         fsparam_u32     ("sb",                  Opt_sb),
1656         fsparam_enum    ("errors",              Opt_errors, ext4_param_errors),
1657         fsparam_flag    ("nouid32",             Opt_nouid32),
1658         fsparam_flag    ("debug",               Opt_debug),
1659         fsparam_flag    ("oldalloc",            Opt_removed),
1660         fsparam_flag    ("orlov",               Opt_removed),
1661         fsparam_flag    ("user_xattr",          Opt_user_xattr),
1662         fsparam_flag    ("acl",                 Opt_acl),
1663         fsparam_flag    ("norecovery",          Opt_noload),
1664         fsparam_flag    ("noload",              Opt_noload),
1665         fsparam_flag    ("bh",                  Opt_removed),
1666         fsparam_flag    ("nobh",                Opt_removed),
1667         fsparam_u32     ("commit",              Opt_commit),
1668         fsparam_u32     ("min_batch_time",      Opt_min_batch_time),
1669         fsparam_u32     ("max_batch_time",      Opt_max_batch_time),
1670         fsparam_u32     ("journal_dev",         Opt_journal_dev),
1671         fsparam_bdev    ("journal_path",        Opt_journal_path),
1672         fsparam_flag    ("journal_checksum",    Opt_journal_checksum),
1673         fsparam_flag    ("nojournal_checksum",  Opt_nojournal_checksum),
1674         fsparam_flag    ("journal_async_commit",Opt_journal_async_commit),
1675         fsparam_flag    ("abort",               Opt_abort),
1676         fsparam_enum    ("data",                Opt_data, ext4_param_data),
1677         fsparam_enum    ("data_err",            Opt_data_err,
1678                                                 ext4_param_data_err),
1679         fsparam_string_empty
1680                         ("usrjquota",           Opt_usrjquota),
1681         fsparam_string_empty
1682                         ("grpjquota",           Opt_grpjquota),
1683         fsparam_enum    ("jqfmt",               Opt_jqfmt, ext4_param_jqfmt),
1684         fsparam_flag    ("grpquota",            Opt_grpquota),
1685         fsparam_flag    ("quota",               Opt_quota),
1686         fsparam_flag    ("noquota",             Opt_noquota),
1687         fsparam_flag    ("usrquota",            Opt_usrquota),
1688         fsparam_flag    ("prjquota",            Opt_prjquota),
1689         fsparam_flag    ("barrier",             Opt_barrier),
1690         fsparam_u32     ("barrier",             Opt_barrier),
1691         fsparam_flag    ("nobarrier",           Opt_nobarrier),
1692         fsparam_flag    ("i_version",           Opt_removed),
1693         fsparam_flag    ("dax",                 Opt_dax),
1694         fsparam_enum    ("dax",                 Opt_dax_type, ext4_param_dax),
1695         fsparam_u32     ("stripe",              Opt_stripe),
1696         fsparam_flag    ("delalloc",            Opt_delalloc),
1697         fsparam_flag    ("nodelalloc",          Opt_nodelalloc),
1698         fsparam_flag    ("warn_on_error",       Opt_warn_on_error),
1699         fsparam_flag    ("nowarn_on_error",     Opt_nowarn_on_error),
1700         fsparam_u32     ("debug_want_extra_isize",
1701                                                 Opt_debug_want_extra_isize),
1702         fsparam_flag    ("mblk_io_submit",      Opt_removed),
1703         fsparam_flag    ("nomblk_io_submit",    Opt_removed),
1704         fsparam_flag    ("block_validity",      Opt_block_validity),
1705         fsparam_flag    ("noblock_validity",    Opt_noblock_validity),
1706         fsparam_u32     ("inode_readahead_blks",
1707                                                 Opt_inode_readahead_blks),
1708         fsparam_u32     ("journal_ioprio",      Opt_journal_ioprio),
1709         fsparam_u32     ("auto_da_alloc",       Opt_auto_da_alloc),
1710         fsparam_flag    ("auto_da_alloc",       Opt_auto_da_alloc),
1711         fsparam_flag    ("noauto_da_alloc",     Opt_noauto_da_alloc),
1712         fsparam_flag    ("dioread_nolock",      Opt_dioread_nolock),
1713         fsparam_flag    ("nodioread_nolock",    Opt_dioread_lock),
1714         fsparam_flag    ("dioread_lock",        Opt_dioread_lock),
1715         fsparam_flag    ("discard",             Opt_discard),
1716         fsparam_flag    ("nodiscard",           Opt_nodiscard),
1717         fsparam_u32     ("init_itable",         Opt_init_itable),
1718         fsparam_flag    ("init_itable",         Opt_init_itable),
1719         fsparam_flag    ("noinit_itable",       Opt_noinit_itable),
1720 #ifdef CONFIG_EXT4_DEBUG
1721         fsparam_flag    ("fc_debug_force",      Opt_fc_debug_force),
1722         fsparam_u32     ("fc_debug_max_replay", Opt_fc_debug_max_replay),
1723 #endif
1724         fsparam_u32     ("max_dir_size_kb",     Opt_max_dir_size_kb),
1725         fsparam_flag    ("test_dummy_encryption",
1726                                                 Opt_test_dummy_encryption),
1727         fsparam_string  ("test_dummy_encryption",
1728                                                 Opt_test_dummy_encryption),
1729         fsparam_flag    ("inlinecrypt",         Opt_inlinecrypt),
1730         fsparam_flag    ("nombcache",           Opt_nombcache),
1731         fsparam_flag    ("no_mbcache",          Opt_nombcache), /* for backward compatibility */
1732         fsparam_flag    ("prefetch_block_bitmaps",
1733                                                 Opt_removed),
1734         fsparam_flag    ("no_prefetch_block_bitmaps",
1735                                                 Opt_no_prefetch_block_bitmaps),
1736         fsparam_s32     ("mb_optimize_scan",    Opt_mb_optimize_scan),
1737         fsparam_string  ("check",               Opt_removed),   /* mount option from ext2/3 */
1738         fsparam_flag    ("nocheck",             Opt_removed),   /* mount option from ext2/3 */
1739         fsparam_flag    ("reservation",         Opt_removed),   /* mount option from ext2/3 */
1740         fsparam_flag    ("noreservation",       Opt_removed),   /* mount option from ext2/3 */
1741         fsparam_u32     ("journal",             Opt_removed),   /* mount option from ext2/3 */
1742         {}
1743 };
1744
1745 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1746
1747 #define MOPT_SET        0x0001
1748 #define MOPT_CLEAR      0x0002
1749 #define MOPT_NOSUPPORT  0x0004
1750 #define MOPT_EXPLICIT   0x0008
1751 #ifdef CONFIG_QUOTA
1752 #define MOPT_Q          0
1753 #define MOPT_QFMT       0x0010
1754 #else
1755 #define MOPT_Q          MOPT_NOSUPPORT
1756 #define MOPT_QFMT       MOPT_NOSUPPORT
1757 #endif
1758 #define MOPT_NO_EXT2    0x0020
1759 #define MOPT_NO_EXT3    0x0040
1760 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1761 #define MOPT_SKIP       0x0080
1762 #define MOPT_2          0x0100
1763
1764 static const struct mount_opts {
1765         int     token;
1766         int     mount_opt;
1767         int     flags;
1768 } ext4_mount_opts[] = {
1769         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1770         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1771         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1772         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1773         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1774         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1775         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1776          MOPT_EXT4_ONLY | MOPT_SET},
1777         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1778          MOPT_EXT4_ONLY | MOPT_CLEAR},
1779         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1780         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1781         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1782          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1783         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1784          MOPT_EXT4_ONLY | MOPT_CLEAR},
1785         {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET},
1786         {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR},
1787         {Opt_commit, 0, MOPT_NO_EXT2},
1788         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1789          MOPT_EXT4_ONLY | MOPT_CLEAR},
1790         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1791          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1792         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1793                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1794          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1795         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1796         {Opt_data_err, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_NO_EXT2},
1797         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1798         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1799         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1800         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1801         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1802         {Opt_dax_type, 0, MOPT_EXT4_ONLY},
1803         {Opt_journal_dev, 0, MOPT_NO_EXT2},
1804         {Opt_journal_path, 0, MOPT_NO_EXT2},
1805         {Opt_journal_ioprio, 0, MOPT_NO_EXT2},
1806         {Opt_data, 0, MOPT_NO_EXT2},
1807         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1808 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1809         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1810 #else
1811         {Opt_acl, 0, MOPT_NOSUPPORT},
1812 #endif
1813         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1814         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1815         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1816         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1817                                                         MOPT_SET | MOPT_Q},
1818         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1819                                                         MOPT_SET | MOPT_Q},
1820         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1821                                                         MOPT_SET | MOPT_Q},
1822         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1823                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1824                                                         MOPT_CLEAR | MOPT_Q},
1825         {Opt_usrjquota, 0, MOPT_Q},
1826         {Opt_grpjquota, 0, MOPT_Q},
1827         {Opt_jqfmt, 0, MOPT_QFMT},
1828         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1829         {Opt_no_prefetch_block_bitmaps, EXT4_MOUNT_NO_PREFETCH_BLOCK_BITMAPS,
1830          MOPT_SET},
1831 #ifdef CONFIG_EXT4_DEBUG
1832         {Opt_fc_debug_force, EXT4_MOUNT2_JOURNAL_FAST_COMMIT,
1833          MOPT_SET | MOPT_2 | MOPT_EXT4_ONLY},
1834 #endif
1835         {Opt_err, 0, 0}
1836 };
1837
1838 #if IS_ENABLED(CONFIG_UNICODE)
1839 static const struct ext4_sb_encodings {
1840         __u16 magic;
1841         char *name;
1842         unsigned int version;
1843 } ext4_sb_encoding_map[] = {
1844         {EXT4_ENC_UTF8_12_1, "utf8", UNICODE_AGE(12, 1, 0)},
1845 };
1846
1847 static const struct ext4_sb_encodings *
1848 ext4_sb_read_encoding(const struct ext4_super_block *es)
1849 {
1850         __u16 magic = le16_to_cpu(es->s_encoding);
1851         int i;
1852
1853         for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++)
1854                 if (magic == ext4_sb_encoding_map[i].magic)
1855                         return &ext4_sb_encoding_map[i];
1856
1857         return NULL;
1858 }
1859 #endif
1860
1861 #define EXT4_SPEC_JQUOTA                        (1 <<  0)
1862 #define EXT4_SPEC_JQFMT                         (1 <<  1)
1863 #define EXT4_SPEC_DATAJ                         (1 <<  2)
1864 #define EXT4_SPEC_SB_BLOCK                      (1 <<  3)
1865 #define EXT4_SPEC_JOURNAL_DEV                   (1 <<  4)
1866 #define EXT4_SPEC_JOURNAL_IOPRIO                (1 <<  5)
1867 #define EXT4_SPEC_s_want_extra_isize            (1 <<  7)
1868 #define EXT4_SPEC_s_max_batch_time              (1 <<  8)
1869 #define EXT4_SPEC_s_min_batch_time              (1 <<  9)
1870 #define EXT4_SPEC_s_inode_readahead_blks        (1 << 10)
1871 #define EXT4_SPEC_s_li_wait_mult                (1 << 11)
1872 #define EXT4_SPEC_s_max_dir_size_kb             (1 << 12)
1873 #define EXT4_SPEC_s_stripe                      (1 << 13)
1874 #define EXT4_SPEC_s_resuid                      (1 << 14)
1875 #define EXT4_SPEC_s_resgid                      (1 << 15)
1876 #define EXT4_SPEC_s_commit_interval             (1 << 16)
1877 #define EXT4_SPEC_s_fc_debug_max_replay         (1 << 17)
1878 #define EXT4_SPEC_s_sb_block                    (1 << 18)
1879 #define EXT4_SPEC_mb_optimize_scan              (1 << 19)
1880
1881 struct ext4_fs_context {
1882         char            *s_qf_names[EXT4_MAXQUOTAS];
1883         struct fscrypt_dummy_policy dummy_enc_policy;
1884         int             s_jquota_fmt;   /* Format of quota to use */
1885 #ifdef CONFIG_EXT4_DEBUG
1886         int s_fc_debug_max_replay;
1887 #endif
1888         unsigned short  qname_spec;
1889         unsigned long   vals_s_flags;   /* Bits to set in s_flags */
1890         unsigned long   mask_s_flags;   /* Bits changed in s_flags */
1891         unsigned long   journal_devnum;
1892         unsigned long   s_commit_interval;
1893         unsigned long   s_stripe;
1894         unsigned int    s_inode_readahead_blks;
1895         unsigned int    s_want_extra_isize;
1896         unsigned int    s_li_wait_mult;
1897         unsigned int    s_max_dir_size_kb;
1898         unsigned int    journal_ioprio;
1899         unsigned int    vals_s_mount_opt;
1900         unsigned int    mask_s_mount_opt;
1901         unsigned int    vals_s_mount_opt2;
1902         unsigned int    mask_s_mount_opt2;
1903         unsigned long   vals_s_mount_flags;
1904         unsigned long   mask_s_mount_flags;
1905         unsigned int    opt_flags;      /* MOPT flags */
1906         unsigned int    spec;
1907         u32             s_max_batch_time;
1908         u32             s_min_batch_time;
1909         kuid_t          s_resuid;
1910         kgid_t          s_resgid;
1911         ext4_fsblk_t    s_sb_block;
1912 };
1913
1914 static void ext4_fc_free(struct fs_context *fc)
1915 {
1916         struct ext4_fs_context *ctx = fc->fs_private;
1917         int i;
1918
1919         if (!ctx)
1920                 return;
1921
1922         for (i = 0; i < EXT4_MAXQUOTAS; i++)
1923                 kfree(ctx->s_qf_names[i]);
1924
1925         fscrypt_free_dummy_policy(&ctx->dummy_enc_policy);
1926         kfree(ctx);
1927 }
1928
1929 int ext4_init_fs_context(struct fs_context *fc)
1930 {
1931         struct ext4_fs_context *ctx;
1932
1933         ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL);
1934         if (!ctx)
1935                 return -ENOMEM;
1936
1937         fc->fs_private = ctx;
1938         fc->ops = &ext4_context_ops;
1939
1940         return 0;
1941 }
1942
1943 #ifdef CONFIG_QUOTA
1944 /*
1945  * Note the name of the specified quota file.
1946  */
1947 static int note_qf_name(struct fs_context *fc, int qtype,
1948                        struct fs_parameter *param)
1949 {
1950         struct ext4_fs_context *ctx = fc->fs_private;
1951         char *qname;
1952
1953         if (param->size < 1) {
1954                 ext4_msg(NULL, KERN_ERR, "Missing quota name");
1955                 return -EINVAL;
1956         }
1957         if (strchr(param->string, '/')) {
1958                 ext4_msg(NULL, KERN_ERR,
1959                          "quotafile must be on filesystem root");
1960                 return -EINVAL;
1961         }
1962         if (ctx->s_qf_names[qtype]) {
1963                 if (strcmp(ctx->s_qf_names[qtype], param->string) != 0) {
1964                         ext4_msg(NULL, KERN_ERR,
1965                                  "%s quota file already specified",
1966                                  QTYPE2NAME(qtype));
1967                         return -EINVAL;
1968                 }
1969                 return 0;
1970         }
1971
1972         qname = kmemdup_nul(param->string, param->size, GFP_KERNEL);
1973         if (!qname) {
1974                 ext4_msg(NULL, KERN_ERR,
1975                          "Not enough memory for storing quotafile name");
1976                 return -ENOMEM;
1977         }
1978         ctx->s_qf_names[qtype] = qname;
1979         ctx->qname_spec |= 1 << qtype;
1980         ctx->spec |= EXT4_SPEC_JQUOTA;
1981         return 0;
1982 }
1983
1984 /*
1985  * Clear the name of the specified quota file.
1986  */
1987 static int unnote_qf_name(struct fs_context *fc, int qtype)
1988 {
1989         struct ext4_fs_context *ctx = fc->fs_private;
1990
1991         if (ctx->s_qf_names[qtype])
1992                 kfree(ctx->s_qf_names[qtype]);
1993
1994         ctx->s_qf_names[qtype] = NULL;
1995         ctx->qname_spec |= 1 << qtype;
1996         ctx->spec |= EXT4_SPEC_JQUOTA;
1997         return 0;
1998 }
1999 #endif
2000
2001 static int ext4_parse_test_dummy_encryption(const struct fs_parameter *param,
2002                                             struct ext4_fs_context *ctx)
2003 {
2004         int err;
2005
2006         if (!IS_ENABLED(CONFIG_FS_ENCRYPTION)) {
2007                 ext4_msg(NULL, KERN_WARNING,
2008                          "test_dummy_encryption option not supported");
2009                 return -EINVAL;
2010         }
2011         err = fscrypt_parse_test_dummy_encryption(param,
2012                                                   &ctx->dummy_enc_policy);
2013         if (err == -EINVAL) {
2014                 ext4_msg(NULL, KERN_WARNING,
2015                          "Value of option \"%s\" is unrecognized", param->key);
2016         } else if (err == -EEXIST) {
2017                 ext4_msg(NULL, KERN_WARNING,
2018                          "Conflicting test_dummy_encryption options");
2019                 return -EINVAL;
2020         }
2021         return err;
2022 }
2023
2024 #define EXT4_SET_CTX(name)                                              \
2025 static inline void ctx_set_##name(struct ext4_fs_context *ctx,          \
2026                                   unsigned long flag)                   \
2027 {                                                                       \
2028         ctx->mask_s_##name |= flag;                                     \
2029         ctx->vals_s_##name |= flag;                                     \
2030 }
2031
2032 #define EXT4_CLEAR_CTX(name)                                            \
2033 static inline void ctx_clear_##name(struct ext4_fs_context *ctx,        \
2034                                     unsigned long flag)                 \
2035 {                                                                       \
2036         ctx->mask_s_##name |= flag;                                     \
2037         ctx->vals_s_##name &= ~flag;                                    \
2038 }
2039
2040 #define EXT4_TEST_CTX(name)                                             \
2041 static inline unsigned long                                             \
2042 ctx_test_##name(struct ext4_fs_context *ctx, unsigned long flag)        \
2043 {                                                                       \
2044         return (ctx->vals_s_##name & flag);                             \
2045 }
2046
2047 EXT4_SET_CTX(flags); /* set only */
2048 EXT4_SET_CTX(mount_opt);
2049 EXT4_CLEAR_CTX(mount_opt);
2050 EXT4_TEST_CTX(mount_opt);
2051 EXT4_SET_CTX(mount_opt2);
2052 EXT4_CLEAR_CTX(mount_opt2);
2053 EXT4_TEST_CTX(mount_opt2);
2054
2055 static inline void ctx_set_mount_flag(struct ext4_fs_context *ctx, int bit)
2056 {
2057         set_bit(bit, &ctx->mask_s_mount_flags);
2058         set_bit(bit, &ctx->vals_s_mount_flags);
2059 }
2060
2061 static int ext4_parse_param(struct fs_context *fc, struct fs_parameter *param)
2062 {
2063         struct ext4_fs_context *ctx = fc->fs_private;
2064         struct fs_parse_result result;
2065         const struct mount_opts *m;
2066         int is_remount;
2067         kuid_t uid;
2068         kgid_t gid;
2069         int token;
2070
2071         token = fs_parse(fc, ext4_param_specs, param, &result);
2072         if (token < 0)
2073                 return token;
2074         is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE;
2075
2076         for (m = ext4_mount_opts; m->token != Opt_err; m++)
2077                 if (token == m->token)
2078                         break;
2079
2080         ctx->opt_flags |= m->flags;
2081
2082         if (m->flags & MOPT_EXPLICIT) {
2083                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
2084                         ctx_set_mount_opt2(ctx, EXT4_MOUNT2_EXPLICIT_DELALLOC);
2085                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
2086                         ctx_set_mount_opt2(ctx,
2087                                        EXT4_MOUNT2_EXPLICIT_JOURNAL_CHECKSUM);
2088                 } else
2089                         return -EINVAL;
2090         }
2091
2092         if (m->flags & MOPT_NOSUPPORT) {
2093                 ext4_msg(NULL, KERN_ERR, "%s option not supported",
2094                          param->key);
2095                 return 0;
2096         }
2097
2098         switch (token) {
2099 #ifdef CONFIG_QUOTA
2100         case Opt_usrjquota:
2101                 if (!*param->string)
2102                         return unnote_qf_name(fc, USRQUOTA);
2103                 else
2104                         return note_qf_name(fc, USRQUOTA, param);
2105         case Opt_grpjquota:
2106                 if (!*param->string)
2107                         return unnote_qf_name(fc, GRPQUOTA);
2108                 else
2109                         return note_qf_name(fc, GRPQUOTA, param);
2110 #endif
2111         case Opt_sb:
2112                 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
2113                         ext4_msg(NULL, KERN_WARNING,
2114                                  "Ignoring %s option on remount", param->key);
2115                 } else {
2116                         ctx->s_sb_block = result.uint_32;
2117                         ctx->spec |= EXT4_SPEC_s_sb_block;
2118                 }
2119                 return 0;
2120         case Opt_removed:
2121                 ext4_msg(NULL, KERN_WARNING, "Ignoring removed %s option",
2122                          param->key);
2123                 return 0;
2124         case Opt_abort:
2125                 ctx_set_mount_flag(ctx, EXT4_MF_FS_ABORTED);
2126                 return 0;
2127         case Opt_inlinecrypt:
2128 #ifdef CONFIG_FS_ENCRYPTION_INLINE_CRYPT
2129                 ctx_set_flags(ctx, SB_INLINECRYPT);
2130 #else
2131                 ext4_msg(NULL, KERN_ERR, "inline encryption not supported");
2132 #endif
2133                 return 0;
2134         case Opt_errors:
2135                 ctx_clear_mount_opt(ctx, EXT4_MOUNT_ERRORS_MASK);
2136                 ctx_set_mount_opt(ctx, result.uint_32);
2137                 return 0;
2138 #ifdef CONFIG_QUOTA
2139         case Opt_jqfmt:
2140                 ctx->s_jquota_fmt = result.uint_32;
2141                 ctx->spec |= EXT4_SPEC_JQFMT;
2142                 return 0;
2143 #endif
2144         case Opt_data:
2145                 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS);
2146                 ctx_set_mount_opt(ctx, result.uint_32);
2147                 ctx->spec |= EXT4_SPEC_DATAJ;
2148                 return 0;
2149         case Opt_commit:
2150                 if (result.uint_32 == 0)
2151                         result.uint_32 = JBD2_DEFAULT_MAX_COMMIT_AGE;
2152                 else if (result.uint_32 > INT_MAX / HZ) {
2153                         ext4_msg(NULL, KERN_ERR,
2154                                  "Invalid commit interval %d, "
2155                                  "must be smaller than %d",
2156                                  result.uint_32, INT_MAX / HZ);
2157                         return -EINVAL;
2158                 }
2159                 ctx->s_commit_interval = HZ * result.uint_32;
2160                 ctx->spec |= EXT4_SPEC_s_commit_interval;
2161                 return 0;
2162         case Opt_debug_want_extra_isize:
2163                 if ((result.uint_32 & 1) || (result.uint_32 < 4)) {
2164                         ext4_msg(NULL, KERN_ERR,
2165                                  "Invalid want_extra_isize %d", result.uint_32);
2166                         return -EINVAL;
2167                 }
2168                 ctx->s_want_extra_isize = result.uint_32;
2169                 ctx->spec |= EXT4_SPEC_s_want_extra_isize;
2170                 return 0;
2171         case Opt_max_batch_time:
2172                 ctx->s_max_batch_time = result.uint_32;
2173                 ctx->spec |= EXT4_SPEC_s_max_batch_time;
2174                 return 0;
2175         case Opt_min_batch_time:
2176                 ctx->s_min_batch_time = result.uint_32;
2177                 ctx->spec |= EXT4_SPEC_s_min_batch_time;
2178                 return 0;
2179         case Opt_inode_readahead_blks:
2180                 if (result.uint_32 &&
2181                     (result.uint_32 > (1 << 30) ||
2182                      !is_power_of_2(result.uint_32))) {
2183                         ext4_msg(NULL, KERN_ERR,
2184                                  "EXT4-fs: inode_readahead_blks must be "
2185                                  "0 or a power of 2 smaller than 2^31");
2186                         return -EINVAL;
2187                 }
2188                 ctx->s_inode_readahead_blks = result.uint_32;
2189                 ctx->spec |= EXT4_SPEC_s_inode_readahead_blks;
2190                 return 0;
2191         case Opt_init_itable:
2192                 ctx_set_mount_opt(ctx, EXT4_MOUNT_INIT_INODE_TABLE);
2193                 ctx->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
2194                 if (param->type == fs_value_is_string)
2195                         ctx->s_li_wait_mult = result.uint_32;
2196                 ctx->spec |= EXT4_SPEC_s_li_wait_mult;
2197                 return 0;
2198         case Opt_max_dir_size_kb:
2199                 ctx->s_max_dir_size_kb = result.uint_32;
2200                 ctx->spec |= EXT4_SPEC_s_max_dir_size_kb;
2201                 return 0;
2202 #ifdef CONFIG_EXT4_DEBUG
2203         case Opt_fc_debug_max_replay:
2204                 ctx->s_fc_debug_max_replay = result.uint_32;
2205                 ctx->spec |= EXT4_SPEC_s_fc_debug_max_replay;
2206                 return 0;
2207 #endif
2208         case Opt_stripe:
2209                 ctx->s_stripe = result.uint_32;
2210                 ctx->spec |= EXT4_SPEC_s_stripe;
2211                 return 0;
2212         case Opt_resuid:
2213                 uid = make_kuid(current_user_ns(), result.uint_32);
2214                 if (!uid_valid(uid)) {
2215                         ext4_msg(NULL, KERN_ERR, "Invalid uid value %d",
2216                                  result.uint_32);
2217                         return -EINVAL;
2218                 }
2219                 ctx->s_resuid = uid;
2220                 ctx->spec |= EXT4_SPEC_s_resuid;
2221                 return 0;
2222         case Opt_resgid:
2223                 gid = make_kgid(current_user_ns(), result.uint_32);
2224                 if (!gid_valid(gid)) {
2225                         ext4_msg(NULL, KERN_ERR, "Invalid gid value %d",
2226                                  result.uint_32);
2227                         return -EINVAL;
2228                 }
2229                 ctx->s_resgid = gid;
2230                 ctx->spec |= EXT4_SPEC_s_resgid;
2231                 return 0;
2232         case Opt_journal_dev:
2233                 if (is_remount) {
2234                         ext4_msg(NULL, KERN_ERR,
2235                                  "Cannot specify journal on remount");
2236                         return -EINVAL;
2237                 }
2238                 ctx->journal_devnum = result.uint_32;
2239                 ctx->spec |= EXT4_SPEC_JOURNAL_DEV;
2240                 return 0;
2241         case Opt_journal_path:
2242         {
2243                 struct inode *journal_inode;
2244                 struct path path;
2245                 int error;
2246
2247                 if (is_remount) {
2248                         ext4_msg(NULL, KERN_ERR,
2249                                  "Cannot specify journal on remount");
2250                         return -EINVAL;
2251                 }
2252
2253                 error = fs_lookup_param(fc, param, 1, LOOKUP_FOLLOW, &path);
2254                 if (error) {
2255                         ext4_msg(NULL, KERN_ERR, "error: could not find "
2256                                  "journal device path");
2257                         return -EINVAL;
2258                 }
2259
2260                 journal_inode = d_inode(path.dentry);
2261                 ctx->journal_devnum = new_encode_dev(journal_inode->i_rdev);
2262                 ctx->spec |= EXT4_SPEC_JOURNAL_DEV;
2263                 path_put(&path);
2264                 return 0;
2265         }
2266         case Opt_journal_ioprio:
2267                 if (result.uint_32 > 7) {
2268                         ext4_msg(NULL, KERN_ERR, "Invalid journal IO priority"
2269                                  " (must be 0-7)");
2270                         return -EINVAL;
2271                 }
2272                 ctx->journal_ioprio =
2273                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, result.uint_32);
2274                 ctx->spec |= EXT4_SPEC_JOURNAL_IOPRIO;
2275                 return 0;
2276         case Opt_test_dummy_encryption:
2277                 return ext4_parse_test_dummy_encryption(param, ctx);
2278         case Opt_dax:
2279         case Opt_dax_type:
2280 #ifdef CONFIG_FS_DAX
2281         {
2282                 int type = (token == Opt_dax) ?
