fs/ext4/super.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/ext4/super.c
   4  *
   5  * Copyright (C) 1992, 1993, 1994, 1995
   6  * Remy Card ([email protected])
   7  * Laboratoire MASI - Institut Blaise Pascal
   8  * Universite Pierre et Marie Curie (Paris VI)
   9  *
  10  *  from
  11  *
  12  *  linux/fs/minix/inode.c
  13  *
  14  *  Copyright (C) 1991, 1992  Linus Torvalds
  15  *
  16  *  Big-endian to little-endian byte-swapping/bitmaps by
  17  *        David S. Miller ([email protected]), 1995
  18  */
  19
  20 #include <linux/module.h>
  21 #include <linux/string.h>
  22 #include <linux/fs.h>
  23 #include <linux/time.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/backing-dev.h>
  29 #include <linux/parser.h>
  30 #include <linux/buffer_head.h>
  31 #include <linux/exportfs.h>
  32 #include <linux/vfs.h>
  33 #include <linux/random.h>
  34 #include <linux/mount.h>
  35 #include <linux/namei.h>
  36 #include <linux/quotaops.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/dax.h>
  42 #include <linux/cleancache.h>
  43 #include <linux/uaccess.h>
  44 #include <linux/iversion.h>
  45 #include <linux/unicode.h>
  46 #include <linux/part_stat.h>
  47 #include <linux/kthread.h>
  48 #include <linux/freezer.h>
  49
  50 #include "ext4.h"
  51 #include "ext4_extents.h"       /* Needed for trace points definition */
  52 #include "ext4_jbd2.h"
  53 #include "xattr.h"
  54 #include "acl.h"
  55 #include "mballoc.h"
  56 #include "fsmap.h"
  57
  58 #define CREATE_TRACE_POINTS
  59 #include <trace/events/ext4.h>
  60
  61 static struct ext4_lazy_init *ext4_li_info;
  62 static DEFINE_MUTEX(ext4_li_mtx);
  63 static struct ratelimit_state ext4_mount_msg_ratelimit;
  64
  65 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  66                              unsigned long journal_devnum);
  67 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  68 static void ext4_update_super(struct super_block *sb);
  69 static int ext4_commit_super(struct super_block *sb);
  70 static int ext4_mark_recovery_complete(struct super_block *sb,
  71                                         struct ext4_super_block *es);
  72 static int ext4_clear_journal_err(struct super_block *sb,
  73                                   struct ext4_super_block *es);
  74 static int ext4_sync_fs(struct super_block *sb, int wait);
  75 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  76 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  77 static int ext4_unfreeze(struct super_block *sb);
  78 static int ext4_freeze(struct super_block *sb);
  79 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  80                        const char *dev_name, void *data);
  81 static inline int ext2_feature_set_ok(struct super_block *sb);
  82 static inline int ext3_feature_set_ok(struct super_block *sb);
  83 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  84 static void ext4_destroy_lazyinit_thread(void);
  85 static void ext4_unregister_li_request(struct super_block *sb);
  86 static void ext4_clear_request_list(void);
  87 static struct inode *ext4_get_journal_inode(struct super_block *sb,
  88                                             unsigned int journal_inum);
  89
  90 /*
  91  * Lock ordering
  92  *
  93  * Note the difference between i_mmap_sem (EXT4_I(inode)->i_mmap_sem) and
  94  * i_mmap_rwsem (inode->i_mmap_rwsem)!
  95  *
  96  * page fault path:
  97  * mmap_lock -> sb_start_pagefault -> i_mmap_sem (r) -> transaction start ->
  98  *   page lock -> i_data_sem (rw)
  99  *
 100  * buffered write path:
 101  * sb_start_write -> i_mutex -> mmap_lock
 102  * sb_start_write -> i_mutex -> transaction start -> page lock ->
 103  *   i_data_sem (rw)
 104  *
 105  * truncate:
 106  * sb_start_write -> i_mutex -> i_mmap_sem (w) -> i_mmap_rwsem (w) -> page lock
 107  * sb_start_write -> i_mutex -> i_mmap_sem (w) -> transaction start ->
 108  *   i_data_sem (rw)
 109  *
 110  * direct IO:
 111  * sb_start_write -> i_mutex -> mmap_lock
 112  * sb_start_write -> i_mutex -> transaction start -> i_data_sem (rw)
 113  *
 114  * writepages:
 115  * transaction start -> page lock(s) -> i_data_sem (rw)
 116  */
 117
 118 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
 119 static struct file_system_type ext2_fs_type = {
 120         .owner          = THIS_MODULE,
 121         .name           = "ext2",
 122         .mount          = ext4_mount,
 123         .kill_sb        = kill_block_super,
 124         .fs_flags       = FS_REQUIRES_DEV,
 125 };
 126 MODULE_ALIAS_FS("ext2");
 127 MODULE_ALIAS("ext2");
 128 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
 129 #else
 130 #define IS_EXT2_SB(sb) (0)
 131 #endif
 132
 133
 134 static struct file_system_type ext3_fs_type = {
 135         .owner          = THIS_MODULE,
 136         .name           = "ext3",
 137         .mount          = ext4_mount,
 138         .kill_sb        = kill_block_super,
 139         .fs_flags       = FS_REQUIRES_DEV,
 140 };
 141 MODULE_ALIAS_FS("ext3");
 142 MODULE_ALIAS("ext3");
 143 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 144
 145
 146 static inline void __ext4_read_bh(struct buffer_head *bh, int op_flags,
 147                                   bh_end_io_t *end_io)
 148 {
 149         /*
 150          * buffer's verified bit is no longer valid after reading from
 151          * disk again due to write out error, clear it to make sure we
 152          * recheck the buffer contents.
 153          */
 154         clear_buffer_verified(bh);
 155
 156         bh->b_end_io = end_io ? end_io : end_buffer_read_sync;
 157         get_bh(bh);
 158         submit_bh(REQ_OP_READ, op_flags, bh);
 159 }
 160
 161 void ext4_read_bh_nowait(struct buffer_head *bh, int op_flags,
 162                          bh_end_io_t *end_io)
 163 {
 164         BUG_ON(!buffer_locked(bh));
 165
 166         if (ext4_buffer_uptodate(bh)) {
 167                 unlock_buffer(bh);
 168                 return;
 169         }
 170         __ext4_read_bh(bh, op_flags, end_io);
 171 }
 172
 173 int ext4_read_bh(struct buffer_head *bh, int op_flags, bh_end_io_t *end_io)
 174 {
 175         BUG_ON(!buffer_locked(bh));
 176
 177         if (ext4_buffer_uptodate(bh)) {
 178                 unlock_buffer(bh);
 179                 return 0;
 180         }
 181
 182         __ext4_read_bh(bh, op_flags, end_io);
 183
 184         wait_on_buffer(bh);
 185         if (buffer_uptodate(bh))
 186                 return 0;