2283                            Opt_dax : result.uint_32;
2284
2285                 switch (type) {
2286                 case Opt_dax:
2287                 case Opt_dax_always:
2288                         ctx_set_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS);
2289                         ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER);
2290                         break;
2291                 case Opt_dax_never:
2292                         ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER);
2293                         ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS);
2294                         break;
2295                 case Opt_dax_inode:
2296                         ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS);
2297                         ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER);
2298                         /* Strictly for printing options */
2299                         ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE);
2300                         break;
2301                 }
2302                 return 0;
2303         }
2304 #else
2305                 ext4_msg(NULL, KERN_INFO, "dax option not supported");
2306                 return -EINVAL;
2307 #endif
2308         case Opt_data_err:
2309                 if (result.uint_32 == Opt_data_err_abort)
2310                         ctx_set_mount_opt(ctx, m->mount_opt);
2311                 else if (result.uint_32 == Opt_data_err_ignore)
2312                         ctx_clear_mount_opt(ctx, m->mount_opt);
2313                 return 0;
2314         case Opt_mb_optimize_scan:
2315                 if (result.int_32 == 1) {
2316                         ctx_set_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN);
2317                         ctx->spec |= EXT4_SPEC_mb_optimize_scan;
2318                 } else if (result.int_32 == 0) {
2319                         ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN);
2320                         ctx->spec |= EXT4_SPEC_mb_optimize_scan;
2321                 } else {
2322                         ext4_msg(NULL, KERN_WARNING,
2323                                  "mb_optimize_scan should be set to 0 or 1.");
2324                         return -EINVAL;
2325                 }
2326                 return 0;
2327         }
2328
2329         /*
2330          * At this point we should only be getting options requiring MOPT_SET,
2331          * or MOPT_CLEAR. Anything else is a bug
2332          */
2333         if (m->token == Opt_err) {
2334                 ext4_msg(NULL, KERN_WARNING, "buggy handling of option %s",
2335                          param->key);
2336                 WARN_ON(1);
2337                 return -EINVAL;
2338         }
2339
2340         else {
2341                 unsigned int set = 0;
2342
2343                 if ((param->type == fs_value_is_flag) ||
2344                     result.uint_32 > 0)
2345                         set = 1;
2346
2347                 if (m->flags & MOPT_CLEAR)
2348                         set = !set;
2349                 else if (unlikely(!(m->flags & MOPT_SET))) {
2350                         ext4_msg(NULL, KERN_WARNING,
2351                                  "buggy handling of option %s",
2352                                  param->key);
2353                         WARN_ON(1);
2354                         return -EINVAL;
2355                 }
2356                 if (m->flags & MOPT_2) {
2357                         if (set != 0)
2358                                 ctx_set_mount_opt2(ctx, m->mount_opt);
2359                         else
2360                                 ctx_clear_mount_opt2(ctx, m->mount_opt);
2361                 } else {
2362                         if (set != 0)
2363                                 ctx_set_mount_opt(ctx, m->mount_opt);
2364                         else
2365                                 ctx_clear_mount_opt(ctx, m->mount_opt);
2366                 }
2367         }
2368
2369         return 0;
2370 }
2371
2372 static int parse_options(struct fs_context *fc, char *options)
2373 {
2374         struct fs_parameter param;
2375         int ret;
2376         char *key;
2377
2378         if (!options)
2379                 return 0;
2380
2381         while ((key = strsep(&options, ",")) != NULL) {
2382                 if (*key) {
2383                         size_t v_len = 0;
2384                         char *value = strchr(key, '=');
2385
2386                         param.type = fs_value_is_flag;
2387                         param.string = NULL;
2388
2389                         if (value) {
2390                                 if (value == key)
2391                                         continue;
2392
2393                                 *value++ = 0;
2394                                 v_len = strlen(value);
2395                                 param.string = kmemdup_nul(value, v_len,
2396                                                            GFP_KERNEL);
2397                                 if (!param.string)
2398                                         return -ENOMEM;
2399                                 param.type = fs_value_is_string;
2400                         }
2401
2402                         param.key = key;
2403                         param.size = v_len;
2404
2405                         ret = ext4_parse_param(fc, &param);
2406                         if (param.string)
2407                                 kfree(param.string);
2408                         if (ret < 0)
2409                                 return ret;
2410                 }
2411         }
2412
2413         ret = ext4_validate_options(fc);
2414         if (ret < 0)
2415                 return ret;
2416
2417         return 0;
2418 }
2419
2420 static int parse_apply_sb_mount_options(struct super_block *sb,
2421                                         struct ext4_fs_context *m_ctx)
2422 {
2423         struct ext4_sb_info *sbi = EXT4_SB(sb);
2424         char *s_mount_opts = NULL;
2425         struct ext4_fs_context *s_ctx = NULL;
2426         struct fs_context *fc = NULL;
2427         int ret = -ENOMEM;
2428
2429         if (!sbi->s_es->s_mount_opts[0])
2430                 return 0;
2431
2432         s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
2433                                 sizeof(sbi->s_es->s_mount_opts),
2434                                 GFP_KERNEL);
2435         if (!s_mount_opts)
2436                 return ret;
2437
2438         fc = kzalloc(sizeof(struct fs_context), GFP_KERNEL);
2439         if (!fc)
2440                 goto out_free;
2441
2442         s_ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL);
2443         if (!s_ctx)
2444                 goto out_free;
2445
2446         fc->fs_private = s_ctx;
2447         fc->s_fs_info = sbi;
2448
2449         ret = parse_options(fc, s_mount_opts);
2450         if (ret < 0)
2451                 goto parse_failed;
2452
2453         ret = ext4_check_opt_consistency(fc, sb);
2454         if (ret < 0) {
2455 parse_failed:
2456                 ext4_msg(sb, KERN_WARNING,
2457                          "failed to parse options in superblock: %s",
2458                          s_mount_opts);
2459                 ret = 0;
2460                 goto out_free;
2461         }
2462
2463         if (s_ctx->spec & EXT4_SPEC_JOURNAL_DEV)
2464                 m_ctx->journal_devnum = s_ctx->journal_devnum;
2465         if (s_ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)
2466                 m_ctx->journal_ioprio = s_ctx->journal_ioprio;
2467
2468         ext4_apply_options(fc, sb);
2469         ret = 0;
2470
2471 out_free:
2472         if (fc) {
2473                 ext4_fc_free(fc);
2474                 kfree(fc);
2475         }
2476         kfree(s_mount_opts);
2477         return ret;
2478 }
2479
2480 static void ext4_apply_quota_options(struct fs_context *fc,
2481                                      struct super_block *sb)
2482 {
2483 #ifdef CONFIG_QUOTA
2484         bool quota_feature = ext4_has_feature_quota(sb);
2485         struct ext4_fs_context *ctx = fc->fs_private;
2486         struct ext4_sb_info *sbi = EXT4_SB(sb);
2487         char *qname;
2488         int i;
2489
2490         if (quota_feature)
2491                 return;
2492
2493         if (ctx->spec & EXT4_SPEC_JQUOTA) {
2494                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2495                         if (!(ctx->qname_spec & (1 << i)))
2496                                 continue;
2497
2498                         qname = ctx->s_qf_names[i]; /* May be NULL */
2499                         if (qname)
2500                                 set_opt(sb, QUOTA);
2501                         ctx->s_qf_names[i] = NULL;
2502                         qname = rcu_replace_pointer(sbi->s_qf_names[i], qname,
2503                                                 lockdep_is_held(&sb->s_umount));
2504                         if (qname)
2505                                 kfree_rcu(qname);
2506                 }
2507         }
2508
2509         if (ctx->spec & EXT4_SPEC_JQFMT)
2510                 sbi->s_jquota_fmt = ctx->s_jquota_fmt;
2511 #endif
2512 }
2513
2514 /*
2515  * Check quota settings consistency.
2516  */
2517 static int ext4_check_quota_consistency(struct fs_context *fc,
2518                                         struct super_block *sb)
2519 {
2520 #ifdef CONFIG_QUOTA
2521         struct ext4_fs_context *ctx = fc->fs_private;
2522         struct ext4_sb_info *sbi = EXT4_SB(sb);
2523         bool quota_feature = ext4_has_feature_quota(sb);
2524         bool quota_loaded = sb_any_quota_loaded(sb);
2525         bool usr_qf_name, grp_qf_name, usrquota, grpquota;
2526         int quota_flags, i;
2527
2528         /*
2529          * We do the test below only for project quotas. 'usrquota' and
2530          * 'grpquota' mount options are allowed even without quota feature
2531          * to support legacy quotas in quota files.
2532          */
2533         if (ctx_test_mount_opt(ctx, EXT4_MOUNT_PRJQUOTA) &&
2534             !ext4_has_feature_project(sb)) {
2535                 ext4_msg(NULL, KERN_ERR, "Project quota feature not enabled. "
2536                          "Cannot enable project quota enforcement.");
2537                 return -EINVAL;
2538         }
2539
2540         quota_flags = EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
2541                       EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA;
2542         if (quota_loaded &&
2543             ctx->mask_s_mount_opt & quota_flags &&
2544             !ctx_test_mount_opt(ctx, quota_flags))
2545                 goto err_quota_change;
2546
2547         if (ctx->spec & EXT4_SPEC_JQUOTA) {
2548
2549                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2550                         if (!(ctx->qname_spec & (1 << i)))
2551                                 continue;
2552
2553                         if (quota_loaded &&
2554                             !!sbi->s_qf_names[i] != !!ctx->s_qf_names[i])
2555                                 goto err_jquota_change;
2556
2557                         if (sbi->s_qf_names[i] && ctx->s_qf_names[i] &&
2558                             strcmp(get_qf_name(sb, sbi, i),
2559                                    ctx->s_qf_names[i]) != 0)
2560                                 goto err_jquota_specified;
2561                 }
2562
2563                 if (quota_feature) {
2564                         ext4_msg(NULL, KERN_INFO,
2565                                  "Journaled quota options ignored when "
2566                                  "QUOTA feature is enabled");
2567                         return 0;
2568                 }
2569         }
2570
2571         if (ctx->spec & EXT4_SPEC_JQFMT) {
2572                 if (sbi->s_jquota_fmt != ctx->s_jquota_fmt && quota_loaded)
2573                         goto err_jquota_change;
2574                 if (quota_feature) {
2575                         ext4_msg(NULL, KERN_INFO, "Quota format mount options "
2576                                  "ignored when QUOTA feature is enabled");
2577                         return 0;
2578                 }
2579         }
2580
2581         /* Make sure we don't mix old and new quota format */
2582         usr_qf_name = (get_qf_name(sb, sbi, USRQUOTA) ||
2583                        ctx->s_qf_names[USRQUOTA]);
2584         grp_qf_name = (get_qf_name(sb, sbi, GRPQUOTA) ||
2585                        ctx->s_qf_names[GRPQUOTA]);
2586
2587         usrquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) ||
2588                     test_opt(sb, USRQUOTA));
2589
2590         grpquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) ||
2591                     test_opt(sb, GRPQUOTA));
2592
2593         if (usr_qf_name) {
2594                 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA);
2595                 usrquota = false;
2596         }
2597         if (grp_qf_name) {
2598                 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA);
2599                 grpquota = false;
2600         }
2601
2602         if (usr_qf_name || grp_qf_name) {
2603                 if (usrquota || grpquota) {
2604                         ext4_msg(NULL, KERN_ERR, "old and new quota "
2605                                  "format mixing");
2606                         return -EINVAL;
2607                 }
2608
2609                 if (!(ctx->spec & EXT4_SPEC_JQFMT || sbi->s_jquota_fmt)) {
2610                         ext4_msg(NULL, KERN_ERR, "journaled quota format "
2611                                  "not specified");
2612                         return -EINVAL;
2613                 }
2614         }
2615
2616         return 0;
2617
2618 err_quota_change:
2619         ext4_msg(NULL, KERN_ERR,
2620                  "Cannot change quota options when quota turned on");
2621         return -EINVAL;
2622 err_jquota_change:
2623         ext4_msg(NULL, KERN_ERR, "Cannot change journaled quota "
2624                  "options when quota turned on");
2625         return -EINVAL;
2626 err_jquota_specified:
2627         ext4_msg(NULL, KERN_ERR, "%s quota file already specified",
2628                  QTYPE2NAME(i));
2629         return -EINVAL;
2630 #else
2631         return 0;
2632 #endif
2633 }
2634
2635 static int ext4_check_test_dummy_encryption(const struct fs_context *fc,
2636                                             struct super_block *sb)
2637 {
2638         const struct ext4_fs_context *ctx = fc->fs_private;
2639         const struct ext4_sb_info *sbi = EXT4_SB(sb);
2640         int err;
2641
2642         if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy))
2643                 return 0;
2644
2645         if (!ext4_has_feature_encrypt(sb)) {
2646                 ext4_msg(NULL, KERN_WARNING,
2647                          "test_dummy_encryption requires encrypt feature");
2648                 return -EINVAL;
2649         }
2650         /*
2651          * This mount option is just for testing, and it's not worthwhile to
2652          * implement the extra complexity (e.g. RCU protection) that would be
2653          * needed to allow it to be set or changed during remount.  We do allow
2654          * it to be specified during remount, but only if there is no change.
2655          */
2656         if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
2657                 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy,
2658                                                  &ctx->dummy_enc_policy))
2659                         return 0;
2660                 ext4_msg(NULL, KERN_WARNING,
2661                          "Can't set or change test_dummy_encryption on remount");
2662                 return -EINVAL;
2663         }
2664         /* Also make sure s_mount_opts didn't contain a conflicting value. */
2665         if (fscrypt_is_dummy_policy_set(&sbi->s_dummy_enc_policy)) {
2666                 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy,
2667                                                  &ctx->dummy_enc_policy))
2668                         return 0;
2669                 ext4_msg(NULL, KERN_WARNING,
2670                          "Conflicting test_dummy_encryption options");
2671                 return -EINVAL;
2672         }
2673         /*
2674          * fscrypt_add_test_dummy_key() technically changes the super_block, so
2675          * technically it should be delayed until ext4_apply_options() like the
2676          * other changes.  But since we never get here for remounts (see above),
2677          * and this is the last chance to report errors, we do it here.
2678          */
2679         err = fscrypt_add_test_dummy_key(sb, &ctx->dummy_enc_policy);
2680         if (err)
2681                 ext4_msg(NULL, KERN_WARNING,
2682                          "Error adding test dummy encryption key [%d]", err);
2683         return err;
2684 }
2685
2686 static void ext4_apply_test_dummy_encryption(struct ext4_fs_context *ctx,
2687                                              struct super_block *sb)
2688 {
2689         if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy) ||
2690             /* if already set, it was already verified to be the same */
2691             fscrypt_is_dummy_policy_set(&EXT4_SB(sb)->s_dummy_enc_policy))
2692                 return;
2693         EXT4_SB(sb)->s_dummy_enc_policy = ctx->dummy_enc_policy;
2694         memset(&ctx->dummy_enc_policy, 0, sizeof(ctx->dummy_enc_policy));
2695         ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled");
2696 }
2697
2698 static int ext4_check_opt_consistency(struct fs_context *fc,
2699                                       struct super_block *sb)
2700 {
2701         struct ext4_fs_context *ctx = fc->fs_private;
2702         struct ext4_sb_info *sbi = fc->s_fs_info;
2703         int is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE;
2704         int err;
2705
2706         if ((ctx->opt_flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
2707                 ext4_msg(NULL, KERN_ERR,
2708                          "Mount option(s) incompatible with ext2");
2709                 return -EINVAL;
2710         }
2711         if ((ctx->opt_flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
2712                 ext4_msg(NULL, KERN_ERR,
2713                          "Mount option(s) incompatible with ext3");
2714                 return -EINVAL;
2715         }
2716
2717         if (ctx->s_want_extra_isize >
2718             (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE)) {
2719                 ext4_msg(NULL, KERN_ERR,
2720                          "Invalid want_extra_isize %d",
2721                          ctx->s_want_extra_isize);
2722                 return -EINVAL;
2723         }
2724
2725         if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DIOREAD_NOLOCK)) {
2726                 int blocksize =
2727                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
2728                 if (blocksize < PAGE_SIZE)
2729                         ext4_msg(NULL, KERN_WARNING, "Warning: mounting with an "
2730                                  "experimental mount option 'dioread_nolock' "
2731                                  "for blocksize < PAGE_SIZE");
2732         }
2733
2734         err = ext4_check_test_dummy_encryption(fc, sb);
2735         if (err)
2736                 return err;
2737
2738         if ((ctx->spec & EXT4_SPEC_DATAJ) && is_remount) {
2739                 if (!sbi->s_journal) {
2740                         ext4_msg(NULL, KERN_WARNING,
2741                                  "Remounting file system with no journal "
2742                                  "so ignoring journalled data option");
2743                         ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS);
2744                 } else if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS) !=
2745                            test_opt(sb, DATA_FLAGS)) {
2746                         ext4_msg(NULL, KERN_ERR, "Cannot change data mode "
2747                                  "on remount");
2748                         return -EINVAL;
2749                 }
2750         }
2751
2752         if (is_remount) {
2753                 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) &&
2754                     (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) {
2755                         ext4_msg(NULL, KERN_ERR, "can't mount with "
2756                                  "both data=journal and dax");
2757                         return -EINVAL;
2758                 }
2759
2760                 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) &&
2761                     (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2762                      (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) {
2763 fail_dax_change_remount:
2764                         ext4_msg(NULL, KERN_ERR, "can't change "
2765                                  "dax mount option while remounting");
2766                         return -EINVAL;
2767                 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER) &&
2768                          (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2769                           (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS))) {
2770                         goto fail_dax_change_remount;
2771                 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE) &&
2772                            ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2773                             (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2774                             !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE))) {
2775                         goto fail_dax_change_remount;
2776                 }
2777         }
2778
2779         return ext4_check_quota_consistency(fc, sb);
2780 }
2781
2782 static void ext4_apply_options(struct fs_context *fc, struct super_block *sb)
2783 {
2784         struct ext4_fs_context *ctx = fc->fs_private;
2785         struct ext4_sb_info *sbi = fc->s_fs_info;
2786
2787         sbi->s_mount_opt &= ~ctx->mask_s_mount_opt;
2788         sbi->s_mount_opt |= ctx->vals_s_mount_opt;
2789         sbi->s_mount_opt2 &= ~ctx->mask_s_mount_opt2;
2790         sbi->s_mount_opt2 |= ctx->vals_s_mount_opt2;
2791         sbi->s_mount_flags &= ~ctx->mask_s_mount_flags;
2792         sbi->s_mount_flags |= ctx->vals_s_mount_flags;
2793         sb->s_flags &= ~ctx->mask_s_flags;
2794         sb->s_flags |= ctx->vals_s_flags;
2795
2796 #define APPLY(X) ({ if (ctx->spec & EXT4_SPEC_##X) sbi->X = ctx->X; })
2797         APPLY(s_commit_interval);
2798         APPLY(s_stripe);
2799         APPLY(s_max_batch_time);
2800         APPLY(s_min_batch_time);
2801         APPLY(s_want_extra_isize);
2802         APPLY(s_inode_readahead_blks);
2803         APPLY(s_max_dir_size_kb);
2804         APPLY(s_li_wait_mult);
2805         APPLY(s_resgid);
2806         APPLY(s_resuid);
2807
2808 #ifdef CONFIG_EXT4_DEBUG
2809         APPLY(s_fc_debug_max_replay);
2810 #endif
2811
2812         ext4_apply_quota_options(fc, sb);
2813         ext4_apply_test_dummy_encryption(ctx, sb);
2814 }
2815
2816
2817 static int ext4_validate_options(struct fs_context *fc)
2818 {
2819 #ifdef CONFIG_QUOTA
2820         struct ext4_fs_context *ctx = fc->fs_private;
2821         char *usr_qf_name, *grp_qf_name;
2822
2823         usr_qf_name = ctx->s_qf_names[USRQUOTA];
2824         grp_qf_name = ctx->s_qf_names[GRPQUOTA];
2825
2826         if (usr_qf_name || grp_qf_name) {
2827                 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) && usr_qf_name)
2828                         ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA);
2829
2830                 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) && grp_qf_name)
2831                         ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA);
2832
2833                 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) ||
2834                     ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA)) {
2835                         ext4_msg(NULL, KERN_ERR, "old and new quota "
2836                                  "format mixing");
2837                         return -EINVAL;
2838                 }
2839         }
2840 #endif
2841         return 1;
2842 }
2843
2844 static inline void ext4_show_quota_options(struct seq_file *seq,
2845                                            struct super_block *sb)
2846 {
2847 #if defined(CONFIG_QUOTA)
2848         struct ext4_sb_info *sbi = EXT4_SB(sb);
2849         char *usr_qf_name, *grp_qf_name;
2850
2851         if (sbi->s_jquota_fmt) {
2852                 char *fmtname = "";
2853
2854                 switch (sbi->s_jquota_fmt) {
2855                 case QFMT_VFS_OLD:
2856                         fmtname = "vfsold";
2857                         break;
2858                 case QFMT_VFS_V0:
2859                         fmtname = "vfsv0";
2860                         break;
2861                 case QFMT_VFS_V1:
2862                         fmtname = "vfsv1";
2863                         break;
2864                 }
2865                 seq_printf(seq, ",jqfmt=%s", fmtname);
2866         }
2867
2868         rcu_read_lock();
2869         usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]);
2870         grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]);
2871         if (usr_qf_name)
2872                 seq_show_option(seq, "usrjquota", usr_qf_name);
2873         if (grp_qf_name)
2874                 seq_show_option(seq, "grpjquota", grp_qf_name);
2875         rcu_read_unlock();
2876 #endif
2877 }
2878
2879 static const char *token2str(int token)
2880 {
2881         const struct fs_parameter_spec *spec;
2882
2883         for (spec = ext4_param_specs; spec->name != NULL; spec++)
2884                 if (spec->opt == token && !spec->type)
2885                         break;
2886         return spec->name;
2887 }
2888
2889 /*
2890  * Show an option if
2891  *  - it's set to a non-default value OR
2892  *  - if the per-sb default is different from the global default
2893  */
2894 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2895                               int nodefs)
2896 {
2897         struct ext4_sb_info *sbi = EXT4_SB(sb);
2898         struct ext4_super_block *es = sbi->s_es;
2899         int def_errors;
2900         const struct mount_opts *m;
2901         char sep = nodefs ? '\n' : ',';
2902
2903 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2904 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2905
2906         if (sbi->s_sb_block != 1)
2907                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2908
2909         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2910                 int want_set = m->flags & MOPT_SET;
2911                 int opt_2 = m->flags & MOPT_2;
2912                 unsigned int mount_opt, def_mount_opt;
2913
2914                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2915                     m->flags & MOPT_SKIP)
2916                         continue;
2917
2918                 if (opt_2) {
2919                         mount_opt = sbi->s_mount_opt2;
2920                         def_mount_opt = sbi->s_def_mount_opt2;
2921                 } else {
2922                         mount_opt = sbi->s_mount_opt;
2923                         def_mount_opt = sbi->s_def_mount_opt;
2924                 }
2925                 /* skip if same as the default */
2926                 if (!nodefs && !(m->mount_opt & (mount_opt ^ def_mount_opt)))
2927                         continue;
2928                 /* select Opt_noFoo vs Opt_Foo */
2929                 if ((want_set &&
2930                      (mount_opt & m->mount_opt) != m->mount_opt) ||
2931                     (!want_set && (mount_opt & m->mount_opt)))
2932                         continue;
2933                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2934         }
2935
2936         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2937             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2938                 SEQ_OPTS_PRINT("resuid=%u",
2939                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2940         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2941             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2942                 SEQ_OPTS_PRINT("resgid=%u",
2943                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2944         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2945         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2946                 SEQ_OPTS_PUTS("errors=remount-ro");
2947         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2948                 SEQ_OPTS_PUTS("errors=continue");
2949         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2950                 SEQ_OPTS_PUTS("errors=panic");
2951         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2952                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2953         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2954                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2955         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2956                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2957         if (nodefs || sbi->s_stripe)
2958                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2959         if (nodefs || EXT4_MOUNT_DATA_FLAGS &
2960                         (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
2961                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2962                         SEQ_OPTS_PUTS("data=journal");
2963                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2964                         SEQ_OPTS_PUTS("data=ordered");
2965                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2966                         SEQ_OPTS_PUTS("data=writeback");
2967         }
2968         if (nodefs ||
2969             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2970                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2971                                sbi->s_inode_readahead_blks);
2972
2973         if (test_opt(sb, INIT_INODE_TABLE) && (nodefs ||
2974                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2975                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2976         if (nodefs || sbi->s_max_dir_size_kb)
2977                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2978         if (test_opt(sb, DATA_ERR_ABORT))
2979                 SEQ_OPTS_PUTS("data_err=abort");
2980
2981         fscrypt_show_test_dummy_encryption(seq, sep, sb);
2982
2983         if (sb->s_flags & SB_INLINECRYPT)
2984                 SEQ_OPTS_PUTS("inlinecrypt");
2985
2986         if (test_opt(sb, DAX_ALWAYS)) {
2987                 if (IS_EXT2_SB(sb))
2988                         SEQ_OPTS_PUTS("dax");
2989                 else
2990                         SEQ_OPTS_PUTS("dax=always");
2991         } else if (test_opt2(sb, DAX_NEVER)) {
2992                 SEQ_OPTS_PUTS("dax=never");
2993         } else if (test_opt2(sb, DAX_INODE)) {
2994                 SEQ_OPTS_PUTS("dax=inode");
2995         }
2996
2997         if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD &&
2998                         !test_opt2(sb, MB_OPTIMIZE_SCAN)) {
2999                 SEQ_OPTS_PUTS("mb_optimize_scan=0");
3000         } else if (sbi->s_groups_count < MB_DEFAULT_LINEAR_SCAN_THRESHOLD &&
3001                         test_opt2(sb, MB_OPTIMIZE_SCAN)) {
3002                 SEQ_OPTS_PUTS("mb_optimize_scan=1");
3003         }
3004
3005         ext4_show_quota_options(seq, sb);
3006         return 0;
3007 }
3008
3009 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
3010 {
3011         return _ext4_show_options(seq, root->d_sb, 0);
3012 }
3013
3014 int ext4_seq_options_show(struct seq_file *seq, void *offset)
3015 {
3016         struct super_block *sb = seq->private;
3017         int rc;
3018
3019         seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
3020         rc = _ext4_show_options(seq, sb, 1);
3021         seq_puts(seq, "\n");
3022         return rc;
3023 }
3024
3025 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
3026                             int read_only)
3027 {
3028         struct ext4_sb_info *sbi = EXT4_SB(sb);
3029         int err = 0;
3030
3031         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
3032                 ext4_msg(sb, KERN_ERR, "revision level too high, "
3033                          "forcing read-only mode");
3034                 err = -EROFS;
3035                 goto done;
3036         }
3037         if (read_only)
3038                 goto done;
3039         if (!(sbi->s_mount_state & EXT4_VALID_FS))
3040                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
3041                          "running e2fsck is recommended");
3042         else if (sbi->s_mount_state & EXT4_ERROR_FS)
3043                 ext4_msg(sb, KERN_WARNING,
3044                          "warning: mounting fs with errors, "
3045                          "running e2fsck is recommended");
3046         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
3047                  le16_to_cpu(es->s_mnt_count) >=
3048                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
3049                 ext4_msg(sb, KERN_WARNING,
3050                          "warning: maximal mount count reached, "
3051                          "running e2fsck is recommended");
3052         else if (le32_to_cpu(es->s_checkinterval) &&
3053                  (ext4_get_tstamp(es, s_lastcheck) +
3054                   le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds()))
3055                 ext4_msg(sb, KERN_WARNING,
3056                          "warning: checktime reached, "
3057                          "running e2fsck is recommended");
3058         if (!sbi->s_journal)
3059                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
3060         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
3061                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
3062         le16_add_cpu(&es->s_mnt_count, 1);
3063         ext4_update_tstamp(es, s_mtime);
3064         if (sbi->s_journal) {
3065                 ext4_set_feature_journal_needs_recovery(sb);
3066                 if (ext4_has_feature_orphan_file(sb))
3067                         ext4_set_feature_orphan_present(sb);
3068         }
3069
3070         err = ext4_commit_super(sb);
3071 done:
3072         if (test_opt(sb, DEBUG))
3073                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
3074                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
3075                         sb->s_blocksize,
3076                         sbi->s_groups_count,
3077                         EXT4_BLOCKS_PER_GROUP(sb),
3078                         EXT4_INODES_PER_GROUP(sb),
3079                         sbi->s_mount_opt, sbi->s_mount_opt2);
3080         return err;
3081 }
3082
3083 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
3084 {
3085         struct ext4_sb_info *sbi = EXT4_SB(sb);
3086         struct flex_groups **old_groups, **new_groups;
3087         int size, i, j;
3088
3089         if (!sbi->s_log_groups_per_flex)
3090                 return 0;
3091
3092         size = ext4_flex_group(sbi, ngroup - 1) + 1;
3093         if (size <= sbi->s_flex_groups_allocated)
3094                 return 0;
3095
3096         new_groups = kvzalloc(roundup_pow_of_two(size *
3097                               sizeof(*sbi->s_flex_groups)), GFP_KERNEL);
3098         if (!new_groups) {
3099                 ext4_msg(sb, KERN_ERR,
3100                          "not enough memory for %d flex group pointers", size);
3101                 return -ENOMEM;
3102         }
3103         for (i = sbi->s_flex_groups_allocated; i < size; i++) {
3104                 new_groups[i] = kvzalloc(roundup_pow_of_two(
3105                                          sizeof(struct flex_groups)),
3106                                          GFP_KERNEL);
3107                 if (!new_groups[i]) {
3108                         for (j = sbi->s_flex_groups_allocated; j < i; j++)
3109                                 kvfree(new_groups[j]);
3110                         kvfree(new_groups);
3111                         ext4_msg(sb, KERN_ERR,
3112                                  "not enough memory for %d flex groups", size);
3113                         return -ENOMEM;
3114                 }
3115         }
3116         rcu_read_lock();
3117         old_groups = rcu_dereference(sbi->s_flex_groups);
3118         if (old_groups)
3119                 memcpy(new_groups, old_groups,
3120                        (sbi->s_flex_groups_allocated *
3121                         sizeof(struct flex_groups *)));
3122         rcu_read_unlock();
3123         rcu_assign_pointer(sbi->s_flex_groups, new_groups);
3124         sbi->s_flex_groups_allocated = size;
3125         if (old_groups)
3126                 ext4_kvfree_array_rcu(old_groups);
3127         return 0;
3128 }
3129
3130 static int ext4_fill_flex_info(struct super_block *sb)
3131 {
3132         struct ext4_sb_info *sbi = EXT4_SB(sb);
3133         struct ext4_group_desc *gdp = NULL;
3134         struct flex_groups *fg;
3135         ext4_group_t flex_group;
3136         int i, err;
3137
3138         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
3139         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
3140                 sbi->s_log_groups_per_flex = 0;
3141                 return 1;
3142         }
3143
3144         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
3145         if (err)
3146                 goto failed;
3147
3148         for (i = 0; i < sbi->s_groups_count; i++) {
3149                 gdp = ext4_get_group_desc(sb, i, NULL);
3150
3151                 flex_group = ext4_flex_group(sbi, i);
3152                 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
3153                 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes);
3154                 atomic64_add(ext4_free_group_clusters(sb, gdp),
3155                              &fg->free_clusters);
3156                 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs);
3157         }
3158
3159         return 1;
3160 failed:
3161         return 0;
3162 }
3163
3164 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
3165                                    struct ext4_group_desc *gdp)
3166 {
3167         int offset = offsetof(struct ext4_group_desc, bg_checksum);
3168         __u16 crc = 0;
3169         __le32 le_group = cpu_to_le32(block_group);
3170         struct ext4_sb_info *sbi = EXT4_SB(sb);
3171
3172         if (ext4_has_metadata_csum(sbi->s_sb)) {
3173                 /* Use new metadata_csum algorithm */
3174                 __u32 csum32;
3175                 __u16 dummy_csum = 0;
3176
3177                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
3178                                      sizeof(le_group));
3179                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
3180                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
3181                                      sizeof(dummy_csum));
3182                 offset += sizeof(dummy_csum);
3183                 if (offset < sbi->s_desc_size)
3184                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
3185                                              sbi->s_desc_size - offset);
3186
3187                 crc = csum32 & 0xFFFF;
3188                 goto out;
3189         }
3190
3191         /* old crc16 code */
3192         if (!ext4_has_feature_gdt_csum(sb))
3193                 return 0;
3194
3195         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
3196         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
3197         crc = crc16(crc, (__u8 *)gdp, offset);
3198         offset += sizeof(gdp->bg_checksum); /* skip checksum */
3199         /* for checksum of struct ext4_group_desc do the rest...*/
3200         if (ext4_has_feature_64bit(sb) && offset < sbi->s_desc_size)
3201                 crc = crc16(crc, (__u8 *)gdp + offset,
3202                             sbi->s_desc_size - offset);
3203
3204 out:
3205         return cpu_to_le16(crc);
3206 }
3207
3208 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
3209                                 struct ext4_group_desc *gdp)
3210 {
3211         if (ext4_has_group_desc_csum(sb) &&
3212             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
3213                 return 0;
3214
3215         return 1;
3216 }
3217
3218 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
3219                               struct ext4_group_desc *gdp)
3220 {
3221         if (!ext4_has_group_desc_csum(sb))
3222                 return;
3223         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
3224 }
3225
3226 /* Called at mount-time, super-block is locked */
3227 static int ext4_check_descriptors(struct super_block *sb,
3228                                   ext4_fsblk_t sb_block,
3229                                   ext4_group_t *first_not_zeroed)
3230 {
3231         struct ext4_sb_info *sbi = EXT4_SB(sb);
3232         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
3233         ext4_fsblk_t last_block;
3234         ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0);
3235         ext4_fsblk_t block_bitmap;
3236         ext4_fsblk_t inode_bitmap;
3237         ext4_fsblk_t inode_table;
3238         int flexbg_flag = 0;
3239         ext4_group_t i, grp = sbi->s_groups_count;
3240
3241         if (ext4_has_feature_flex_bg(sb))
3242                 flexbg_flag = 1;
3243
3244         ext4_debug("Checking group descriptors");
3245
3246         for (i = 0; i < sbi->s_groups_count; i++) {
3247                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
3248
3249                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
3250                         last_block = ext4_blocks_count(sbi->s_es) - 1;
3251                 else
3252                         last_block = first_block +
3253                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
3254
3255                 if ((grp == sbi->s_groups_count) &&
3256                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3257                         grp = i;
3258
3259                 block_bitmap = ext4_block_bitmap(sb, gdp);
3260                 if (block_bitmap == sb_block) {
3261                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3262                                  "Block bitmap for group %u overlaps "
3263                                  "superblock", i);
3264                         if (!sb_rdonly(sb))
3265                                 return 0;
3266                 }
3267                 if (block_bitmap >= sb_block + 1 &&
3268                     block_bitmap <= last_bg_block) {
3269                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3270                                  "Block bitmap for group %u overlaps "
3271                                  "block group descriptors", i);
3272                         if (!sb_rdonly(sb))
3273                                 return 0;
3274                 }
3275                 if (block_bitmap < first_block || block_bitmap > last_block) {
3276                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3277                                "Block bitmap for group %u not in group "
3278                                "(block %llu)!", i, block_bitmap);
3279                         return 0;
3280                 }
3281                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
3282                 if (inode_bitmap == sb_block) {
3283                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3284                                  "Inode bitmap for group %u overlaps "
3285                                  "superblock", i);
3286                         if (!sb_rdonly(sb))
3287                                 return 0;
3288                 }
3289                 if (inode_bitmap >= sb_block + 1 &&
3290                     inode_bitmap <= last_bg_block) {
3291                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3292                                  "Inode bitmap for group %u overlaps "
3293                                  "block group descriptors", i);
3294                         if (!sb_rdonly(sb))
3295                                 return 0;
3296                 }
3297                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
3298                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3299                                "Inode bitmap for group %u not in group "
3300                                "(block %llu)!", i, inode_bitmap);
3301                         return 0;
3302                 }
3303                 inode_table = ext4_inode_table(sb, gdp);
3304                 if (inode_table == sb_block) {
3305                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3306                                  "Inode table for group %u overlaps "
3307                                  "superblock", i);
3308                         if (!sb_rdonly(sb))
3309                                 return 0;
3310                 }
3311                 if (inode_table >= sb_block + 1 &&
3312                     inode_table <= last_bg_block) {
3313                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3314                                  "Inode table for group %u overlaps "
3315                                  "block group descriptors", i);
3316                         if (!sb_rdonly(sb))
3317                                 return 0;
3318                 }
3319                 if (inode_table < first_block ||
3320                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
3321                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3322                                "Inode table for group %u not in group "
3323                                "(block %llu)!", i, inode_table);
3324                         return 0;
3325                 }
3326                 ext4_lock_group(sb, i);
3327                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
3328                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3329                                  "Checksum for group %u failed (%u!=%u)",
3330                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
3331                                      gdp)), le16_to_cpu(gdp->bg_checksum));
3332                         if (!sb_rdonly(sb)) {
3333                                 ext4_unlock_group(sb, i);
3334                                 return 0;
3335                         }
3336                 }
3337                 ext4_unlock_group(sb, i);
3338                 if (!flexbg_flag)
3339                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
3340         }
3341         if (NULL != first_not_zeroed)
3342                 *first_not_zeroed = grp;
3343         return 1;
3344 }
3345
3346 /*
3347  * Maximal extent format file size.
3348  * Resulting logical blkno at s_maxbytes must fit in our on-disk
3349  * extent format containers, within a sector_t, and within i_blocks
3350  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
3351  * so that won't be a limiting factor.
3352  *
3353  * However there is other limiting factor. We do store extents in the form
3354  * of starting block and length, hence the resulting length of the extent
3355  * covering maximum file size must fit into on-disk format containers as
3356  * well. Given that length is always by 1 unit bigger than max unit (because
3357  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
3358  *
3359  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
3360  */
3361 static loff_t ext4_max_size(int blkbits, int has_huge_files)
3362 {
3363         loff_t res;
3364         loff_t upper_limit = MAX_LFS_FILESIZE;
3365
3366         BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64));
3367
3368         if (!has_huge_files) {
3369                 upper_limit = (1LL << 32) - 1;
3370
3371                 /* total blocks in file system block size */
3372                 upper_limit >>= (blkbits - 9);
3373                 upper_limit <<= blkbits;
3374         }
3375
3376         /*
3377          * 32-bit extent-start container, ee_block. We lower the maxbytes
3378          * by one fs block, so ee_len can cover the extent of maximum file
3379          * size
3380          */
3381         res = (1LL << 32) - 1;
3382         res <<= blkbits;
3383
3384         /* Sanity check against vm- & vfs- imposed limits */
3385         if (res > upper_limit)
3386                 res = upper_limit;
3387
3388         return res;
3389 }
3390
3391 /*
3392  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
3393  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
3394  * We need to be 1 filesystem block less than the 2^48 sector limit.
3395  */
3396 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
3397 {
3398         loff_t upper_limit, res = EXT4_NDIR_BLOCKS;
3399         int meta_blocks;
3400         unsigned int ppb = 1 << (bits - 2);
3401
3402         /*
3403          * This is calculated to be the largest file size for a dense, block
3404          * mapped file such that the file's total number of 512-byte sectors,
3405          * including data and all indirect blocks, does not exceed (2^48 - 1).
3406          *
3407          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
3408          * number of 512-byte sectors of the file.
3409          */
3410         if (!has_huge_files) {
3411                 /*
3412                  * !has_huge_files or implies that the inode i_block field
3413                  * represents total file blocks in 2^32 512-byte sectors ==
3414                  * size of vfs inode i_blocks * 8
3415                  */
3416                 upper_limit = (1LL << 32) - 1;
3417
3418                 /* total blocks in file system block size */
3419                 upper_limit >>= (bits - 9);
3420
3421         } else {
3422                 /*
3423                  * We use 48 bit ext4_inode i_blocks
3424                  * With EXT4_HUGE_FILE_FL set the i_blocks
3425                  * represent total number of blocks in
3426                  * file system block size
3427                  */
3428                 upper_limit = (1LL << 48) - 1;
3429
3430         }
3431
3432         /* Compute how many blocks we can address by block tree */
3433         res += ppb;
3434         res += ppb * ppb;
3435         res += ((loff_t)ppb) * ppb * ppb;
3436         /* Compute how many metadata blocks are needed */
3437         meta_blocks = 1;
3438         meta_blocks += 1 + ppb;
3439         meta_blocks += 1 + ppb + ppb * ppb;
3440         /* Does block tree limit file size? */
3441         if (res + meta_blocks <= upper_limit)
3442                 goto check_lfs;
3443
3444         res = upper_limit;
3445         /* How many metadata blocks are needed for addressing upper_limit? */
3446         upper_limit -= EXT4_NDIR_BLOCKS;
3447         /* indirect blocks */
3448         meta_blocks = 1;
3449         upper_limit -= ppb;
3450         /* double indirect blocks */
3451         if (upper_limit < ppb * ppb) {
3452                 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb);
3453                 res -= meta_blocks;
3454                 goto check_lfs;
3455         }
3456         meta_blocks += 1 + ppb;
3457         upper_limit -= ppb * ppb;
3458         /* tripple indirect blocks for the rest */
3459         meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb) +
3460                 DIV_ROUND_UP_ULL(upper_limit, ppb*ppb);
3461         res -= meta_blocks;
3462 check_lfs:
3463         res <<= bits;
3464         if (res > MAX_LFS_FILESIZE)
3465                 res = MAX_LFS_FILESIZE;
3466
3467         return res;
3468 }
3469
3470 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
3471                                    ext4_fsblk_t logical_sb_block, int nr)
3472 {
3473         struct ext4_sb_info *sbi = EXT4_SB(sb);
3474         ext4_group_t bg, first_meta_bg;
3475         int has_super = 0;
3476
3477         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
3478
3479         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
3480                 return logical_sb_block + nr + 1;
3481         bg = sbi->s_desc_per_block * nr;
3482         if (ext4_bg_has_super(sb, bg))
3483                 has_super = 1;
3484
3485         /*
3486          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
3487          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
3488          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
3489          * compensate.
3490          */
3491         if (sb->s_blocksize == 1024 && nr == 0 &&
3492             le32_to_cpu(sbi->s_es->s_first_data_block) == 0)
3493                 has_super++;
3494
3495         return (has_super + ext4_group_first_block_no(sb, bg));
3496 }
3497
3498 /**
3499  * ext4_get_stripe_size: Get the stripe size.
3500  * @sbi: In memory super block info
3501  *
3502  * If we have specified it via mount option, then
3503  * use the mount option value. If the value specified at mount time is
3504  * greater than the blocks per group use the super block value.
3505  * If the super block value is greater than blocks per group return 0.
3506  * Allocator needs it be less than blocks per group.
3507  *
3508  */
3509 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
3510 {
3511         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
3512         unsigned long stripe_width =
3513                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
3514         int ret;
3515
3516         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
3517                 ret = sbi->s_stripe;
3518         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
3519                 ret = stripe_width;
3520         else if (stride && stride <= sbi->s_blocks_per_group)
3521                 ret = stride;
3522         else
3523                 ret = 0;
3524
3525         /*
3526          * If the stripe width is 1, this makes no sense and
3527          * we set it to 0 to turn off stripe handling code.
3528          */
3529         if (ret <= 1)
3530                 ret = 0;
3531
3532         return ret;
3533 }
3534
3535 /*
3536  * Check whether this filesystem can be mounted based on
3537  * the features present and the RDONLY/RDWR mount requested.
3538  * Returns 1 if this filesystem can be mounted as requested,
3539  * 0 if it cannot be.
3540  */
3541 int ext4_feature_set_ok(struct super_block *sb, int readonly)
3542 {
3543         if (ext4_has_unknown_ext4_incompat_features(sb)) {
3544                 ext4_msg(sb, KERN_ERR,
3545                         "Couldn't mount because of "
3546                         "unsupported optional features (%x)",
3547                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
3548                         ~EXT4_FEATURE_INCOMPAT_SUPP));
3549                 return 0;
3550         }
3551
3552 #if !IS_ENABLED(CONFIG_UNICODE)
3553         if (ext4_has_feature_casefold(sb)) {
3554                 ext4_msg(sb, KERN_ERR,
3555                          "Filesystem with casefold feature cannot be "
3556                          "mounted without CONFIG_UNICODE");
3557                 return 0;
3558         }
3559 #endif
3560
3561         if (readonly)
3562                 return 1;
3563
3564         if (ext4_has_feature_readonly(sb)) {
3565                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
3566                 sb->s_flags |= SB_RDONLY;
3567                 return 1;
3568         }
3569
3570         /* Check that feature set is OK for a read-write mount */
3571         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
3572                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
3573                          "unsupported optional features (%x)",
3574                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
3575                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
3576                 return 0;
3577         }
3578         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
3579                 ext4_msg(sb, KERN_ERR,
3580                          "Can't support bigalloc feature without "
3581                          "extents feature\n");
3582                 return 0;
3583         }
3584
3585 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2)
3586         if (!readonly && (ext4_has_feature_quota(sb) ||
3587                           ext4_has_feature_project(sb))) {
3588                 ext4_msg(sb, KERN_ERR,
3589                          "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2");
3590                 return 0;
3591         }
3592 #endif  /* CONFIG_QUOTA */
3593         return 1;
3594 }
3595
3596 /*
3597  * This function is called once a day if we have errors logged
3598  * on the file system
3599  */
3600 static void print_daily_error_info(struct timer_list *t)
3601 {
3602         struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report);
3603         struct super_block *sb = sbi->s_sb;
3604         struct ext4_super_block *es = sbi->s_es;
3605
3606         if (es->s_error_count)
3607                 /* fsck newer than v1.41.13 is needed to clean this condition. */
3608                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
3609                          le32_to_cpu(es->s_error_count));
3610         if (es->s_first_error_time) {
3611                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d",
3612                        sb->s_id,
3613                        ext4_get_tstamp(es, s_first_error_time),
3614                        (int) sizeof(es->s_first_error_func),
3615                        es->s_first_error_func,
3616                        le32_to_cpu(es->s_first_error_line));
3617                 if (es->s_first_error_ino)
3618                         printk(KERN_CONT ": inode %u",
3619                                le32_to_cpu(es->s_first_error_ino));
3620                 if (es->s_first_error_block)
3621                         printk(KERN_CONT ": block %llu", (unsigned long long)
3622                                le64_to_cpu(es->s_first_error_block));
3623                 printk(KERN_CONT "\n");
3624         }
3625         if (es->s_last_error_time) {
3626                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d",
3627                        sb->s_id,
3628                        ext4_get_tstamp(es, s_last_error_time),
3629                        (int) sizeof(es->s_last_error_func),
3630                        es->s_last_error_func,
3631                        le32_to_cpu(es->s_last_error_line));
3632                 if (es->s_last_error_ino)
3633                         printk(KERN_CONT ": inode %u",
3634                                le32_to_cpu(es->s_last_error_ino));
3635                 if (es->s_last_error_block)
3636                         printk(KERN_CONT ": block %llu", (unsigned long long)
3637                                le64_to_cpu(es->s_last_error_block));
3638                 printk(KERN_CONT "\n");
3639         }
3640         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
3641 }
3642
3643 /* Find next suitable group and run ext4_init_inode_table */
3644 static int ext4_run_li_request(struct ext4_li_request *elr)
3645 {
3646         struct ext4_group_desc *gdp = NULL;
3647         struct super_block *sb = elr->lr_super;
3648         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3649         ext4_group_t group = elr->lr_next_group;
3650         unsigned int prefetch_ios = 0;
3651         int ret = 0;
3652         u64 start_time;
3653
3654         if (elr->lr_mode == EXT4_LI_MODE_PREFETCH_BBITMAP) {
3655                 elr->lr_next_group = ext4_mb_prefetch(sb, group,
3656                                 EXT4_SB(sb)->s_mb_prefetch, &prefetch_ios);
3657                 if (prefetch_ios)
3658                         ext4_mb_prefetch_fini(sb, elr->lr_next_group,
3659                                               prefetch_ios);
3660                 trace_ext4_prefetch_bitmaps(sb, group, elr->lr_next_group,
3661                                             prefetch_ios);
3662                 if (group >= elr->lr_next_group) {
3663                         ret = 1;
3664                         if (elr->lr_first_not_zeroed != ngroups &&
3665                             !sb_rdonly(sb) && test_opt(sb, INIT_INODE_TABLE)) {
3666                                 elr->lr_next_group = elr->lr_first_not_zeroed;
3667                                 elr->lr_mode = EXT4_LI_MODE_ITABLE;
3668                                 ret = 0;
3669                         }
3670                 }
3671                 return ret;
3672         }
3673
3674         for (; group < ngroups; group++) {
3675                 gdp = ext4_get_group_desc(sb, group, NULL);
3676                 if (!gdp) {
3677                         ret = 1;
3678                         break;
3679                 }
3680
3681                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3682                         break;
3683         }
3684
3685         if (group >= ngroups)
3686                 ret = 1;
3687
3688         if (!ret) {
3689                 start_time = ktime_get_real_ns();
3690                 ret = ext4_init_inode_table(sb, group,
3691                                             elr->lr_timeout ? 0 : 1);
3692                 trace_ext4_lazy_itable_init(sb, group);
3693                 if (elr->lr_timeout == 0) {
3694                         elr->lr_timeout = nsecs_to_jiffies((ktime_get_real_ns() - start_time) *
3695                                 EXT4_SB(elr->lr_super)->s_li_wait_mult);
3696                 }
3697                 elr->lr_next_sched = jiffies + elr->lr_timeout;
3698                 elr->lr_next_group = group + 1;
3699         }
3700         return ret;
3701 }
3702
3703 /*
3704  * Remove lr_request from the list_request and free the
3705  * request structure. Should be called with li_list_mtx held
3706  */
3707 static void ext4_remove_li_request(struct ext4_li_request *elr)
3708 {
3709         if (!elr)
3710                 return;
3711
3712         list_del(&elr->lr_request);
3713         EXT4_SB(elr->lr_super)->s_li_request = NULL;
3714         kfree(elr);
3715 }
3716
3717 static void ext4_unregister_li_request(struct super_block *sb)
3718 {
3719         mutex_lock(&ext4_li_mtx);
3720         if (!ext4_li_info) {
3721                 mutex_unlock(&ext4_li_mtx);
3722                 return;
3723         }
3724
3725         mutex_lock(&ext4_li_info->li_list_mtx);
3726         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
3727         mutex_unlock(&ext4_li_info->li_list_mtx);
3728         mutex_unlock(&ext4_li_mtx);
3729 }
3730
3731 static struct task_struct *ext4_lazyinit_task;
3732
3733 /*
3734  * This is the function where ext4lazyinit thread lives. It walks
3735  * through the request list searching for next scheduled filesystem.
3736  * When such a fs is found, run the lazy initialization request
3737  * (ext4_rn_li_request) and keep track of the time spend in this
3738  * function. Based on that time we compute next schedule time of
3739  * the request. When walking through the list is complete, compute
3740  * next waking time and put itself into sleep.
3741  */
3742 static int ext4_lazyinit_thread(void *arg)
3743 {
3744         struct ext4_lazy_init *eli = arg;
3745         struct list_head *pos, *n;
3746         struct ext4_li_request *elr;
3747         unsigned long next_wakeup, cur;
3748
3749         BUG_ON(NULL == eli);
3750         set_freezable();
3751
3752 cont_thread:
3753         while (true) {
3754                 next_wakeup = MAX_JIFFY_OFFSET;
3755
3756                 mutex_lock(&eli->li_list_mtx);
3757                 if (list_empty(&eli->li_request_list)) {
3758                         mutex_unlock(&eli->li_list_mtx);
3759                         goto exit_thread;
3760                 }
3761                 list_for_each_safe(pos, n, &eli->li_request_list) {
3762                         int err = 0;
3763                         int progress = 0;
3764                         elr = list_entry(pos, struct ext4_li_request,
3765                                          lr_request);
3766
3767                         if (time_before(jiffies, elr->lr_next_sched)) {
3768                                 if (time_before(elr->lr_next_sched, next_wakeup))
3769                                         next_wakeup = elr->lr_next_sched;
3770                                 continue;
3771                         }
3772                         if (down_read_trylock(&elr->lr_super->s_umount)) {
3773                                 if (sb_start_write_trylock(elr->lr_super)) {
3774                                         progress = 1;
3775                                         /*
3776                                          * We hold sb->s_umount, sb can not
3777                                          * be removed from the list, it is
3778                                          * now safe to drop li_list_mtx
3779                                          */
3780                                         mutex_unlock(&eli->li_list_mtx);
3781                                         err = ext4_run_li_request(elr);
3782                                         sb_end_write(elr->lr_super);
3783                                         mutex_lock(&eli->li_list_mtx);
3784                                         n = pos->next;
3785                                 }
3786                                 up_read((&elr->lr_super->s_umount));
3787                         }
3788                         /* error, remove the lazy_init job */
3789                         if (err) {
3790                                 ext4_remove_li_request(elr);
3791                                 continue;
3792                         }
3793                         if (!progress) {
3794                                 elr->lr_next_sched = jiffies +
3795                                         prandom_u32_max(EXT4_DEF_LI_MAX_START_DELAY * HZ);
3796                         }
3797                         if (time_before(elr->lr_next_sched, next_wakeup))
3798                                 next_wakeup = elr->lr_next_sched;
3799                 }
3800                 mutex_unlock(&eli->li_list_mtx);
3801
3802                 try_to_freeze();
3803
3804                 cur = jiffies;
3805                 if ((time_after_eq(cur, next_wakeup)) ||
3806                     (MAX_JIFFY_OFFSET == next_wakeup)) {
3807                         cond_resched();
3808                         continue;
3809                 }
3810
3811                 schedule_timeout_interruptible(next_wakeup - cur);
3812
3813                 if (kthread_should_stop()) {
3814                         ext4_clear_request_list();
3815                         goto exit_thread;
3816                 }
3817         }
3818
3819 exit_thread:
3820         /*
3821          * It looks like the request list is empty, but we need
3822          * to check it under the li_list_mtx lock, to prevent any
3823          * additions into it, and of course we should lock ext4_li_mtx
3824          * to atomically free the list and ext4_li_info, because at
3825          * this point another ext4 filesystem could be registering
3826          * new one.
3827          */
3828         mutex_lock(&ext4_li_mtx);
3829         mutex_lock(&eli->li_list_mtx);
3830         if (!list_empty(&eli->li_request_list)) {
3831                 mutex_unlock(&eli->li_list_mtx);
3832                 mutex_unlock(&ext4_li_mtx);
3833                 goto cont_thread;
3834         }
3835         mutex_unlock(&eli->li_list_mtx);
3836         kfree(ext4_li_info);
3837         ext4_li_info = NULL;
3838         mutex_unlock(&ext4_li_mtx);
3839
3840         return 0;
3841 }
3842
3843 static void ext4_clear_request_list(void)
3844 {
3845         struct list_head *pos, *n;
3846         struct ext4_li_request *elr;
3847
3848         mutex_lock(&ext4_li_info->li_list_mtx);
3849         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3850                 elr = list_entry(pos, struct ext4_li_request,
3851                                  lr_request);
3852                 ext4_remove_li_request(elr);
3853         }
3854         mutex_unlock(&ext4_li_info->li_list_mtx);
3855 }
3856
3857 static int ext4_run_lazyinit_thread(void)
3858 {
3859         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3860                                          ext4_li_info, "ext4lazyinit");
3861         if (IS_ERR(ext4_lazyinit_task)) {
3862                 int err = PTR_ERR(ext4_lazyinit_task);
3863                 ext4_clear_request_list();
3864                 kfree(ext4_li_info);
3865                 ext4_li_info = NULL;
3866                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3867                                  "initialization thread\n",
3868                                  err);
3869                 return err;
3870         }
3871         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3872         return 0;
3873 }
3874
3875 /*
3876  * Check whether it make sense to run itable init. thread or not.
3877  * If there is at least one uninitialized inode table, return
3878  * corresponding group number, else the loop goes through all
3879  * groups and return total number of groups.
3880  */
3881 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3882 {
3883         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3884         struct ext4_group_desc *gdp = NULL;
3885
3886         if (!ext4_has_group_desc_csum(sb))
3887                 return ngroups;
3888
3889         for (group = 0; group < ngroups; group++) {
3890                 gdp = ext4_get_group_desc(sb, group, NULL);
3891                 if (!gdp)
3892                         continue;
3893
3894                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3895                         break;
3896         }
3897
3898         return group;
3899 }
3900
3901 static int ext4_li_info_new(void)
3902 {
3903         struct ext4_lazy_init *eli = NULL;
3904
3905         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3906         if (!eli)
3907                 return -ENOMEM;
3908
3909         INIT_LIST_HEAD(&eli->li_request_list);
3910         mutex_init(&eli->li_list_mtx);
3911
3912         eli->li_state |= EXT4_LAZYINIT_QUIT;
3913
3914         ext4_li_info = eli;
3915
3916         return 0;
3917 }
3918
3919 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3920                                             ext4_group_t start)
3921 {
3922         struct ext4_li_request *elr;
3923
3924         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3925         if (!elr)
3926                 return NULL;
3927
3928         elr->lr_super = sb;
3929         elr->lr_first_not_zeroed = start;
3930         if (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS)) {
3931                 elr->lr_mode = EXT4_LI_MODE_ITABLE;
3932                 elr->lr_next_group = start;
3933         } else {
3934                 elr->lr_mode = EXT4_LI_MODE_PREFETCH_BBITMAP;
3935         }
3936
3937         /*
3938          * Randomize first schedule time of the request to
3939          * spread the inode table initialization requests
3940          * better.
3941          */
3942         elr->lr_next_sched = jiffies + prandom_u32_max(
3943                                 EXT4_DEF_LI_MAX_START_DELAY * HZ);
3944         return elr;
3945 }
3946
3947 int ext4_register_li_request(struct super_block *sb,
3948                              ext4_group_t first_not_zeroed)
3949 {
3950         struct ext4_sb_info *sbi = EXT4_SB(sb);
3951         struct ext4_li_request *elr = NULL;
3952         ext4_group_t ngroups = sbi->s_groups_count;
3953         int ret = 0;
3954
3955         mutex_lock(&ext4_li_mtx);
3956         if (sbi->s_li_request != NULL) {
3957                 /*
3958                  * Reset timeout so it can be computed again, because
3959                  * s_li_wait_mult might have changed.
3960                  */
3961                 sbi->s_li_request->lr_timeout = 0;
3962                 goto out;
3963         }
3964
3965         if (sb_rdonly(sb) ||
3966             (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS) &&
3967              (first_not_zeroed == ngroups || !test_opt(sb, INIT_INODE_TABLE))))
3968                 goto out;
3969
3970         elr = ext4_li_request_new(sb, first_not_zeroed);
3971         if (!elr) {
3972                 ret = -ENOMEM;
3973                 goto out;
3974         }
3975
3976         if (NULL == ext4_li_info) {
3977                 ret = ext4_li_info_new();
3978                 if (ret)
3979                         goto out;
3980         }
3981
3982         mutex_lock(&ext4_li_info->li_list_mtx);
3983         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3984         mutex_unlock(&ext4_li_info->li_list_mtx);
3985
3986         sbi->s_li_request = elr;
3987         /*
3988          * set elr to NULL here since it has been inserted to
3989          * the request_list and the removal and free of it is
3990          * handled by ext4_clear_request_list from now on.
3991          */
3992         elr = NULL;
3993
3994         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3995                 ret = ext4_run_lazyinit_thread();
3996                 if (ret)
3997                         goto out;
3998         }
3999 out:
4000         mutex_unlock(&ext4_li_mtx);
4001         if (ret)
4002                 kfree(elr);
4003         return ret;
4004 }
4005
4006 /*
4007  * We do not need to lock anything since this is called on
4008  * module unload.
4009  */
4010 static void ext4_destroy_lazyinit_thread(void)
4011 {
4012         /*
4013          * If thread exited earlier
4014          * there's nothing to be done.
4015          */
4016         if (!ext4_li_info || !ext4_lazyinit_task)
4017                 return;
4018
4019         kthread_stop(ext4_lazyinit_task);
4020 }
4021
4022 static int set_journal_csum_feature_set(struct super_block *sb)
4023 {
4024         int ret = 1;
4025         int compat, incompat;
4026         struct ext4_sb_info *sbi = EXT4_SB(sb);
4027
4028         if (ext4_has_metadata_csum(sb)) {
4029                 /* journal checksum v3 */
4030                 compat = 0;
4031                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
4032         } else {
4033                 /* journal checksum v1 */
4034                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
4035                 incompat = 0;
4036         }
4037
4038         jbd2_journal_clear_features(sbi->s_journal,
4039                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
4040                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
4041                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
4042         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4043                 ret = jbd2_journal_set_features(sbi->s_journal,
4044                                 compat, 0,
4045                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
4046                                 incompat);
4047         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
4048                 ret = jbd2_journal_set_features(sbi->s_journal,
4049                                 compat, 0,
4050                                 incompat);
4051                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
4052                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
4053         } else {
4054                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
4055                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
4056         }
4057
4058         return ret;
4059 }
4060
4061 /*
4062  * Note: calculating the overhead so we can be compatible with
4063  * historical BSD practice is quite difficult in the face of
4064  * clusters/bigalloc.  This is because multiple metadata blocks from
4065  * different block group can end up in the same allocation cluster.
4066  * Calculating the exact overhead in the face of clustered allocation
4067  * requires either O(all block bitmaps) in memory or O(number of block
4068  * groups**2) in time.  We will still calculate the superblock for
4069  * older file systems --- and if we come across with a bigalloc file
4070  * system with zero in s_overhead_clusters the estimate will be close to
4071  * correct especially for very large cluster sizes --- but for newer
4072  * file systems, it's better to calculate this figure once at mkfs
4073  * time, and store it in the superblock.  If the superblock value is
4074  * present (even for non-bigalloc file systems), we will use it.
4075  */
4076 static int count_overhead(struct super_block *sb, ext4_group_t grp,
4077                           char *buf)
4078 {
4079         struct ext4_sb_info     *sbi = EXT4_SB(sb);
4080         struct ext4_group_desc  *gdp;
4081         ext4_fsblk_t            first_block, last_block, b;
4082         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
4083         int                     s, j, count = 0;
4084         int                     has_super = ext4_bg_has_super(sb, grp);
4085
4086         if (!ext4_has_feature_bigalloc(sb))
4087                 return (has_super + ext4_bg_num_gdb(sb, grp) +
4088                         (has_super ? le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0) +
4089                         sbi->s_itb_per_group + 2);
4090
4091         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
4092                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
4093         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
4094         for (i = 0; i < ngroups; i++) {
4095                 gdp = ext4_get_group_desc(sb, i, NULL);
4096                 b = ext4_block_bitmap(sb, gdp);
4097                 if (b >= first_block && b <= last_block) {
4098                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
4099                         count++;
4100                 }
4101                 b = ext4_inode_bitmap(sb, gdp);
4102                 if (b >= first_block && b <= last_block) {
4103                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
4104                         count++;
4105                 }
4106                 b = ext4_inode_table(sb, gdp);
4107                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
4108                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
4109                                 int c = EXT4_B2C(sbi, b - first_block);
4110                                 ext4_set_bit(c, buf);
4111                                 count++;
4112                         }
4113                 if (i != grp)
4114                         continue;
4115                 s = 0;
4116                 if (ext4_bg_has_super(sb, grp)) {
4117                         ext4_set_bit(s++, buf);
4118                         count++;
4119                 }
4120                 j = ext4_bg_num_gdb(sb, grp);
4121                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
4122                         ext4_error(sb, "Invalid number of block group "
4123                                    "descriptor blocks: %d", j);
4124                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
4125                 }
4126                 count += j;
4127                 for (; j > 0; j--)
4128                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
4129         }
4130         if (!count)
4131                 return 0;
4132         return EXT4_CLUSTERS_PER_GROUP(sb) -
4133                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
4134 }
4135
4136 /*
4137  * Compute the overhead and stash it in sbi->s_overhead
4138  */
4139 int ext4_calculate_overhead(struct super_block *sb)
4140 {
4141         struct ext4_sb_info *sbi = EXT4_SB(sb);
4142         struct ext4_super_block *es = sbi->s_es;
4143         struct inode *j_inode;
4144         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
4145         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4146         ext4_fsblk_t overhead = 0;
4147         char *buf = (char *) get_zeroed_page(GFP_NOFS);
4148
4149         if (!buf)
4150                 return -ENOMEM;
4151
4152         /*
4153          * Compute the overhead (FS structures).  This is constant
4154          * for a given filesystem unless the number of block groups
4155          * changes so we cache the previous value until it does.
4156          */
4157
4158         /*
4159          * All of the blocks before first_data_block are overhead
4160          */
4161         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
4162
4163         /*
4164          * Add the overhead found in each block group
4165          */
4166         for (i = 0; i < ngroups; i++) {
4167                 int blks;
4168
4169                 blks = count_overhead(sb, i, buf);
4170                 overhead += blks;
4171                 if (blks)
4172                         memset(buf, 0, PAGE_SIZE);
4173                 cond_resched();
4174         }
4175
4176         /*
4177          * Add the internal journal blocks whether the journal has been
4178          * loaded or not
4179          */
4180         if (sbi->s_journal && !sbi->s_journal_bdev)
4181                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_total_len);
4182         else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) {
4183                 /* j_inum for internal journal is non-zero */
4184                 j_inode = ext4_get_journal_inode(sb, j_inum);
4185                 if (j_inode) {
4186                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
4187                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
4188                         iput(j_inode);
4189                 } else {
4190                         ext4_msg(sb, KERN_ERR, "can't get journal size");
4191                 }
4192         }
4193         sbi->s_overhead = overhead;
4194         smp_wmb();
4195         free_page((unsigned long) buf);
4196         return 0;
4197 }
4198
4199 static void ext4_set_resv_clusters(struct super_block *sb)
4200 {
4201         ext4_fsblk_t resv_clusters;
4202         struct ext4_sb_info *sbi = EXT4_SB(sb);
4203
4204         /*
4205          * There's no need to reserve anything when we aren't using extents.
4206          * The space estimates are exact, there are no unwritten extents,
4207          * hole punching doesn't need new metadata... This is needed especially
4208          * to keep ext2/3 backward compatibility.
4209          */
4210         if (!ext4_has_feature_extents(sb))
4211                 return;
4212         /*
4213          * By default we reserve 2% or 4096 clusters, whichever is smaller.
4214          * This should cover the situations where we can not afford to run
4215          * out of space like for example punch hole, or converting
4216          * unwritten extents in delalloc path. In most cases such
4217          * allocation would require 1, or 2 blocks, higher numbers are
4218          * very rare.
4219          */
4220         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
4221                          sbi->s_cluster_bits);
4222
4223         do_div(resv_clusters, 50);
4224         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
4225
4226         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
4227 }
4228
4229 static const char *ext4_quota_mode(struct super_block *sb)
4230 {
4231 #ifdef CONFIG_QUOTA
4232         if (!ext4_quota_capable(sb))
4233                 return "none";
4234
4235         if (EXT4_SB(sb)->s_journal && ext4_is_quota_journalled(sb))
4236                 return "journalled";
4237         else
4238                 return "writeback";
4239 #else
4240         return "disabled";
4241 #endif
4242 }
4243
4244 static void ext4_setup_csum_trigger(struct super_block *sb,
4245                                     enum ext4_journal_trigger_type type,
4246                                     void (*trigger)(
4247                                         struct jbd2_buffer_trigger_type *type,
4248                                         struct buffer_head *bh,
4249                                         void *mapped_data,
4250                                         size_t size))
4251 {
4252         struct ext4_sb_info *sbi = EXT4_SB(sb);
4253
4254         sbi->s_journal_triggers[type].sb = sb;
4255         sbi->s_journal_triggers[type].tr_triggers.t_frozen = trigger;
4256 }
4257
4258 static void ext4_free_sbi(struct ext4_sb_info *sbi)
4259 {
4260         if (!sbi)
4261                 return;
4262
4263         kfree(sbi->s_blockgroup_lock);
4264         fs_put_dax(sbi->s_daxdev, NULL);
4265         kfree(sbi);
4266 }
4267
4268 static struct ext4_sb_info *ext4_alloc_sbi(struct super_block *sb)
4269 {
4270         struct ext4_sb_info *sbi;
4271
4272         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
4273         if (!sbi)
4274                 return NULL;
4275
4276         sbi->s_daxdev = fs_dax_get_by_bdev(sb->s_bdev, &sbi->s_dax_part_off,
4277                                            NULL, NULL);
4278
4279         sbi->s_blockgroup_lock =
4280                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
4281
4282         if (!sbi->s_blockgroup_lock)
4283                 goto err_out;
4284
4285         sb->s_fs_info = sbi;
4286         sbi->s_sb = sb;
4287         return sbi;
4288 err_out:
4289         fs_put_dax(sbi->s_daxdev, NULL);
4290         kfree(sbi);
4291         return NULL;
4292 }
4293
4294 static void ext4_set_def_opts(struct super_block *sb,
4295                               struct ext4_super_block *es)
4296 {
4297         unsigned long def_mount_opts;
4298
4299         /* Set defaults before we parse the mount options */
4300         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
4301         set_opt(sb, INIT_INODE_TABLE);
4302         if (def_mount_opts & EXT4_DEFM_DEBUG)
4303                 set_opt(sb, DEBUG);
4304         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
4305                 set_opt(sb, GRPID);
4306         if (def_mount_opts & EXT4_DEFM_UID16)
4307                 set_opt(sb, NO_UID32);
4308         /* xattr user namespace & acls are now defaulted on */
4309         set_opt(sb, XATTR_USER);
4310 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4311         set_opt(sb, POSIX_ACL);
4312 #endif
4313         if (ext4_has_feature_fast_commit(sb))
4314                 set_opt2(sb, JOURNAL_FAST_COMMIT);
4315         /* don't forget to enable journal_csum when metadata_csum is enabled. */
4316         if (ext4_has_metadata_csum(sb))
4317                 set_opt(sb, JOURNAL_CHECKSUM);
4318
4319         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
4320                 set_opt(sb, JOURNAL_DATA);
4321         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
4322                 set_opt(sb, ORDERED_DATA);
4323         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
4324                 set_opt(sb, WRITEBACK_DATA);
4325
4326         if (le16_to_cpu(es->s_errors) == EXT4_ERRORS_PANIC)
4327                 set_opt(sb, ERRORS_PANIC);
4328         else if (le16_to_cpu(es->s_errors) == EXT4_ERRORS_CONTINUE)
4329                 set_opt(sb, ERRORS_CONT);
4330         else
4331                 set_opt(sb, ERRORS_RO);
4332         /* block_validity enabled by default; disable with noblock_validity */
4333         set_opt(sb, BLOCK_VALIDITY);
4334         if (def_mount_opts & EXT4_DEFM_DISCARD)
4335                 set_opt(sb, DISCARD);
4336
4337         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
4338                 set_opt(sb, BARRIER);
4339
4340         /*
4341          * enable delayed allocation by default
4342          * Use -o nodelalloc to turn it off
4343          */
4344         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
4345             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
4346                 set_opt(sb, DELALLOC);
4347
4348         if (sb->s_blocksize == PAGE_SIZE)
4349                 set_opt(sb, DIOREAD_NOLOCK);
4350 }
4351
4352 static int ext4_handle_clustersize(struct super_block *sb)
4353 {
4354         struct ext4_sb_info *sbi = EXT4_SB(sb);
4355         struct ext4_super_block *es = sbi->s_es;
4356         int clustersize;
4357
4358         /* Handle clustersize */
4359         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
4360         if (ext4_has_feature_bigalloc(sb)) {
4361                 if (clustersize < sb->s_blocksize) {
4362                         ext4_msg(sb, KERN_ERR,
4363                                  "cluster size (%d) smaller than "
4364                                  "block size (%lu)", clustersize, sb->s_blocksize);
4365                         return -EINVAL;
4366                 }
4367                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
4368                         le32_to_cpu(es->s_log_block_size);
4369                 sbi->s_clusters_per_group =
4370                         le32_to_cpu(es->s_clusters_per_group);
4371                 if (sbi->s_clusters_per_group > sb->s_blocksize * 8) {
4372                         ext4_msg(sb, KERN_ERR,
4373                                  "#clusters per group too big: %lu",
4374                                  sbi->s_clusters_per_group);
4375                         return -EINVAL;
4376                 }
4377                 if (sbi->s_blocks_per_group !=
4378                     (sbi->s_clusters_per_group * (clustersize / sb->s_blocksize))) {
4379                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
4380                                  "clusters per group (%lu) inconsistent",
4381                                  sbi->s_blocks_per_group,
4382                                  sbi->s_clusters_per_group);
4383                         return -EINVAL;
4384                 }
4385         } else {
4386                 if (clustersize != sb->s_blocksize) {
4387                         ext4_msg(sb, KERN_ERR,
4388                                  "fragment/cluster size (%d) != "
4389                                  "block size (%lu)", clustersize, sb->s_blocksize);
4390                         return -EINVAL;
4391                 }
4392                 if (sbi->s_blocks_per_group > sb->s_blocksize * 8) {
4393                         ext4_msg(sb, KERN_ERR,
4394                                  "#blocks per group too big: %lu",
4395                                  sbi->s_blocks_per_group);
4396                         return -EINVAL;
4397                 }
4398                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
4399                 sbi->s_cluster_bits = 0;
4400         }
4401         sbi->s_cluster_ratio = clustersize / sb->s_blocksize;
4402
4403         /* Do we have standard group size of clustersize * 8 blocks ? */
4404         if (sbi->s_blocks_per_group == clustersize << 3)
4405                 set_opt2(sb, STD_GROUP_SIZE);
4406
4407         return 0;
4408 }
4409
4410 static void ext4_fast_commit_init(struct super_block *sb)
4411 {
4412         struct ext4_sb_info *sbi = EXT4_SB(sb);
4413
4414         /* Initialize fast commit stuff */
4415         atomic_set(&sbi->s_fc_subtid, 0);
4416         INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_MAIN]);
4417         INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_STAGING]);
4418         INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_MAIN]);
4419         INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_STAGING]);
4420         sbi->s_fc_bytes = 0;
4421         ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE);
4422         sbi->s_fc_ineligible_tid = 0;
4423         spin_lock_init(&sbi->s_fc_lock);
4424         memset(&sbi->s_fc_stats, 0, sizeof(sbi->s_fc_stats));
4425         sbi->s_fc_replay_state.fc_regions = NULL;
4426         sbi->s_fc_replay_state.fc_regions_size = 0;
4427         sbi->s_fc_replay_state.fc_regions_used = 0;
4428         sbi->s_fc_replay_state.fc_regions_valid = 0;
4429         sbi->s_fc_replay_state.fc_modified_inodes = NULL;
4430         sbi->s_fc_replay_state.fc_modified_inodes_size = 0;
4431         sbi->s_fc_replay_state.fc_modified_inodes_used = 0;
4432 }
4433
4434 static int ext4_inode_info_init(struct super_block *sb,
4435                                 struct ext4_super_block *es)
4436 {
4437         struct ext4_sb_info *sbi = EXT4_SB(sb);
4438
4439         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
4440                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
4441                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
4442         } else {
4443                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
4444                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
4445                 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) {
4446                         ext4_msg(sb, KERN_ERR, "invalid first ino: %u",
4447                                  sbi->s_first_ino);
4448                         return -EINVAL;
4449                 }
4450                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
4451                     (!is_power_of_2(sbi->s_inode_size)) ||
4452                     (sbi->s_inode_size > sb->s_blocksize)) {
4453                         ext4_msg(sb, KERN_ERR,
4454                                "unsupported inode size: %d",
4455                                sbi->s_inode_size);
4456                         ext4_msg(sb, KERN_ERR, "blocksize: %lu", sb->s_blocksize);
4457                         return -EINVAL;
4458                 }
4459                 /*
4460                  * i_atime_extra is the last extra field available for
4461                  * [acm]times in struct ext4_inode. Checking for that
4462                  * field should suffice to ensure we have extra space
4463                  * for all three.
4464                  */
4465                 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) +
4466                         sizeof(((struct ext4_inode *)0)->i_atime_extra)) {
4467                         sb->s_time_gran = 1;
4468                         sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX;
4469                 } else {
4470                         sb->s_time_gran = NSEC_PER_SEC;
4471                         sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX;
4472                 }
4473                 sb->s_time_min = EXT4_TIMESTAMP_MIN;
4474         }
4475
4476         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
4477                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4478                         EXT4_GOOD_OLD_INODE_SIZE;
4479                 if (ext4_has_feature_extra_isize(sb)) {
4480                         unsigned v, max = (sbi->s_inode_size -
4481                                            EXT4_GOOD_OLD_INODE_SIZE);
4482
4483                         v = le16_to_cpu(es->s_want_extra_isize);
4484                         if (v > max) {
4485                                 ext4_msg(sb, KERN_ERR,
4486                                          "bad s_want_extra_isize: %d", v);
4487                                 return -EINVAL;
4488                         }
4489                         if (sbi->s_want_extra_isize < v)
4490                                 sbi->s_want_extra_isize = v;
4491
4492                         v = le16_to_cpu(es->s_min_extra_isize);
4493                         if (v > max) {
4494                                 ext4_msg(sb, KERN_ERR,
4495                                          "bad s_min_extra_isize: %d", v);
4496                                 return -EINVAL;
4497                         }
4498                         if (sbi->s_want_extra_isize < v)
4499                                 sbi->s_want_extra_isize = v;
4500                 }
4501         }
4502
4503         return 0;
4504 }
4505
4506 #if IS_ENABLED(CONFIG_UNICODE)
4507 static int ext4_encoding_init(struct super_block *sb, struct ext4_super_block *es)
4508 {
4509         const struct ext4_sb_encodings *encoding_info;
4510         struct unicode_map *encoding;
4511         __u16 encoding_flags = le16_to_cpu(es->s_encoding_flags);
4512
4513         if (!ext4_has_feature_casefold(sb) || sb->s_encoding)
4514                 return 0;
4515
4516         encoding_info = ext4_sb_read_encoding(es);
4517         if (!encoding_info) {
4518                 ext4_msg(sb, KERN_ERR,
4519                         "Encoding requested by superblock is unknown");
4520                 return -EINVAL;
4521         }
4522
4523         encoding = utf8_load(encoding_info->version);
4524         if (IS_ERR(encoding)) {
4525                 ext4_msg(sb, KERN_ERR,
4526                         "can't mount with superblock charset: %s-%u.%u.%u "
4527                         "not supported by the kernel. flags: 0x%x.",
4528                         encoding_info->name,
4529                         unicode_major(encoding_info->version),
4530                         unicode_minor(encoding_info->version),
4531                         unicode_rev(encoding_info->version),
4532                         encoding_flags);
4533                 return -EINVAL;
4534         }
4535         ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: "
4536                 "%s-%u.%u.%u with flags 0x%hx", encoding_info->name,
4537                 unicode_major(encoding_info->version),
4538                 unicode_minor(encoding_info->version),
4539                 unicode_rev(encoding_info->version),
4540                 encoding_flags);
4541
4542         sb->s_encoding = encoding;
4543         sb->s_encoding_flags = encoding_flags;
4544
4545         return 0;
4546 }
4547 #else
4548 static inline int ext4_encoding_init(struct super_block *sb, struct ext4_super_block *es)
4549 {
4550         return 0;
4551 }
4552 #endif
4553
4554 static int ext4_init_metadata_csum(struct super_block *sb, struct ext4_super_block *es)
4555 {
4556         struct ext4_sb_info *sbi = EXT4_SB(sb);
4557
4558         /* Warn if metadata_csum and gdt_csum are both set. */
4559         if (ext4_has_feature_metadata_csum(sb) &&
4560             ext4_has_feature_gdt_csum(sb))
4561                 ext4_warning(sb, "metadata_csum and uninit_bg are "
4562                              "redundant flags; please run fsck.");
4563
4564         /* Check for a known checksum algorithm */
4565         if (!ext4_verify_csum_type(sb, es)) {
4566                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
4567                          "unknown checksum algorithm.");
4568                 return -EINVAL;
4569         }
4570         ext4_setup_csum_trigger(sb, EXT4_JTR_ORPHAN_FILE,
4571                                 ext4_orphan_file_block_trigger);
4572
4573         /* Load the checksum driver */
4574         sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
4575         if (IS_ERR(sbi->s_chksum_driver)) {
4576                 int ret = PTR_ERR(sbi->s_chksum_driver);
4577                 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
4578                 sbi->s_chksum_driver = NULL;
4579                 return ret;
4580         }
4581
4582         /* Check superblock checksum */
4583         if (!ext4_superblock_csum_verify(sb, es)) {
4584                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
4585                          "invalid superblock checksum.  Run e2fsck?");
4586                 return -EFSBADCRC;
4587         }
4588
4589         /* Precompute checksum seed for all metadata */
4590         if (ext4_has_feature_csum_seed(sb))
4591                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
4592         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
4593                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
4594                                                sizeof(es->s_uuid));
4595         return 0;
4596 }
4597
4598 static int ext4_check_feature_compatibility(struct super_block *sb,
4599                                             struct ext4_super_block *es,
4600                                             int silent)
4601 {
4602         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
4603             (ext4_has_compat_features(sb) ||
4604              ext4_has_ro_compat_features(sb) ||
4605              ext4_has_incompat_features(sb)))
4606                 ext4_msg(sb, KERN_WARNING,
4607                        "feature flags set on rev 0 fs, "
4608                        "running e2fsck is recommended");
4609
4610         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
4611                 set_opt2(sb, HURD_COMPAT);
4612                 if (ext4_has_feature_64bit(sb)) {
4613                         ext4_msg(sb, KERN_ERR,
4614                                  "The Hurd can't support 64-bit file systems");
4615                         return -EINVAL;
4616                 }
4617
4618                 /*
4619                  * ea_inode feature uses l_i_version field which is not
4620                  * available in HURD_COMPAT mode.
4621                  */
4622                 if (ext4_has_feature_ea_inode(sb)) {
4623                         ext4_msg(sb, KERN_ERR,
4624                                  "ea_inode feature is not supported for Hurd");
4625                         return -EINVAL;
4626                 }
4627         }
4628
4629         if (IS_EXT2_SB(sb)) {
4630                 if (ext2_feature_set_ok(sb))
4631                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
4632                                  "using the ext4 subsystem");
4633                 else {
4634                         /*
4635                          * If we're probing be silent, if this looks like
4636                          * it's actually an ext[34] filesystem.
4637                          */
4638                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4639                                 return -EINVAL;
4640                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
4641                                  "to feature incompatibilities");
4642                         return -EINVAL;
4643                 }
4644         }
4645
4646         if (IS_EXT3_SB(sb)) {
4647                 if (ext3_feature_set_ok(sb))
4648                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
4649                                  "using the ext4 subsystem");
4650                 else {
4651                         /*
4652                          * If we're probing be silent, if this looks like
4653                          * it's actually an ext4 filesystem.
4654                          */
4655                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4656                                 return -EINVAL;
4657                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
4658                                  "to feature incompatibilities");
4659                         return -EINVAL;
4660                 }
4661         }
4662
4663         /*
4664          * Check feature flags regardless of the revision level, since we
4665          * previously didn't change the revision level when setting the flags,
4666          * so there is a chance incompat flags are set on a rev 0 filesystem.
4667          */
4668         if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
4669                 return -EINVAL;
4670
4671         return 0;
4672 }
4673
4674 static int ext4_geometry_check(struct super_block *sb,
4675                                struct ext4_super_block *es)
4676 {
4677         struct ext4_sb_info *sbi = EXT4_SB(sb);
4678         __u64 blocks_count;
4679
4680         /* check blocks count against device size */
4681         blocks_count = sb_bdev_nr_blocks(sb);
4682         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
4683                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
4684                        "exceeds size of device (%llu blocks)",
4685                        ext4_blocks_count(es), blocks_count);
4686                 return -EINVAL;
4687         }
4688
4689         /*
4690          * It makes no sense for the first data block to be beyond the end
4691          * of the filesystem.
4692          */
4693         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
4694                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4695                          "block %u is beyond end of filesystem (%llu)",
4696                          le32_to_cpu(es->s_first_data_block),
4697                          ext4_blocks_count(es));
4698                 return -EINVAL;
4699         }
4700         if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) &&
4701             (sbi->s_cluster_ratio == 1)) {
4702                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4703                          "block is 0 with a 1k block and cluster size");
4704                 return -EINVAL;
4705         }
4706
4707         blocks_count = (ext4_blocks_count(es) -
4708                         le32_to_cpu(es->s_first_data_block) +
4709                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
4710         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
4711         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
4712                 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu "
4713                        "(block count %llu, first data block %u, "
4714                        "blocks per group %lu)", blocks_count,
4715                        ext4_blocks_count(es),
4716                        le32_to_cpu(es->s_first_data_block),
4717                        EXT4_BLOCKS_PER_GROUP(sb));
4718                 return -EINVAL;
4719         }
4720         sbi->s_groups_count = blocks_count;
4721         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
4722                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
4723         if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) !=
4724             le32_to_cpu(es->s_inodes_count)) {
4725                 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu",
4726                          le32_to_cpu(es->s_inodes_count),
4727                          ((u64)sbi->s_groups_count * sbi->s_inodes_per_group));
4728                 return -EINVAL;
4729         }
4730
4731         return 0;
4732 }
4733
4734 static void ext4_group_desc_free(struct ext4_sb_info *sbi)
4735 {
4736         struct buffer_head **group_desc;
4737         int i;
4738
4739         rcu_read_lock();
4740         group_desc = rcu_dereference(sbi->s_group_desc);
4741         for (i = 0; i < sbi->s_gdb_count; i++)
4742                 brelse(group_desc[i]);
4743         kvfree(group_desc);
4744         rcu_read_unlock();
4745 }
4746
4747 static int ext4_group_desc_init(struct super_block *sb,
4748                                 struct ext4_super_block *es,
4749                                 ext4_fsblk_t logical_sb_block,
4750                                 ext4_group_t *first_not_zeroed)
4751 {
4752         struct ext4_sb_info *sbi = EXT4_SB(sb);
4753         unsigned int db_count;
4754         ext4_fsblk_t block;
4755         int ret;
4756         int i;
4757
4758         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
4759                    EXT4_DESC_PER_BLOCK(sb);
4760         if (ext4_has_feature_meta_bg(sb)) {
4761                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
4762                         ext4_msg(sb, KERN_WARNING,
4763                                  "first meta block group too large: %u "
4764                                  "(group descriptor block count %u)",
4765                                  le32_to_cpu(es->s_first_meta_bg), db_count);
4766                         return -EINVAL;
4767                 }
4768         }
4769         rcu_assign_pointer(sbi->s_group_desc,
4770                            kvmalloc_array(db_count,
4771                                           sizeof(struct buffer_head *),
4772                                           GFP_KERNEL));
4773         if (sbi->s_group_desc == NULL) {
4774                 ext4_msg(sb, KERN_ERR, "not enough memory");
4775                 return -ENOMEM;
4776         }
4777
4778         bgl_lock_init(sbi->s_blockgroup_lock);
4779
4780         /* Pre-read the descriptors into the buffer cache */
4781         for (i = 0; i < db_count; i++) {
4782                 block = descriptor_loc(sb, logical_sb_block, i);
4783                 ext4_sb_breadahead_unmovable(sb, block);
4784         }
4785
4786         for (i = 0; i < db_count; i++) {
4787                 struct buffer_head *bh;
4788
4789                 block = descriptor_loc(sb, logical_sb_block, i);
4790                 bh = ext4_sb_bread_unmovable(sb, block);
4791                 if (IS_ERR(bh)) {
4792                         ext4_msg(sb, KERN_ERR,
4793                                "can't read group descriptor %d", i);
4794                         sbi->s_gdb_count = i;
4795                         ret = PTR_ERR(bh);
4796                         goto out;
4797                 }
4798                 rcu_read_lock();
4799                 rcu_dereference(sbi->s_group_desc)[i] = bh;
4800                 rcu_read_unlock();
4801         }
4802         sbi->s_gdb_count = db_count;
4803         if (!ext4_check_descriptors(sb, logical_sb_block, first_not_zeroed)) {
4804                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
4805                 ret = -EFSCORRUPTED;
4806                 goto out;
4807         }
4808         return 0;
4809 out:
4810         ext4_group_desc_free(sbi);
4811         return ret;
4812 }
4813
4814 static int ext4_load_and_init_journal(struct super_block *sb,
4815                                       struct ext4_super_block *es,
4816                                       struct ext4_fs_context *ctx)
4817 {
4818         struct ext4_sb_info *sbi = EXT4_SB(sb);
4819         int err;
4820
4821         err = ext4_load_journal(sb, es, ctx->journal_devnum);
4822         if (err)
4823                 return err;
4824
4825         if (ext4_has_feature_64bit(sb) &&
4826             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4827                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4828                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4829                 goto out;
4830         }
4831
4832         if (!set_journal_csum_feature_set(sb)) {
4833                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4834                          "feature set");
4835                 goto out;
4836         }
4837
4838         if (test_opt2(sb, JOURNAL_FAST_COMMIT) &&
4839                 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4840                                           JBD2_FEATURE_INCOMPAT_FAST_COMMIT)) {
4841                 ext4_msg(sb, KERN_ERR,
4842                         "Failed to set fast commit journal feature");
4843                 goto out;
4844         }
4845
4846         /* We have now updated the journal if required, so we can
4847          * validate the data journaling mode. */
4848         switch (test_opt(sb, DATA_FLAGS)) {
4849         case 0:
4850                 /* No mode set, assume a default based on the journal
4851                  * capabilities: ORDERED_DATA if the journal can
4852                  * cope, else JOURNAL_DATA
4853                  */
4854                 if (jbd2_journal_check_available_features
4855                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4856                         set_opt(sb, ORDERED_DATA);
4857                         sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
4858                 } else {
4859                         set_opt(sb, JOURNAL_DATA);
4860                         sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
4861                 }
4862                 break;
4863
4864         case EXT4_MOUNT_ORDERED_DATA:
4865         case EXT4_MOUNT_WRITEBACK_DATA:
4866                 if (!jbd2_journal_check_available_features
4867                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4868                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4869                                "requested data journaling mode");
4870                         goto out;
4871                 }
4872                 break;
4873         default:
4874                 break;
4875         }
4876
4877         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4878             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4879                 ext4_msg(sb, KERN_ERR, "can't mount with "
4880                         "journal_async_commit in data=ordered mode");
4881                 goto out;
4882         }
4883
4884         set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio);
4885
4886         sbi->s_journal->j_submit_inode_data_buffers =
4887                 ext4_journal_submit_inode_data_buffers;
4888         sbi->s_journal->j_finish_inode_data_buffers =
4889                 ext4_journal_finish_inode_data_buffers;
4890
4891         return 0;
4892
4893 out:
4894         /* flush s_error_work before journal destroy. */
4895         flush_work(&sbi->s_error_work);
4896         jbd2_journal_destroy(sbi->s_journal);
4897         sbi->s_journal = NULL;
4898         return -EINVAL;
4899 }
4900
4901 static int ext4_journal_data_mode_check(struct super_block *sb)
4902 {
4903         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4904                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with "
4905                             "data=journal disables delayed allocation, "
4906                             "dioread_nolock, O_DIRECT and fast_commit support!\n");
4907                 /* can't mount with both data=journal and dioread_nolock. */
4908                 clear_opt(sb, DIOREAD_NOLOCK);
4909                 clear_opt2(sb, JOURNAL_FAST_COMMIT);
4910                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4911                         ext4_msg(sb, KERN_ERR, "can't mount with "
4912                                  "both data=journal and delalloc");
4913                         return -EINVAL;
4914                 }
4915                 if (test_opt(sb, DAX_ALWAYS)) {
4916                         ext4_msg(sb, KERN_ERR, "can't mount with "
4917                                  "both data=journal and dax");
4918                         return -EINVAL;
4919                 }
4920                 if (ext4_has_feature_encrypt(sb)) {
4921                         ext4_msg(sb, KERN_WARNING,
4922                                  "encrypted files will use data=ordered "
4923                                  "instead of data journaling mode");
4924                 }
4925                 if (test_opt(sb, DELALLOC))
4926                         clear_opt(sb, DELALLOC);
4927         } else {
4928                 sb->s_iflags |= SB_I_CGROUPWB;
4929         }
4930
4931         return 0;
4932 }
4933
4934 static int ext4_load_super(struct super_block *sb, ext4_fsblk_t *lsb,
4935                            int silent)
4936 {
4937         struct ext4_sb_info *sbi = EXT4_SB(sb);
4938         struct ext4_super_block *es;
4939         ext4_fsblk_t logical_sb_block;
4940         unsigned long offset = 0;
4941         struct buffer_head *bh;
4942         int ret = -EINVAL;
4943         int blocksize;
4944
4945         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
4946         if (!blocksize) {
4947                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
4948                 return -EINVAL;
4949         }
4950
4951         /*
4952          * The ext4 superblock will not be buffer aligned for other than 1kB
4953          * block sizes.  We need to calculate the offset from buffer start.
4954          */
4955         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
4956                 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE;
4957                 offset = do_div(logical_sb_block, blocksize);
4958         } else {
4959                 logical_sb_block = sbi->s_sb_block;
4960         }
4961
4962         bh = ext4_sb_bread_unmovable(sb, logical_sb_block);
4963         if (IS_ERR(bh)) {
4964                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
4965                 return PTR_ERR(bh);
4966         }
4967         /*
4968          * Note: s_es must be initialized as soon as possible because
4969          *       some ext4 macro-instructions depend on its value
4970          */
4971         es = (struct ext4_super_block *) (bh->b_data + offset);
4972         sbi->s_es = es;
4973         sb->s_magic = le16_to_cpu(es->s_magic);
4974         if (sb->s_magic != EXT4_SUPER_MAGIC) {
4975                 if (!silent)
4976                         ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4977                 goto out;
4978         }
4979
4980         if (le32_to_cpu(es->s_log_block_size) >
4981             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4982                 ext4_msg(sb, KERN_ERR,
4983                          "Invalid log block size: %u",
4984                          le32_to_cpu(es->s_log_block_size));
4985                 goto out;
4986         }
4987         if (le32_to_cpu(es->s_log_cluster_size) >
4988             (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4989                 ext4_msg(sb, KERN_ERR,
4990                          "Invalid log cluster size: %u",
4991                          le32_to_cpu(es->s_log_cluster_size));
4992                 goto out;
4993         }
4994
4995         blocksize = EXT4_MIN_BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
4996
4997         /*
4998          * If the default block size is not the same as the real block size,
4999          * we need to reload it.
5000          */
5001         if (sb->s_blocksize == blocksize) {
5002                 *lsb = logical_sb_block;
5003                 sbi->s_sbh = bh;
5004                 return 0;
5005         }
5006
5007         /*
5008          * bh must be released before kill_bdev(), otherwise
5009          * it won't be freed and its page also. kill_bdev()
5010          * is called by sb_set_blocksize().
5011          */
5012         brelse(bh);
5013         /* Validate the filesystem blocksize */
5014         if (!sb_set_blocksize(sb, blocksize)) {
5015                 ext4_msg(sb, KERN_ERR, "bad block size %d",
5016                                 blocksize);
5017                 bh = NULL;
5018                 goto out;
5019         }
5020
5021         logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE;
5022         offset = do_div(logical_sb_block, blocksize);
5023         bh = ext4_sb_bread_unmovable(sb, logical_sb_block);
5024         if (IS_ERR(bh)) {
5025                 ext4_msg(sb, KERN_ERR, "Can't read superblock on 2nd try");
5026                 ret = PTR_ERR(bh);
5027                 bh = NULL;
5028                 goto out;
5029         }
5030         es = (struct ext4_super_block *)(bh->b_data + offset);
5031         sbi->s_es = es;
5032         if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
5033                 ext4_msg(sb, KERN_ERR, "Magic mismatch, very weird!");
5034                 goto out;
5035         }
5036         *lsb = logical_sb_block;
5037         sbi->s_sbh = bh;
5038         return 0;
5039 out:
5040         brelse(bh);
5041         return ret;
5042 }
5043
5044 static int __ext4_fill_super(struct fs_context *fc, struct super_block *sb)
5045 {
5046         struct ext4_super_block *es = NULL;
5047         struct ext4_sb_info *sbi = EXT4_SB(sb);
5048         struct flex_groups **flex_groups;
5049         ext4_fsblk_t block;
5050         ext4_fsblk_t logical_sb_block;
5051         struct inode *root;
5052         int ret = -ENOMEM;
5053         unsigned int i;
5054         int needs_recovery, has_huge_files;
5055         int err = 0;
5056         ext4_group_t first_not_zeroed;
5057         struct ext4_fs_context *ctx = fc->fs_private;
5058         int silent = fc->sb_flags & SB_SILENT;
5059
5060         /* Set defaults for the variables that will be set during parsing */
5061         if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO))
5062                 ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
5063
5064         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
5065         sbi->s_sectors_written_start =
5066                 part_stat_read(sb->s_bdev, sectors[STAT_WRITE]);
5067
5068         /* -EINVAL is default */
5069         ret = -EINVAL;
5070         err = ext4_load_super(sb, &logical_sb_block, silent);
5071         if (err)
5072                 goto out_fail;
5073
5074         es = sbi->s_es;
5075         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
5076
5077         err = ext4_init_metadata_csum(sb, es);
5078         if (err)
5079                 goto failed_mount;
5080
5081         ext4_set_def_opts(sb, es);
5082
5083         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
5084         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
5085         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
5086         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
5087         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
5088
5089         /*
5090          * set default s_li_wait_mult for lazyinit, for the case there is
5091          * no mount option specified.
5092          */
5093         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
5094
5095         if (ext4_inode_info_init(sb, es))
5096                 goto failed_mount;
5097
5098         err = parse_apply_sb_mount_options(sb, ctx);
5099         if (err < 0)
5100                 goto failed_mount;
5101
5102         sbi->s_def_mount_opt = sbi->s_mount_opt;
5103         sbi->s_def_mount_opt2 = sbi->s_mount_opt2;
5104
5105         err = ext4_check_opt_consistency(fc, sb);
5106         if (err < 0)
5107                 goto failed_mount;
5108
5109         ext4_apply_options(fc, sb);
5110
5111         if (ext4_encoding_init(sb, es))
5112                 goto failed_mount;
5113
5114         if (ext4_journal_data_mode_check(sb))
5115                 goto failed_mount;
5116
5117         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
5118                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
5119
5120         /* i_version is always enabled now */
5121         sb->s_flags |= SB_I_VERSION;
5122
5123         if (ext4_check_feature_compatibility(sb, es, silent))
5124                 goto failed_mount;
5125
5126         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (sb->s_blocksize / 4)) {
5127                 ext4_msg(sb, KERN_ERR,
5128                          "Number of reserved GDT blocks insanely large: %d",
5129                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
5130                 goto failed_mount;
5131         }
5132
5133         if (sbi->s_daxdev) {
5134                 if (sb->s_blocksize == PAGE_SIZE)
5135                         set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags);
5136                 else
5137                         ext4_msg(sb, KERN_ERR, "unsupported blocksize for DAX\n");
5138         }
5139
5140         if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) {
5141                 if (ext4_has_feature_inline_data(sb)) {
5142                         ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
5143                                         " that may contain inline data");
5144                         goto failed_mount;
5145                 }
5146                 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) {
5147                         ext4_msg(sb, KERN_ERR,
5148                                 "DAX unsupported by block device.");
5149                         goto failed_mount;
5150                 }
5151         }
5152
5153         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
5154                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
5155                          es->s_encryption_level);
5156                 goto failed_mount;
5157         }
5158
5159         has_huge_files = ext4_has_feature_huge_file(sb);
5160         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
5161                                                       has_huge_files);
5162         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
5163
5164         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
5165         if (ext4_has_feature_64bit(sb)) {
5166                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
5167                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
5168                     !is_power_of_2(sbi->s_desc_size)) {
5169                         ext4_msg(sb, KERN_ERR,
5170                                "unsupported descriptor size %lu",
5171                                sbi->s_desc_size);
5172                         goto failed_mount;
5173                 }
5174         } else
5175                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
5176
5177         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
5178         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
5179
5180         sbi->s_inodes_per_block = sb->s_blocksize / EXT4_INODE_SIZE(sb);
5181         if (sbi->s_inodes_per_block == 0 || sbi->s_blocks_per_group == 0) {
5182                 if (!silent)
5183                         ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
5184                 goto failed_mount;
5185         }
5186         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
5187             sbi->s_inodes_per_group > sb->s_blocksize * 8) {
5188                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
5189                          sbi->s_inodes_per_group);
5190                 goto failed_mount;
5191         }
5192         sbi->s_itb_per_group = sbi->s_inodes_per_group /
5193                                         sbi->s_inodes_per_block;
5194         sbi->s_desc_per_block = sb->s_blocksize / EXT4_DESC_SIZE(sb);
5195         sbi->s_mount_state = le16_to_cpu(es->s_state) & ~EXT4_FC_REPLAY;
5196         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
5197         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
5198
5199         for (i = 0; i < 4; i++)
5200                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
5201         sbi->s_def_hash_version = es->s_def_hash_version;
5202         if (ext4_has_feature_dir_index(sb)) {
5203                 i = le32_to_cpu(es->s_flags);
5204                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
5205                         sbi->s_hash_unsigned = 3;
5206                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
5207 #ifdef __CHAR_UNSIGNED__
5208                         if (!sb_rdonly(sb))
5209                                 es->s_flags |=
5210                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
5211                         sbi->s_hash_unsigned = 3;
5212 #else
5213                         if (!sb_rdonly(sb))
5214                                 es->s_flags |=
5215                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
5216 #endif
5217                 }
5218         }
5219
5220         if (ext4_handle_clustersize(sb))
5221                 goto failed_mount;
5222
5223         /*
5224          * Test whether we have more sectors than will fit in sector_t,
5225          * and whether the max offset is addressable by the page cache.
5226          */
5227         err = generic_check_addressable(sb->s_blocksize_bits,
5228                                         ext4_blocks_count(es));
5229         if (err) {
5230                 ext4_msg(sb, KERN_ERR, "filesystem"
5231                          " too large to mount safely on this system");
5232                 goto failed_mount;
5233         }
5234
5235         if (ext4_geometry_check(sb, es))
5236                 goto failed_mount;
5237
5238         err = ext4_group_desc_init(sb, es, logical_sb_block, &first_not_zeroed);
5239         if (err)
5240                 goto failed_mount;
5241
5242         timer_setup(&sbi->s_err_report, print_daily_error_info, 0);
5243         spin_lock_init(&sbi->s_error_lock);
5244         INIT_WORK(&sbi->s_error_work, flush_stashed_error_work);
5245
5246         /* Register extent status tree shrinker */
5247         if (ext4_es_register_shrinker(sbi))
5248                 goto failed_mount3;
5249
5250         sbi->s_stripe = ext4_get_stripe_size(sbi);
5251         sbi->s_extent_max_zeroout_kb = 32;
5252
5253         /*
5254          * set up enough so that it can read an inode
5255          */
5256         sb->s_op = &ext4_sops;
5257         sb->s_export_op = &ext4_export_ops;
5258         sb->s_xattr = ext4_xattr_handlers;
5259 #ifdef CONFIG_FS_ENCRYPTION
5260         sb->s_cop = &ext4_cryptops;
5261 #endif
5262 #ifdef CONFIG_FS_VERITY
5263         sb->s_vop = &ext4_verityops;
5264 #endif
5265 #ifdef CONFIG_QUOTA
5266         sb->dq_op = &ext4_quota_operations;
5267         if (ext4_has_feature_quota(sb))
5268                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
5269         else
5270                 sb->s_qcop = &ext4_qctl_operations;
5271         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
5272 #endif
5273         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
5274
5275         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
5276         mutex_init(&sbi->s_orphan_lock);
5277
5278         ext4_fast_commit_init(sb);
5279
5280         sb->s_root = NULL;
5281
5282         needs_recovery = (es->s_last_orphan != 0 ||
5283                           ext4_has_feature_orphan_present(sb) ||
5284                           ext4_has_feature_journal_needs_recovery(sb));
5285
5286         if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb)) {
5287                 err = ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block));
5288                 if (err)
5289                         goto failed_mount3a;
5290         }
5291
5292         /*
5293          * The first inode we look at is the journal inode.  Don't try
5294          * root first: it may be modified in the journal!
5295          */
5296         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
5297                 err = ext4_load_and_init_journal(sb, es, ctx);
5298                 if (err)
5299                         goto failed_mount3a;
5300         } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
5301                    ext4_has_feature_journal_needs_recovery(sb)) {
5302                 ext4_msg(sb, KERN_ERR, "required journal recovery "
5303                        "suppressed and not mounted read-only");
5304                 goto failed_mount3a;
5305         } else {
5306                 /* Nojournal mode, all journal mount options are illegal */
5307                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5308                         ext4_msg(sb, KERN_ERR, "can't mount with "
5309                                  "journal_async_commit, fs mounted w/o journal");
5310                         goto failed_mount3a;
5311                 }
5312
5313                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
5314                         ext4_msg(sb, KERN_ERR, "can't mount with "
5315                                  "journal_checksum, fs mounted w/o journal");
5316                         goto failed_mount3a;
5317                 }
5318                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
5319                         ext4_msg(sb, KERN_ERR, "can't mount with "
5320                                  "commit=%lu, fs mounted w/o journal",
5321                                  sbi->s_commit_interval / HZ);
5322                         goto failed_mount3a;
5323                 }
5324                 if (EXT4_MOUNT_DATA_FLAGS &
5325                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
5326                         ext4_msg(sb, KERN_ERR, "can't mount with "
5327                                  "data=, fs mounted w/o journal");
5328                         goto failed_mount3a;
5329                 }
5330                 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM;
5331                 clear_opt(sb, JOURNAL_CHECKSUM);
5332                 clear_opt(sb, DATA_FLAGS);
5333                 clear_opt2(sb, JOURNAL_FAST_COMMIT);
5334                 sbi->s_journal = NULL;
5335                 needs_recovery = 0;
5336         }
5337
5338         if (!test_opt(sb, NO_MBCACHE)) {
5339                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
5340                 if (!sbi->s_ea_block_cache) {
5341                         ext4_msg(sb, KERN_ERR,
5342                                  "Failed to create ea_block_cache");
5343                         goto failed_mount_wq;
5344                 }
5345
5346                 if (ext4_has_feature_ea_inode(sb)) {
5347                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
5348                         if (!sbi->s_ea_inode_cache) {
5349                                 ext4_msg(sb, KERN_ERR,
5350                                          "Failed to create ea_inode_cache");
5351                                 goto failed_mount_wq;
5352                         }
5353                 }
5354         }
5355
5356         if (ext4_has_feature_verity(sb) && sb->s_blocksize != PAGE_SIZE) {
5357                 ext4_msg(sb, KERN_ERR, "Unsupported blocksize for fs-verity");
5358                 goto failed_mount_wq;
5359         }
5360
5361         /*
5362          * Get the # of file system overhead blocks from the
5363          * superblock if present.
5364          */
5365         sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
5366         /* ignore the precalculated value if it is ridiculous */
5367         if (sbi->s_overhead > ext4_blocks_count(es))
5368                 sbi->s_overhead = 0;
5369         /*
5370          * If the bigalloc feature is not enabled recalculating the
5371          * overhead doesn't take long, so we might as well just redo
5372          * it to make sure we are using the correct value.
5373          */
5374         if (!ext4_has_feature_bigalloc(sb))
5375                 sbi->s_overhead = 0;
5376         if (sbi->s_overhead == 0) {
5377                 err = ext4_calculate_overhead(sb);
5378                 if (err)
5379                         goto failed_mount_wq;
5380         }
5381
5382         /*
5383          * The maximum number of concurrent works can be high and
5384          * concurrency isn't really necessary.  Limit it to 1.
5385          */
5386         EXT4_SB(sb)->rsv_conversion_wq =
5387                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
5388         if (!EXT4_SB(sb)->rsv_conversion_wq) {
5389                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
5390                 ret = -ENOMEM;
5391                 goto failed_mount4;
5392         }
5393
5394         /*
5395          * The jbd2_journal_load will have done any necessary log recovery,
5396          * so we can safely mount the rest of the filesystem now.
5397          */
5398
5399         root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL);
5400         if (IS_ERR(root)) {
5401                 ext4_msg(sb, KERN_ERR, "get root inode failed");
5402                 ret = PTR_ERR(root);
5403                 root = NULL;
5404                 goto failed_mount4;
5405         }
5406         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
5407                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
5408                 iput(root);
5409                 goto failed_mount4;
5410         }
5411
5412         sb->s_root = d_make_root(root);
5413         if (!sb->s_root) {
5414                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
5415                 ret = -ENOMEM;
5416                 goto failed_mount4;
5417         }
5418
5419         ret = ext4_setup_super(sb, es, sb_rdonly(sb));
5420         if (ret == -EROFS) {
5421                 sb->s_flags |= SB_RDONLY;
5422                 ret = 0;
5423         } else if (ret)
5424                 goto failed_mount4a;
5425
5426         ext4_set_resv_clusters(sb);
5427
5428         if (test_opt(sb, BLOCK_VALIDITY)) {
5429                 err = ext4_setup_system_zone(sb);
5430                 if (err) {
5431                         ext4_msg(sb, KERN_ERR, "failed to initialize system "
5432                                  "zone (%d)", err);
5433                         goto failed_mount4a;
5434                 }
5435         }
5436         ext4_fc_replay_cleanup(sb);
5437
5438         ext4_ext_init(sb);
5439
5440         /*
5441          * Enable optimize_scan if number of groups is > threshold. This can be
5442          * turned off by passing "mb_optimize_scan=0". This can also be
5443          * turned on forcefully by passing "mb_optimize_scan=1".
5444          */
5445         if (!(ctx->spec & EXT4_SPEC_mb_optimize_scan)) {
5446                 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD)
5447                         set_opt2(sb, MB_OPTIMIZE_SCAN);
5448                 else
5449                         clear_opt2(sb, MB_OPTIMIZE_SCAN);
5450         }
5451
5452         err = ext4_mb_init(sb);
5453         if (err) {
5454                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
5455                          err);
5456                 goto failed_mount5;
5457         }
5458
5459         /*
5460          * We can only set up the journal commit callback once
5461          * mballoc is initialized
5462          */
5463         if (sbi->s_journal)
5464                 sbi->s_journal->j_commit_callback =
5465                         ext4_journal_commit_callback;
5466
5467         block = ext4_count_free_clusters(sb);
5468         ext4_free_blocks_count_set(sbi->s_es,
5469                                    EXT4_C2B(sbi, block));
5470         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
5471                                   GFP_KERNEL);
5472         if (!err) {
5473                 unsigned long freei = ext4_count_free_inodes(sb);
5474                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
5475                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
5476                                           GFP_KERNEL);
5477         }
5478         if (!err)
5479                 err = percpu_counter_init(&sbi->s_dirs_counter,
5480                                           ext4_count_dirs(sb), GFP_KERNEL);
5481         if (!err)
5482                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
5483                                           GFP_KERNEL);
5484         if (!err)
5485                 err = percpu_counter_init(&sbi->s_sra_exceeded_retry_limit, 0,
5486                                           GFP_KERNEL);
5487         if (!err)
5488                 err = percpu_init_rwsem(&sbi->s_writepages_rwsem);
5489
5490         if (err) {
5491                 ext4_msg(sb, KERN_ERR, "insufficient memory");
5492                 goto failed_mount6;
5493         }
5494
5495         if (ext4_has_feature_flex_bg(sb))
5496                 if (!ext4_fill_flex_info(sb)) {
5497                         ext4_msg(sb, KERN_ERR,
5498                                "unable to initialize "
5499                                "flex_bg meta info!");
5500                         ret = -ENOMEM;
5501                         goto failed_mount6;
5502                 }
5503
5504         err = ext4_register_li_request(sb, first_not_zeroed);
5505         if (err)
5506                 goto failed_mount6;
5507
5508         err = ext4_register_sysfs(sb);
5509         if (err)
5510                 goto failed_mount7;
5511
5512         err = ext4_init_orphan_info(sb);
5513         if (err)
5514                 goto failed_mount8;
5515 #ifdef CONFIG_QUOTA
5516         /* Enable quota usage during mount. */
5517         if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
5518                 err = ext4_enable_quotas(sb);
5519                 if (err)
5520                         goto failed_mount9;
5521         }
5522 #endif  /* CONFIG_QUOTA */
5523
5524         /*
5525          * Save the original bdev mapping's wb_err value which could be
5526          * used to detect the metadata async write error.
5527          */
5528         spin_lock_init(&sbi->s_bdev_wb_lock);
5529         errseq_check_and_advance(&sb->s_bdev->bd_inode->i_mapping->wb_err,
5530                                  &sbi->s_bdev_wb_err);
5531         sb->s_bdev->bd_super = sb;
5532         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
5533         ext4_orphan_cleanup(sb, es);
5534         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
5535         /*
5536          * Update the checksum after updating free space/inode counters and
5537          * ext4_orphan_cleanup. Otherwise the superblock can have an incorrect
5538          * checksum in the buffer cache until it is written out and
5539          * e2fsprogs programs trying to open a file system immediately
5540          * after it is mounted can fail.
5541          */
5542         ext4_superblock_csum_set(sb);
5543         if (needs_recovery) {
5544                 ext4_msg(sb, KERN_INFO, "recovery complete");
5545                 err = ext4_mark_recovery_complete(sb, es);
5546                 if (err)
5547                         goto failed_mount10;
5548         }
5549
5550         if (test_opt(sb, DISCARD) && !bdev_max_discard_sectors(sb->s_bdev))
5551                 ext4_msg(sb, KERN_WARNING,
5552                          "mounting with \"discard\" option, but the device does not support discard");
5553
5554         if (es->s_error_count)
5555                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
5556
5557         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
5558         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
5559         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
5560         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
5561         atomic_set(&sbi->s_warning_count, 0);
5562         atomic_set(&sbi->s_msg_count, 0);
5563
5564         return 0;
5565
5566 failed_mount10:
5567         ext4_quota_off_umount(sb);
5568 failed_mount9: __maybe_unused
5569         ext4_release_orphan_info(sb);
5570 failed_mount8:
5571         ext4_unregister_sysfs(sb);
5572         kobject_put(&sbi->s_kobj);
5573 failed_mount7:
5574         ext4_unregister_li_request(sb);
5575 failed_mount6:
5576         ext4_mb_release(sb);
5577         rcu_read_lock();
5578         flex_groups = rcu_dereference(sbi->s_flex_groups);
5579         if (flex_groups) {
5580                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
5581                         kvfree(flex_groups[i]);
5582                 kvfree(flex_groups);
5583         }
5584         rcu_read_unlock();
5585         percpu_counter_destroy(&sbi->s_freeclusters_counter);
5586         percpu_counter_destroy(&sbi->s_freeinodes_counter);
5587         percpu_counter_destroy(&sbi->s_dirs_counter);
5588         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
5589         percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit);
5590         percpu_free_rwsem(&sbi->s_writepages_rwsem);
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         invalidate_bdev(sb->s_bdev);
5640         sb->s_fs_info = NULL;
5641         return err ? err : ret;
5642 }
5643
5644 static int ext4_fill_super(struct super_block *sb, struct fs_context *fc)
5645 {
5646         struct ext4_fs_context *ctx = fc->fs_private;
5647         struct ext4_sb_info *sbi;
5648         const char *descr;
5649         int ret;
5650
5651         sbi = ext4_alloc_sbi(sb);
5652         if (!sbi)
5653                 return -ENOMEM;
5654
5655         fc->s_fs_info = sbi;
5656
5657         /* Cleanup superblock name */
5658         strreplace(sb->s_id, '/', '!');
5659
5660         sbi->s_sb_block = 1;    /* Default super block location */
5661         if (ctx->spec & EXT4_SPEC_s_sb_block)
5662                 sbi->s_sb_block = ctx->s_sb_block;
5663
5664         ret = __ext4_fill_super(fc, sb);
5665         if (ret < 0)
5666                 goto free_sbi;
5667
5668         if (sbi->s_journal) {
5669                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
5670                         descr = " journalled data mode";
5671                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
5672                         descr = " ordered data mode";
5673                 else
5674                         descr = " writeback data mode";
5675         } else
5676                 descr = "out journal";
5677
5678         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
5679                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
5680                          "Quota mode: %s.", descr, 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 journal_t *ext4_get_journal(struct super_block *sb,
5756                                    unsigned int journal_inum)
5757 {
5758         struct inode *journal_inode;
5759         journal_t *journal;
5760
5761         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5762                 return NULL;
5763
5764         journal_inode = ext4_get_journal_inode(sb, journal_inum);
5765         if (!journal_inode)
5766                 return NULL;
5767
5768         journal = jbd2_journal_init_inode(journal_inode);
5769         if (!journal) {
5770                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
5771                 iput(journal_inode);
5772                 return NULL;
5773         }
5774         journal->j_private = sb;
5775         ext4_init_journal_params(sb, journal);
5776         return journal;
5777 }
5778
5779 static journal_t *ext4_get_dev_journal(struct super_block *sb,
5780                                        dev_t j_dev)
5781 {
5782         struct buffer_head *bh;
5783         journal_t *journal;
5784         ext4_fsblk_t start;
5785         ext4_fsblk_t len;
5786         int hblock, blocksize;
5787         ext4_fsblk_t sb_block;
5788         unsigned long offset;
5789         struct ext4_super_block *es;
5790         struct block_device *bdev;
5791
5792         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5793                 return NULL;
5794
5795         bdev = ext4_blkdev_get(j_dev, sb);
5796         if (bdev == NULL)
5797                 return NULL;
5798
5799         blocksize = sb->s_blocksize;
5800         hblock = bdev_logical_block_size(bdev);
5801         if (blocksize < hblock) {
5802                 ext4_msg(sb, KERN_ERR,
5803                         "blocksize too small for journal device");
5804                 goto out_bdev;
5805         }
5806
5807         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
5808         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
5809         set_blocksize(bdev, blocksize);
5810         if (!(bh = __bread(bdev, sb_block, blocksize))) {
5811                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
5812                        "external journal");
5813                 goto out_bdev;
5814         }
5815
5816         es = (struct ext4_super_block *) (bh->b_data + offset);
5817         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
5818             !(le32_to_cpu(es->s_feature_incompat) &
5819               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
5820                 ext4_msg(sb, KERN_ERR, "external journal has "
5821                                         "bad superblock");
5822                 brelse(bh);
5823                 goto out_bdev;
5824         }
5825
5826         if ((le32_to_cpu(es->s_feature_ro_compat) &
5827              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
5828             es->s_checksum != ext4_superblock_csum(sb, es)) {
5829                 ext4_msg(sb, KERN_ERR, "external journal has "
5830                                        "corrupt superblock");
5831                 brelse(bh);
5832                 goto out_bdev;
5833         }
5834
5835         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
5836                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
5837                 brelse(bh);
5838                 goto out_bdev;
5839         }
5840
5841         len = ext4_blocks_count(es);
5842         start = sb_block + 1;
5843         brelse(bh);     /* we're done with the superblock */
5844
5845         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
5846                                         start, len, blocksize);
5847         if (!journal) {
5848                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
5849                 goto out_bdev;
5850         }
5851         journal->j_private = sb;
5852         if (ext4_read_bh_lock(journal->j_sb_buffer, REQ_META | REQ_PRIO, true)) {
5853                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
5854                 goto out_journal;
5855         }
5856         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
5857                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
5858                                         "user (unsupported) - %d",
5859                         be32_to_cpu(journal->j_superblock->s_nr_users));
5860                 goto out_journal;
5861         }
5862         EXT4_SB(sb)->s_journal_bdev = bdev;
5863         ext4_init_journal_params(sb, journal);
5864         return journal;
5865
5866 out_journal:
5867         jbd2_journal_destroy(journal);
5868 out_bdev:
5869         ext4_blkdev_put(bdev);
5870         return NULL;
5871 }
5872
5873 static int ext4_load_journal(struct super_block *sb,
5874                              struct ext4_super_block *es,
5875                              unsigned long journal_devnum)
5876 {
5877         journal_t *journal;
5878         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
5879         dev_t journal_dev;
5880         int err = 0;
5881         int really_read_only;
5882         int journal_dev_ro;
5883
5884         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5885                 return -EFSCORRUPTED;
5886
5887         if (journal_devnum &&
5888             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5889                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
5890                         "numbers have changed");
5891                 journal_dev = new_decode_dev(journal_devnum);
5892         } else
5893                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
5894
5895         if (journal_inum && journal_dev) {
5896                 ext4_msg(sb, KERN_ERR,
5897                          "filesystem has both journal inode and journal device!");
5898                 return -EINVAL;
5899         }
5900
5901         if (journal_inum) {
5902                 journal = ext4_get_journal(sb, journal_inum);
5903                 if (!journal)
5904                         return -EINVAL;
5905         } else {
5906                 journal = ext4_get_dev_journal(sb, journal_dev);
5907                 if (!journal)
5908                         return -EINVAL;
5909         }
5910
5911         journal_dev_ro = bdev_read_only(journal->j_dev);
5912         really_read_only = bdev_read_only(sb->s_bdev) | journal_dev_ro;
5913
5914         if (journal_dev_ro && !sb_rdonly(sb)) {
5915                 ext4_msg(sb, KERN_ERR,
5916                          "journal device read-only, try mounting with '-o ro'");
5917                 err = -EROFS;
5918                 goto err_out;
5919         }
5920
5921         /*
5922          * Are we loading a blank journal or performing recovery after a
5923          * crash?  For recovery, we need to check in advance whether we
5924          * can get read-write access to the device.
5925          */
5926         if (ext4_has_feature_journal_needs_recovery(sb)) {
5927                 if (sb_rdonly(sb)) {
5928                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
5929                                         "required on readonly filesystem");
5930                         if (really_read_only) {
5931                                 ext4_msg(sb, KERN_ERR, "write access "
5932                                         "unavailable, cannot proceed "
5933                                         "(try mounting with noload)");
5934                                 err = -EROFS;
5935                                 goto err_out;
5936                         }
5937                         ext4_msg(sb, KERN_INFO, "write access will "
5938                                "be enabled during recovery");
5939                 }
5940         }
5941
5942         if (!(journal->j_flags & JBD2_BARRIER))
5943                 ext4_msg(sb, KERN_INFO, "barriers disabled");
5944
5945         if (!ext4_has_feature_journal_needs_recovery(sb))
5946                 err = jbd2_journal_wipe(journal, !really_read_only);
5947         if (!err) {
5948                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
5949                 if (save)
5950                         memcpy(save, ((char *) es) +
5951                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
5952                 err = jbd2_journal_load(journal);
5953                 if (save)
5954                         memcpy(((char *) es) + EXT4_S_ERR_START,
5955                                save, EXT4_S_ERR_LEN);
5956                 kfree(save);
5957         }
5958
5959         if (err) {
5960                 ext4_msg(sb, KERN_ERR, "error loading journal");
5961                 goto err_out;
5962         }
5963
5964         EXT4_SB(sb)->s_journal = journal;
5965         err = ext4_clear_journal_err(sb, es);
5966         if (err) {
5967                 EXT4_SB(sb)->s_journal = NULL;
5968                 jbd2_journal_destroy(journal);
5969                 return err;
5970         }
5971
5972         if (!really_read_only && journal_devnum &&
5973             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5974                 es->s_journal_dev = cpu_to_le32(journal_devnum);
5975                 ext4_commit_super(sb);
5976         }
5977         if (!really_read_only && journal_inum &&
5978             journal_inum != le32_to_cpu(es->s_journal_inum)) {
5979                 es->s_journal_inum = cpu_to_le32(journal_inum);
5980                 ext4_commit_super(sb);
5981         }
5982
5983         return 0;
5984
5985 err_out:
5986         jbd2_journal_destroy(journal);
5987         return err;
5988 }
5989
5990 /* Copy state of EXT4_SB(sb) into buffer for on-disk superblock */
5991 static void ext4_update_super(struct super_block *sb)
5992 {
5993         struct ext4_sb_info *sbi = EXT4_SB(sb);
5994         struct ext4_super_block *es = sbi->s_es;
5995         struct buffer_head *sbh = sbi->s_sbh;
5996
5997         lock_buffer(sbh);
5998         /*
5999          * If the file system is mounted read-only, don't update the
6000          * superblock write time.  This avoids updating the superblock
6001          * write time when we are mounting the root file system
6002          * read/only but we need to replay the journal; at that point,
6003          * for people who are east of GMT and who make their clock
6004          * tick in localtime for Windows bug-for-bug compatibility,
6005          * the clock is set in the future, and this will cause e2fsck
6006          * to complain and force a full file system check.
6007          */
6008         if (!(sb->s_flags & SB_RDONLY))
6009                 ext4_update_tstamp(es, s_wtime);
6010         es->s_kbytes_written =
6011                 cpu_to_le64(sbi->s_kbytes_written +
6012                     ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) -
6013                       sbi->s_sectors_written_start) >> 1));
6014         if (percpu_counter_initialized(&sbi->s_freeclusters_counter))
6015                 ext4_free_blocks_count_set(es,
6016                         EXT4_C2B(sbi, percpu_counter_sum_positive(
6017                                 &sbi->s_freeclusters_counter)));
6018         if (percpu_counter_initialized(&sbi->s_freeinodes_counter))
6019                 es->s_free_inodes_count =
6020                         cpu_to_le32(percpu_counter_sum_positive(
6021                                 &sbi->s_freeinodes_counter));
6022         /* Copy error information to the on-disk superblock */
6023         spin_lock(&sbi->s_error_lock);
6024         if (sbi->s_add_error_count > 0) {
6025                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
6026                 if (!es->s_first_error_time && !es->s_first_error_time_hi) {
6027                         __ext4_update_tstamp(&es->s_first_error_time,
6028                                              &es->s_first_error_time_hi,
6029                                              sbi->s_first_error_time);
6030                         strncpy(es->s_first_error_func, sbi->s_first_error_func,
6031                                 sizeof(es->s_first_error_func));
6032                         es->s_first_error_line =
6033                                 cpu_to_le32(sbi->s_first_error_line);
6034                         es->s_first_error_ino =
6035                                 cpu_to_le32(sbi->s_first_error_ino);
6036                         es->s_first_error_block =
6037                                 cpu_to_le64(sbi->s_first_error_block);
6038                         es->s_first_error_errcode =
6039                                 ext4_errno_to_code(sbi->s_first_error_code);
6040                 }
6041                 __ext4_update_tstamp(&es->s_last_error_time,
6042                                      &es->s_last_error_time_hi,
6043                                      sbi->s_last_error_time);
6044                 strncpy(es->s_last_error_func, sbi->s_last_error_func,
6045                         sizeof(es->s_last_error_func));
6046                 es->s_last_error_line = cpu_to_le32(sbi->s_last_error_line);
6047                 es->s_last_error_ino = cpu_to_le32(sbi->s_last_error_ino);
6048                 es->s_last_error_block = cpu_to_le64(sbi->s_last_error_block);
6049                 es->s_last_error_errcode =
6050                                 ext4_errno_to_code(sbi->s_last_error_code);
6051                 /*
6052                  * Start the daily error reporting function if it hasn't been
6053                  * started already
6054                  */
6055                 if (!es->s_error_count)
6056                         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);
6057                 le32_add_cpu(&es->s_error_count, sbi->s_add_error_count);
6058                 sbi->s_add_error_count = 0;
6059         }
6060         spin_unlock(&sbi->s_error_lock);
6061
6062         ext4_superblock_csum_set(sb);
6063         unlock_buffer(sbh);
6064 }
6065
6066 static int ext4_commit_super(struct super_block *sb)
6067 {
6068         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
6069
6070         if (!sbh)
6071                 return -EINVAL;
6072         if (block_device_ejected(sb))
6073                 return -ENODEV;
6074
6075         ext4_update_super(sb);
6076
6077         lock_buffer(sbh);
6078         /* Buffer got discarded which means block device got invalidated */
6079         if (!buffer_mapped(sbh)) {
6080                 unlock_buffer(sbh);
6081                 return -EIO;
6082         }
6083
6084         if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
6085                 /*
6086                  * Oh, dear.  A previous attempt to write the
6087                  * superblock failed.  This could happen because the
6088                  * USB device was yanked out.  Or it could happen to
6089                  * be a transient write error and maybe the block will
6090                  * be remapped.  Nothing we can do but to retry the
6091                  * write and hope for the best.
6092                  */
6093                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
6094                        "superblock detected");
6095                 clear_buffer_write_io_error(sbh);
6096                 set_buffer_uptodate(sbh);
6097         }
6098         get_bh(sbh);
6099         /* Clear potential dirty bit if it was journalled update */
6100         clear_buffer_dirty(sbh);
6101         sbh->b_end_io = end_buffer_write_sync;
6102         submit_bh(REQ_OP_WRITE | REQ_SYNC |
6103                   (test_opt(sb, BARRIER) ? REQ_FUA : 0), sbh);
6104         wait_on_buffer(sbh);
6105         if (buffer_write_io_error(sbh)) {
6106                 ext4_msg(sb, KERN_ERR, "I/O error while writing "
6107                        "superblock");
6108                 clear_buffer_write_io_error(sbh);
6109                 set_buffer_uptodate(sbh);
6110                 return -EIO;
6111         }
6112         return 0;
6113 }
6114
6115 /*
6116  * Have we just finished recovery?  If so, and if we are mounting (or
6117  * remounting) the filesystem readonly, then we will end up with a
6118  * consistent fs on disk.  Record that fact.
6119  */
6120 static int ext4_mark_recovery_complete(struct super_block *sb,
6121                                        struct ext4_super_block *es)
6122 {
6123         int err;
6124         journal_t *journal = EXT4_SB(sb)->s_journal;
6125
6126         if (!ext4_has_feature_journal(sb)) {
6127                 if (journal != NULL) {
6128                         ext4_error(sb, "Journal got removed while the fs was "
6129                                    "mounted!");
6130                         return -EFSCORRUPTED;
6131                 }
6132                 return 0;
6133         }
6134         jbd2_journal_lock_updates(journal);
6135         err = jbd2_journal_flush(journal, 0);
6136         if (err < 0)
6137                 goto out;
6138
6139         if (sb_rdonly(sb) && (ext4_has_feature_journal_needs_recovery(sb) ||
6140             ext4_has_feature_orphan_present(sb))) {
6141                 if (!ext4_orphan_file_empty(sb)) {
6142                         ext4_error(sb, "Orphan file not empty on read-only fs.");
6143                         err = -EFSCORRUPTED;
6144                         goto out;
6145                 }
6146                 ext4_clear_feature_journal_needs_recovery(sb);
6147                 ext4_clear_feature_orphan_present(sb);
6148                 ext4_commit_super(sb);
6149         }
6150 out:
6151         jbd2_journal_unlock_updates(journal);
6152         return err;
6153 }
6154
6155 /*
6156  * If we are mounting (or read-write remounting) a filesystem whose journal
6157  * has recorded an error from a previous lifetime, move that error to the
6158  * main filesystem now.
6159  */
6160 static int ext4_clear_journal_err(struct super_block *sb,
6161                                    struct ext4_super_block *es)
6162 {
6163         journal_t *journal;
6164         int j_errno;
6165         const char *errstr;
6166
6167         if (!ext4_has_feature_journal(sb)) {
6168                 ext4_error(sb, "Journal got removed while the fs was mounted!");
6169                 return -EFSCORRUPTED;
6170         }
6171
6172         journal = EXT4_SB(sb)->s_journal;
6173
6174         /*
6175          * Now check for any error status which may have been recorded in the
6176          * journal by a prior ext4_error() or ext4_abort()
6177          */
6178
6179         j_errno = jbd2_journal_errno(journal);
6180         if (j_errno) {
6181                 char nbuf[16];
6182
6183                 errstr = ext4_decode_error(sb, j_errno, nbuf);
6184                 ext4_warning(sb, "Filesystem error recorded "
6185                              "from previous mount: %s", errstr);
6186                 ext4_warning(sb, "Marking fs in need of filesystem check.");
6187
6188                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
6189                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
6190                 ext4_commit_super(sb);
6191
6192                 jbd2_journal_clear_err(journal);
6193                 jbd2_journal_update_sb_errno(journal);
6194         }
6195         return 0;
6196 }
6197
6198 /*
6199  * Force the running and committing transactions to commit,
6200  * and wait on the commit.
6201  */
6202 int ext4_force_commit(struct super_block *sb)
6203 {
6204         journal_t *journal;
6205
6206         if (sb_rdonly(sb))
6207                 return 0;
6208
6209         journal = EXT4_SB(sb)->s_journal;
6210         return ext4_journal_force_commit(journal);
6211 }
6212
6213 static int ext4_sync_fs(struct super_block *sb, int wait)
6214 {
6215         int ret = 0;
6216         tid_t target;
6217         bool needs_barrier = false;
6218         struct ext4_sb_info *sbi = EXT4_SB(sb);
6219
6220         if (unlikely(ext4_forced_shutdown(sbi)))
6221                 return 0;
6222
6223         trace_ext4_sync_fs(sb, wait);
6224         flush_workqueue(sbi->rsv_conversion_wq);
6225         /*
6226          * Writeback quota in non-journalled quota case - journalled quota has
6227          * no dirty dquots
6228          */
6229         dquot_writeback_dquots(sb, -1);
6230         /*
6231          * Data writeback is possible w/o journal transaction, so barrier must
6232          * being sent at the end of the function. But we can skip it if
6233          * transaction_commit will do it for us.
6234          */
6235         if (sbi->s_journal) {
6236                 target = jbd2_get_latest_transaction(sbi->s_journal);
6237                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
6238                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
6239                         needs_barrier = true;
6240
6241                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
6242                         if (wait)
6243                                 ret = jbd2_log_wait_commit(sbi->s_journal,
6244                                                            target);
6245                 }
6246         } else if (wait && test_opt(sb, BARRIER))
6247                 needs_barrier = true;
6248         if (needs_barrier) {
6249                 int err;
6250                 err = blkdev_issue_flush(sb->s_bdev);
6251                 if (!ret)
6252                         ret = err;
6253         }
6254
6255         return ret;
6256 }
6257
6258 /*
6259  * LVM calls this function before a (read-only) snapshot is created.  This
6260  * gives us a chance to flush the journal completely and mark the fs clean.
6261  *
6262  * Note that only this function cannot bring a filesystem to be in a clean
6263  * state independently. It relies on upper layer to stop all data & metadata
6264  * modifications.
6265  */
6266 static int ext4_freeze(struct super_block *sb)
6267 {
6268         int error = 0;
6269         journal_t *journal;
6270
6271         if (sb_rdonly(sb))
6272                 return 0;
6273
6274         journal = EXT4_SB(sb)->s_journal;
6275
6276         if (journal) {
6277                 /* Now we set up the journal barrier. */
6278                 jbd2_journal_lock_updates(journal);
6279
6280                 /*
6281                  * Don't clear the needs_recovery flag if we failed to
6282                  * flush the journal.
6283                  */
6284                 error = jbd2_journal_flush(journal, 0);
6285                 if (error < 0)
6286                         goto out;
6287
6288                 /* Journal blocked and flushed, clear needs_recovery flag. */
6289                 ext4_clear_feature_journal_needs_recovery(sb);
6290                 if (ext4_orphan_file_empty(sb))
6291                         ext4_clear_feature_orphan_present(sb);
6292         }
6293
6294         error = ext4_commit_super(sb);
6295 out:
6296         if (journal)
6297                 /* we rely on upper layer to stop further updates */
6298                 jbd2_journal_unlock_updates(journal);
6299         return error;
6300 }
6301
6302 /*
6303  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
6304  * flag here, even though the filesystem is not technically dirty yet.
6305  */
6306 static int ext4_unfreeze(struct super_block *sb)
6307 {
6308         if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
6309                 return 0;
6310
6311         if (EXT4_SB(sb)->s_journal) {
6312                 /* Reset the needs_recovery flag before the fs is unlocked. */
6313                 ext4_set_feature_journal_needs_recovery(sb);
6314                 if (ext4_has_feature_orphan_file(sb))
6315                         ext4_set_feature_orphan_present(sb);
6316         }
6317
6318         ext4_commit_super(sb);
6319         return 0;
6320 }
6321
6322 /*
6323  * Structure to save mount options for ext4_remount's benefit
6324  */
6325 struct ext4_mount_options {
6326         unsigned long s_mount_opt;
6327         unsigned long s_mount_opt2;
6328         kuid_t s_resuid;
6329         kgid_t s_resgid;
6330         unsigned long s_commit_interval;
6331         u32 s_min_batch_time, s_max_batch_time;
6332 #ifdef CONFIG_QUOTA
6333         int s_jquota_fmt;
6334         char *s_qf_names[EXT4_MAXQUOTAS];
6335 #endif
6336 };
6337
6338 static int __ext4_remount(struct fs_context *fc, struct super_block *sb)
6339 {
6340         struct ext4_fs_context *ctx = fc->fs_private;
6341         struct ext4_super_block *es;
6342         struct ext4_sb_info *sbi = EXT4_SB(sb);
6343         unsigned long old_sb_flags;
6344         struct ext4_mount_options old_opts;
6345         ext4_group_t g;
6346         int err = 0;
6347 #ifdef CONFIG_QUOTA
6348         int enable_quota = 0;
6349         int i, j;
6350         char *to_free[EXT4_MAXQUOTAS];
6351 #endif
6352
6353
6354         /* Store the original options */
6355         old_sb_flags = sb->s_flags;
6356         old_opts.s_mount_opt = sbi->s_mount_opt;
6357         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
6358         old_opts.s_resuid = sbi->s_resuid;
6359         old_opts.s_resgid = sbi->s_resgid;
6360         old_opts.s_commit_interval = sbi->s_commit_interval;
6361         old_opts.s_min_batch_time = sbi->s_min_batch_time;
6362         old_opts.s_max_batch_time = sbi->s_max_batch_time;
6363 #ifdef CONFIG_QUOTA
6364         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
6365         for (i = 0; i < EXT4_MAXQUOTAS; i++)
6366                 if (sbi->s_qf_names[i]) {
6367                         char *qf_name = get_qf_name(sb, sbi, i);
6368
6369                         old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL);
6370                         if (!old_opts.s_qf_names[i]) {
6371                                 for (j = 0; j < i; j++)
6372                                         kfree(old_opts.s_qf_names[j]);
6373                                 return -ENOMEM;
6374                         }
6375                 } else
6376                         old_opts.s_qf_names[i] = NULL;
6377 #endif
6378         if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)) {
6379                 if (sbi->s_journal && sbi->s_journal->j_task->io_context)
6380                         ctx->journal_ioprio =
6381                                 sbi->s_journal->j_task->io_context->ioprio;
6382                 else
6383                         ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
6384
6385         }
6386
6387         ext4_apply_options(fc, sb);
6388
6389         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
6390             test_opt(sb, JOURNAL_CHECKSUM)) {
6391                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
6392                          "during remount not supported; ignoring");
6393                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
6394         }
6395
6396         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
6397                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
6398                         ext4_msg(sb, KERN_ERR, "can't mount with "
6399                                  "both data=journal and delalloc");
6400                         err = -EINVAL;
6401                         goto restore_opts;
6402                 }
6403                 if (test_opt(sb, DIOREAD_NOLOCK)) {
6404                         ext4_msg(sb, KERN_ERR, "can't mount with "
6405                                  "both data=journal and dioread_nolock");
6406                         err = -EINVAL;
6407                         goto restore_opts;
6408                 }
6409         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
6410                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
6411                         ext4_msg(sb, KERN_ERR, "can't mount with "
6412                                 "journal_async_commit in data=ordered mode");
6413                         err = -EINVAL;
6414                         goto restore_opts;
6415                 }
6416         }
6417
6418         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
6419                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
6420                 err = -EINVAL;
6421                 goto restore_opts;
6422         }
6423
6424         if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED))
6425                 ext4_abort(sb, ESHUTDOWN, "Abort forced by user");
6426
6427         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
6428                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
6429
6430         es = sbi->s_es;
6431
6432         if (sbi->s_journal) {
6433                 ext4_init_journal_params(sb, sbi->s_journal);
6434                 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio);
6435         }
6436
6437         /* Flush outstanding errors before changing fs state */
6438         flush_work(&sbi->s_error_work);
6439
6440         if ((bool)(fc->sb_flags & SB_RDONLY) != sb_rdonly(sb)) {
6441                 if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED)) {
6442                         err = -EROFS;
6443                         goto restore_opts;
6444                 }
6445
6446                 if (fc->sb_flags & SB_RDONLY) {
6447                         err = sync_filesystem(sb);
6448                         if (err < 0)
6449                                 goto restore_opts;
6450                         err = dquot_suspend(sb, -1);
6451                         if (err < 0)
6452                                 goto restore_opts;
6453
6454                         /*
6455                          * First of all, the unconditional stuff we have to do
6456                          * to disable replay of the journal when we next remount
6457                          */
6458                         sb->s_flags |= SB_RDONLY;
6459
6460                         /*
6461                          * OK, test if we are remounting a valid rw partition
6462                          * readonly, and if so set the rdonly flag and then
6463                          * mark the partition as valid again.
6464                          */
6465                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
6466                             (sbi->s_mount_state & EXT4_VALID_FS))
6467                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
6468
6469                         if (sbi->s_journal) {
6470                                 /*
6471                                  * We let remount-ro finish even if marking fs
6472                                  * as clean failed...
6473                                  */
6474                                 ext4_mark_recovery_complete(sb, es);
6475                         }
6476                 } else {
6477                         /* Make sure we can mount this feature set readwrite */
6478                         if (ext4_has_feature_readonly(sb) ||
6479                             !ext4_feature_set_ok(sb, 0)) {
6480                                 err = -EROFS;
6481                                 goto restore_opts;
6482                         }
6483                         /*
6484                          * Make sure the group descriptor checksums
6485                          * are sane.  If they aren't, refuse to remount r/w.
6486                          */
6487                         for (g = 0; g < sbi->s_groups_count; g++) {
6488                                 struct ext4_group_desc *gdp =
6489                                         ext4_get_group_desc(sb, g, NULL);
6490
6491                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
6492                                         ext4_msg(sb, KERN_ERR,
6493                "ext4_remount: Checksum for group %u failed (%u!=%u)",
6494                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
6495                                                le16_to_cpu(gdp->bg_checksum));
6496                                         err = -EFSBADCRC;
6497                                         goto restore_opts;
6498                                 }
6499                         }
6500
6501                         /*
6502                          * If we have an unprocessed orphan list hanging
6503                          * around from a previously readonly bdev mount,
6504                          * require a full umount/remount for now.
6505                          */
6506                         if (es->s_last_orphan || !ext4_orphan_file_empty(sb)) {
6507                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
6508                                        "remount RDWR because of unprocessed "
6509                                        "orphan inode list.  Please "
6510                                        "umount/remount instead");
6511                                 err = -EINVAL;
6512                                 goto restore_opts;
6513                         }
6514
6515                         /*
6516                          * Mounting a RDONLY partition read-write, so reread
6517                          * and store the current valid flag.  (It may have
6518                          * been changed by e2fsck since we originally mounted
6519                          * the partition.)
6520                          */
6521                         if (sbi->s_journal) {
6522                                 err = ext4_clear_journal_err(sb, es);
6523                                 if (err)
6524                                         goto restore_opts;
6525                         }
6526                         sbi->s_mount_state = (le16_to_cpu(es->s_state) &
6527                                               ~EXT4_FC_REPLAY);
6528
6529                         err = ext4_setup_super(sb, es, 0);
6530                         if (err)
6531                                 goto restore_opts;
6532
6533                         sb->s_flags &= ~SB_RDONLY;
6534                         if (ext4_has_feature_mmp(sb)) {
6535                                 err = ext4_multi_mount_protect(sb,
6536                                                 le64_to_cpu(es->s_mmp_block));
6537                                 if (err)
6538                                         goto restore_opts;
6539                         }
6540 #ifdef CONFIG_QUOTA
6541                         enable_quota = 1;
6542 #endif
6543                 }
6544         }
6545
6546         /*
6547          * Handle creation of system zone data early because it can fail.
6548          * Releasing of existing data is done when we are sure remount will
6549          * succeed.
6550          */
6551         if (test_opt(sb, BLOCK_VALIDITY) && !sbi->s_system_blks) {
6552                 err = ext4_setup_system_zone(sb);
6553                 if (err)
6554                         goto restore_opts;
6555         }
6556
6557         if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) {
6558                 err = ext4_commit_super(sb);
6559                 if (err)
6560                         goto restore_opts;
6561         }
6562
6563 #ifdef CONFIG_QUOTA
6564         if (enable_quota) {
6565                 if (sb_any_quota_suspended(sb))
6566                         dquot_resume(sb, -1);
6567                 else if (ext4_has_feature_quota(sb)) {
6568                         err = ext4_enable_quotas(sb);
6569                         if (err)
6570                                 goto restore_opts;
6571                 }
6572         }
6573         /* Release old quota file names */
6574         for (i = 0; i < EXT4_MAXQUOTAS; i++)
6575                 kfree(old_opts.s_qf_names[i]);
6576 #endif
6577         if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks)
6578                 ext4_release_system_zone(sb);
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         if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb))
6593                 ext4_stop_mmpd(sbi);
6594
6595         return 0;
6596
6597 restore_opts:
6598         /*
6599          * If there was a failing r/w to ro transition, we may need to
6600          * re-enable quota
6601          */
6602         if ((sb->s_flags & SB_RDONLY) && !(old_sb_flags & SB_RDONLY) &&
6603             sb_any_quota_suspended(sb))
6604                 dquot_resume(sb, -1);
6605         sb->s_flags = old_sb_flags;
6606         sbi->s_mount_opt = old_opts.s_mount_opt;
6607         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
6608         sbi->s_resuid = old_opts.s_resuid;
6609         sbi->s_resgid = old_opts.s_resgid;
6610         sbi->s_commit_interval = old_opts.s_commit_interval;
6611         sbi->s_min_batch_time = old_opts.s_min_batch_time;
6612         sbi->s_max_batch_time = old_opts.s_max_batch_time;
6613         if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks)
6614                 ext4_release_system_zone(sb);
6615 #ifdef CONFIG_QUOTA
6616         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
6617         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
6618                 to_free[i] = get_qf_name(sb, sbi, i);
6619                 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]);
6620         }
6621         synchronize_rcu();
6622         for (i = 0; i < EXT4_MAXQUOTAS; i++)
6623                 kfree(to_free[i]);
6624 #endif
6625         if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb))
6626                 ext4_stop_mmpd(sbi);
6627         return err;
6628 }
6629
6630 static int ext4_reconfigure(struct fs_context *fc)
6631 {
6632         struct super_block *sb = fc->root->d_sb;
6633         int ret;
6634
6635         fc->s_fs_info = EXT4_SB(sb);
6636
6637         ret = ext4_check_opt_consistency(fc, sb);
6638         if (ret < 0)
6639                 return ret;
6640
6641         ret = __ext4_remount(fc, sb);
6642         if (ret < 0)
6643                 return ret;
6644
6645         ext4_msg(sb, KERN_INFO, "re-mounted. Quota mode: %s.",
6646                  ext4_quota_mode(sb));
6647
6648         return 0;
6649 }
6650
6651 #ifdef CONFIG_QUOTA
6652 static int ext4_statfs_project(struct super_block *sb,
6653                                kprojid_t projid, struct kstatfs *buf)
6654 {
6655         struct kqid qid;
6656         struct dquot *dquot;
6657         u64 limit;
6658         u64 curblock;
6659
6660         qid = make_kqid_projid(projid);
6661         dquot = dqget(sb, qid);
6662         if (IS_ERR(dquot))
6663                 return PTR_ERR(dquot);
6664         spin_lock(&dquot->dq_dqb_lock);
6665
6666         limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit,
6667                              dquot->dq_dqb.dqb_bhardlimit);
6668         limit >>= sb->s_blocksize_bits;
6669
6670         if (limit && buf->f_blocks > limit) {
6671                 curblock = (dquot->dq_dqb.dqb_curspace +
6672                             dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits;
6673                 buf->f_blocks = limit;
6674                 buf->f_bfree = buf->f_bavail =
6675                         (buf->f_blocks > curblock) ?
6676                          (buf->f_blocks - curblock) : 0;
6677         }
6678
6679         limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit,
6680                              dquot->dq_dqb.dqb_ihardlimit);
6681         if (limit && buf->f_files > limit) {
6682                 buf->f_files = limit;
6683                 buf->f_ffree =
6684                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
6685                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
6686         }
6687
6688         spin_unlock(&dquot->dq_dqb_lock);
6689         dqput(dquot);
6690         return 0;
6691 }
6692 #endif
6693
6694 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
6695 {
6696         struct super_block *sb = dentry->d_sb;
6697         struct ext4_sb_info *sbi = EXT4_SB(sb);
6698         struct ext4_super_block *es = sbi->s_es;
6699         ext4_fsblk_t overhead = 0, resv_blocks;
6700         s64 bfree;
6701         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
6702
6703         if (!test_opt(sb, MINIX_DF))
6704                 overhead = sbi->s_overhead;
6705
6706         buf->f_type = EXT4_SUPER_MAGIC;
6707         buf->f_bsize = sb->s_blocksize;
6708         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
6709         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
6710                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
6711         /* prevent underflow in case that few free space is available */
6712         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
6713         buf->f_bavail = buf->f_bfree -
6714                         (ext4_r_blocks_count(es) + resv_blocks);
6715         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
6716                 buf->f_bavail = 0;
6717         buf->f_files = le32_to_cpu(es->s_inodes_count);
6718         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
6719         buf->f_namelen = EXT4_NAME_LEN;
6720         buf->f_fsid = uuid_to_fsid(es->s_uuid);
6721
6722 #ifdef CONFIG_QUOTA
6723         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
6724             sb_has_quota_limits_enabled(sb, PRJQUOTA))
6725                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
6726 #endif
6727         return 0;
6728 }
6729
6730
6731 #ifdef CONFIG_QUOTA
6732
6733 /*
6734  * Helper functions so that transaction is started before we acquire dqio_sem
6735  * to keep correct lock ordering of transaction > dqio_sem
6736  */
6737 static inline struct inode *dquot_to_inode(struct dquot *dquot)
6738 {
6739         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
6740 }
6741
6742 static int ext4_write_dquot(struct dquot *dquot)
6743 {
6744         int ret, err;
6745         handle_t *handle;
6746         struct inode *inode;
6747
6748         inode = dquot_to_inode(dquot);
6749         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
6750                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
6751         if (IS_ERR(handle))
6752                 return PTR_ERR(handle);
6753         ret = dquot_commit(dquot);
6754         if (ret < 0)
6755                 ext4_error_err(dquot->dq_sb, -ret,
6756                                "Failed to commit dquot type %d",
6757                                dquot->dq_id.type);
6758         err = ext4_journal_stop(handle);
6759         if (!ret)
6760                 ret = err;
6761         return ret;
6762 }
6763
6764 static int ext4_acquire_dquot(struct dquot *dquot)
6765 {
6766         int ret, err;
6767         handle_t *handle;
6768
6769         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
6770                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
6771         if (IS_ERR(handle))
6772                 return PTR_ERR(handle);
6773         ret = dquot_acquire(dquot);
6774         if (ret < 0)
6775                 ext4_error_err(dquot->dq_sb, -ret,
6776                               "Failed to acquire dquot type %d",
6777                               dquot->dq_id.type);
6778         err = ext4_journal_stop(handle);
6779         if (!ret)
6780                 ret = err;
6781         return ret;
6782 }
6783
6784 static int ext4_release_dquot(struct dquot *dquot)
6785 {
6786         int ret, err;
6787         handle_t *handle;
6788
6789         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
6790                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
6791         if (IS_ERR(handle)) {
6792                 /* Release dquot anyway to avoid endless cycle in dqput() */
6793                 dquot_release(dquot);
6794                 return PTR_ERR(handle);
6795         }
6796         ret = dquot_release(dquot);
6797         if (ret < 0)
6798                 ext4_error_err(dquot->dq_sb, -ret,
6799                                "Failed to release dquot type %d",
6800                                dquot->dq_id.type);
6801         err = ext4_journal_stop(handle);
6802         if (!ret)
6803                 ret = err;
6804         return ret;
6805 }
6806
6807 static int ext4_mark_dquot_dirty(struct dquot *dquot)
6808 {
6809         struct super_block *sb = dquot->dq_sb;
6810
6811         if (ext4_is_quota_journalled(sb)) {
6812                 dquot_mark_dquot_dirty(dquot);
6813                 return ext4_write_dquot(dquot);
6814         } else {
6815                 return dquot_mark_dquot_dirty(dquot);
6816         }
6817 }
6818
6819 static int ext4_write_info(struct super_block *sb, int type)
6820 {
6821         int ret, err;
6822         handle_t *handle;
6823
6824         /* Data block + inode block */
6825         handle = ext4_journal_start_sb(sb, EXT4_HT_QUOTA, 2);
6826         if (IS_ERR(handle))
6827                 return PTR_ERR(handle);
6828         ret = dquot_commit_info(sb, type);
6829         err = ext4_journal_stop(handle);
6830         if (!ret)
6831                 ret = err;
6832         return ret;
6833 }
6834
6835 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
6836 {
6837         struct ext4_inode_info *ei = EXT4_I(inode);
6838
6839         /* The first argument of lockdep_set_subclass has to be
6840          * *exactly* the same as the argument to init_rwsem() --- in
6841          * this case, in init_once() --- or lockdep gets unhappy
6842          * because the name of the lock is set using the
6843          * stringification of the argument to init_rwsem().
6844          */
6845         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
6846         lockdep_set_subclass(&ei->i_data_sem, subclass);
6847 }
6848
6849 /*
6850  * Standard function to be called on quota_on
6851  */
6852 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
6853                          const struct path *path)
6854 {
6855         int err;
6856
6857         if (!test_opt(sb, QUOTA))
6858                 return -EINVAL;
6859
6860         /* Quotafile not on the same filesystem? */
6861         if (path->dentry->d_sb != sb)
6862                 return -EXDEV;
6863
6864         /* Quota already enabled for this file? */
6865         if (IS_NOQUOTA(d_inode(path->dentry)))
6866                 return -EBUSY;
6867
6868         /* Journaling quota? */
6869         if (EXT4_SB(sb)->s_qf_names[type]) {
6870                 /* Quotafile not in fs root? */
6871                 if (path->dentry->d_parent != sb->s_root)
6872                         ext4_msg(sb, KERN_WARNING,
6873                                 "Quota file not on filesystem root. "
6874                                 "Journaled quota will not work");
6875                 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
6876         } else {
6877                 /*
6878                  * Clear the flag just in case mount options changed since
6879                  * last time.
6880                  */
6881                 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
6882         }
6883
6884         /*
6885          * When we journal data on quota file, we have to flush journal to see
6886          * all updates to the file when we bypass pagecache...
6887          */
6888         if (EXT4_SB(sb)->s_journal &&
6889             ext4_should_journal_data(d_inode(path->dentry))) {
6890                 /*
6891                  * We don't need to lock updates but journal_flush() could
6892                  * otherwise be livelocked...
6893                  */
6894                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
6895                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal, 0);
6896                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
6897                 if (err)
6898                         return err;
6899         }
6900
6901         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
6902         err = dquot_quota_on(sb, type, format_id, path);
6903         if (!err) {
6904                 struct inode *inode = d_inode(path->dentry);
6905                 handle_t *handle;
6906
6907                 /*
6908                  * Set inode flags to prevent userspace from messing with quota
6909                  * files. If this fails, we return success anyway since quotas
6910                  * are already enabled and this is not a hard failure.
6911                  */
6912                 inode_lock(inode);
6913                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
6914                 if (IS_ERR(handle))
6915                         goto unlock_inode;
6916                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
6917                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
6918                                 S_NOATIME | S_IMMUTABLE);
6919                 err = ext4_mark_inode_dirty(handle, inode);
6920                 ext4_journal_stop(handle);
6921         unlock_inode:
6922                 inode_unlock(inode);
6923                 if (err)
6924                         dquot_quota_off(sb, type);
6925         }
6926         if (err)
6927                 lockdep_set_quota_inode(path->dentry->d_inode,
6928                                              I_DATA_SEM_NORMAL);
6929         return err;
6930 }
6931
6932 static inline bool ext4_check_quota_inum(int type, unsigned long qf_inum)
6933 {
6934         switch (type) {
6935         case USRQUOTA:
6936                 return qf_inum == EXT4_USR_QUOTA_INO;
6937         case GRPQUOTA:
6938                 return qf_inum == EXT4_GRP_QUOTA_INO;
6939         case PRJQUOTA:
6940                 return qf_inum >= EXT4_GOOD_OLD_FIRST_INO;
6941         default:
6942                 BUG();
6943         }
6944 }
6945
6946 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
6947                              unsigned int flags)
6948 {
6949         int err;
6950         struct inode *qf_inode;
6951         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
6952                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
6953                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
6954                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
6955         };
6956
6957         BUG_ON(!ext4_has_feature_quota(sb));
6958
6959         if (!qf_inums[type])
6960                 return -EPERM;
6961
6962         if (!ext4_check_quota_inum(type, qf_inums[type])) {
6963                 ext4_error(sb, "Bad quota inum: %lu, type: %d",
6964                                 qf_inums[type], type);
6965                 return -EUCLEAN;
6966         }
6967
6968         qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL);
6969         if (IS_ERR(qf_inode)) {
6970                 ext4_error(sb, "Bad quota inode: %lu, type: %d",
6971                                 qf_inums[type], type);
6972                 return PTR_ERR(qf_inode);
6973         }
6974
6975         /* Don't account quota for quota files to avoid recursion */
6976         qf_inode->i_flags |= S_NOQUOTA;
6977         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
6978         err = dquot_load_quota_inode(qf_inode, type, format_id, flags);
6979         if (err)
6980                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
6981         iput(qf_inode);
6982
6983         return err;
6984 }
6985
6986 /* Enable usage tracking for all quota types. */
6987 int ext4_enable_quotas(struct super_block *sb)
6988 {
6989         int type, err = 0;
6990         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
6991                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
6992                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
6993                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
6994         };
6995         bool quota_mopt[EXT4_MAXQUOTAS] = {
6996                 test_opt(sb, USRQUOTA),
6997                 test_opt(sb, GRPQUOTA),
6998                 test_opt(sb, PRJQUOTA),
6999         };
7000
7001         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
7002         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
7003                 if (qf_inums[type]) {
7004                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
7005                                 DQUOT_USAGE_ENABLED |
7006                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
7007                         if (err) {
7008                                 ext4_warning(sb,
7009                                         "Failed to enable quota tracking "
7010                                         "(type=%d, err=%d, ino=%lu). "
7011                                         "Please run e2fsck to fix.", type,
7012                                         err, qf_inums[type]);
7013                                 for (type--; type >= 0; type--) {
7014                                         struct inode *inode;
7015
7016                                         inode = sb_dqopt(sb)->files[type];
7017                                         if (inode)
7018                                                 inode = igrab(inode);
7019                                         dquot_quota_off(sb, type);
7020                                         if (inode) {
7021                                                 lockdep_set_quota_inode(inode,
7022                                                         I_DATA_SEM_NORMAL);
7023                                                 iput(inode);
7024                                         }
7025                                 }
7026
7027                                 return err;
7028                         }
7029                 }
7030         }
7031         return 0;
7032 }
7033
7034 static int ext4_quota_off(struct super_block *sb, int type)
7035 {
7036         struct inode *inode = sb_dqopt(sb)->files[type];
7037         handle_t *handle;
7038         int err;
7039
7040         /* Force all delayed allocation blocks to be allocated.
7041          * Caller already holds s_umount sem */
7042         if (test_opt(sb, DELALLOC))
7043                 sync_filesystem(sb);
7044
7045         if (!inode || !igrab(inode))
7046                 goto out;
7047
7048         err = dquot_quota_off(sb, type);
7049         if (err || ext4_has_feature_quota(sb))
7050                 goto out_put;
7051
7052         inode_lock(inode);
7053         /*
7054          * Update modification times of quota files when userspace can
7055          * start looking at them. If we fail, we return success anyway since
7056          * this is not a hard failure and quotas are already disabled.
7057          */
7058         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
7059         if (IS_ERR(handle)) {
7060                 err = PTR_ERR(handle);
7061                 goto out_unlock;
7062         }
7063         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
7064         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
7065         inode->i_mtime = inode->i_ctime = current_time(inode);
7066         err = ext4_mark_inode_dirty(handle, inode);
7067         ext4_journal_stop(handle);
7068 out_unlock:
7069         inode_unlock(inode);
7070 out_put:
7071         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
7072         iput(inode);
7073         return err;
7074 out:
7075         return dquot_quota_off(sb, type);
7076 }
7077
7078 /* Read data from quotafile - avoid pagecache and such because we cannot afford
7079  * acquiring the locks... As quota files are never truncated and quota code
7080  * itself serializes the operations (and no one else should touch the files)
7081  * we don't have to be afraid of races */
7082 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
7083                                size_t len, loff_t off)
7084 {
7085         struct inode *inode = sb_dqopt(sb)->files[type];
7086         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
7087         int offset = off & (sb->s_blocksize - 1);
7088         int tocopy;
7089         size_t toread;
7090         struct buffer_head *bh;
7091         loff_t i_size = i_size_read(inode);
7092
7093         if (off > i_size)
7094                 return 0;
7095         if (off+len > i_size)
7096                 len = i_size-off;
7097         toread = len;
7098         while (toread > 0) {
7099                 tocopy = sb->s_blocksize - offset < toread ?
7100                                 sb->s_blocksize - offset : toread;
7101                 bh = ext4_bread(NULL, inode, blk, 0);
7102                 if (IS_ERR(bh))
7103                         return PTR_ERR(bh);
7104                 if (!bh)        /* A hole? */
7105                         memset(data, 0, tocopy);
7106                 else
7107                         memcpy(data, bh->b_data+offset, tocopy);
7108                 brelse(bh);
7109                 offset = 0;
7110                 toread -= tocopy;
7111                 data += tocopy;
7112                 blk++;
7113         }
7114         return len;
7115 }
7116
7117 /* Write to quotafile (we know the transaction is already started and has
7118  * enough credits) */
7119 static ssize_t ext4_quota_write(struct super_block *sb, int type,
7120                                 const char *data, size_t len, loff_t off)
7121 {
7122         struct inode *inode = sb_dqopt(sb)->files[type];
7123         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
7124         int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1);
7125         int retries = 0;
7126         struct buffer_head *bh;
7127         handle_t *handle = journal_current_handle();
7128
7129         if (!handle) {
7130                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
7131                         " cancelled because transaction is not started",
7132                         (unsigned long long)off, (unsigned long long)len);
7133                 return -EIO;
7134         }
7135         /*
7136          * Since we account only one data block in transaction credits,
7137          * then it is impossible to cross a block boundary.
7138          */
7139         if (sb->s_blocksize - offset < len) {
7140                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
7141                         " cancelled because not block aligned",
7142                         (unsigned long long)off, (unsigned long long)len);
7143                 return -EIO;
7144         }
7145
7146         do {
7147                 bh = ext4_bread(handle, inode, blk,
7148                                 EXT4_GET_BLOCKS_CREATE |
7149                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
7150         } while (PTR_ERR(bh) == -ENOSPC &&
7151                  ext4_should_retry_alloc(inode->i_sb, &retries));
7152         if (IS_ERR(bh))
7153                 return PTR_ERR(bh);
7154         if (!bh)
7155                 goto out;
7156         BUFFER_TRACE(bh, "get write access");
7157         err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
7158         if (err) {
7159                 brelse(bh);
7160                 return err;
7161         }
7162         lock_buffer(bh);
7163         memcpy(bh->b_data+offset, data, len);
7164         flush_dcache_page(bh->b_page);
7165         unlock_buffer(bh);
7166         err = ext4_handle_dirty_metadata(handle, NULL, bh);
7167         brelse(bh);
7168 out:
7169         if (inode->i_size < off + len) {
7170                 i_size_write(inode, off + len);
7171                 EXT4_I(inode)->i_disksize = inode->i_size;
7172                 err2 = ext4_mark_inode_dirty(handle, inode);
7173                 if (unlikely(err2 && !err))
7174                         err = err2;
7175         }
7176         return err ? err : len;
7177 }
7178 #endif
7179
7180 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
7181 static inline void register_as_ext2(void)
7182 {
7183         int err = register_filesystem(&ext2_fs_type);
7184         if (err)
7185                 printk(KERN_WARNING
7186                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
7187 }
7188
7189 static inline void unregister_as_ext2(void)
7190 {
7191         unregister_filesystem(&ext2_fs_type);
7192 }
7193
7194 static inline int ext2_feature_set_ok(struct super_block *sb)
7195 {
7196         if (ext4_has_unknown_ext2_incompat_features(sb))
7197                 return 0;
7198         if (sb_rdonly(sb))
7199                 return 1;
7200         if (ext4_has_unknown_ext2_ro_compat_features(sb))
7201                 return 0;
7202         return 1;
7203 }
7204 #else
7205 static inline void register_as_ext2(void) { }
7206 static inline void unregister_as_ext2(void) { }
7207 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
7208 #endif
7209
7210 static inline void register_as_ext3(void)
7211 {
7212         int err = register_filesystem(&ext3_fs_type);
7213         if (err)
7214                 printk(KERN_WARNING
7215                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
7216 }
7217
7218 static inline void unregister_as_ext3(void)
7219 {
7220         unregister_filesystem(&ext3_fs_type);
7221 }
7222
7223 static inline int ext3_feature_set_ok(struct super_block *sb)
7224 {
7225         if (ext4_has_unknown_ext3_incompat_features(sb))
7226                 return 0;
7227         if (!ext4_has_feature_journal(sb))
7228                 return 0;
7229         if (sb_rdonly(sb))
7230                 return 1;
7231         if (ext4_has_unknown_ext3_ro_compat_features(sb))
7232                 return 0;
7233         return 1;
7234 }
7235
7236 static struct file_system_type ext4_fs_type = {
7237         .owner                  = THIS_MODULE,
7238         .name                   = "ext4",
7239         .init_fs_context        = ext4_init_fs_context,
7240         .parameters             = ext4_param_specs,
7241         .kill_sb                = kill_block_super,
7242         .fs_flags               = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
7243 };
7244 MODULE_ALIAS_FS("ext4");
7245
7246 /* Shared across all ext4 file systems */
7247 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
7248
7249 static int __init ext4_init_fs(void)
7250 {
7251         int i, err;
7252
7253         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
7254         ext4_li_info = NULL;
7255
7256         /* Build-time check for flags consistency */
7257         ext4_check_flag_values();
7258
7259         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
7260                 init_waitqueue_head(&ext4__ioend_wq[i]);
7261
7262         err = ext4_init_es();
7263         if (err)
7264                 return err;
7265
7266         err = ext4_init_pending();
7267         if (err)
7268                 goto out7;
7269
7270         err = ext4_init_post_read_processing();
7271         if (err)
7272                 goto out6;
7273
7274         err = ext4_init_pageio();
7275         if (err)
7276                 goto out5;
7277
7278         err = ext4_init_system_zone();
7279         if (err)
7280                 goto out4;
7281
7282         err = ext4_init_sysfs();
7283         if (err)
7284                 goto out3;
7285
7286         err = ext4_init_mballoc();
7287         if (err)
7288                 goto out2;
7289         err = init_inodecache();
7290         if (err)
7291                 goto out1;
7292
7293         err = ext4_fc_init_dentry_cache();
7294         if (err)
7295                 goto out05;
7296
7297         register_as_ext3();
7298         register_as_ext2();
7299         err = register_filesystem(&ext4_fs_type);
7300         if (err)
7301                 goto out;
7302
7303         return 0;
7304 out:
7305         unregister_as_ext2();
7306         unregister_as_ext3();
7307         ext4_fc_destroy_dentry_cache();
7308 out05:
7309         destroy_inodecache();
7310 out1:
7311         ext4_exit_mballoc();
7312 out2:
7313         ext4_exit_sysfs();
7314 out3:
7315         ext4_exit_system_zone();
7316 out4:
7317         ext4_exit_pageio();
7318 out5:
7319         ext4_exit_post_read_processing();
7320 out6:
7321         ext4_exit_pending();
7322 out7:
7323         ext4_exit_es();
7324
7325         return err;
7326 }
7327
7328 static void __exit ext4_exit_fs(void)
7329 {
7330         ext4_destroy_lazyinit_thread();
7331         unregister_as_ext2();
7332         unregister_as_ext3();
7333         unregister_filesystem(&ext4_fs_type);
7334         ext4_fc_destroy_dentry_cache();
7335         destroy_inodecache();
7336         ext4_exit_mballoc();
7337         ext4_exit_sysfs();
7338         ext4_exit_system_zone();
7339         ext4_exit_pageio();
7340         ext4_exit_post_read_processing();
7341         ext4_exit_es();
7342         ext4_exit_pending();
7343 }
7344
7345 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
7346 MODULE_DESCRIPTION("Fourth Extended Filesystem");
7347 MODULE_LICENSE("GPL");
7348 MODULE_SOFTDEP("pre: crc32c");
7349 module_init(ext4_init_fs)
7350 module_exit(ext4_exit_fs)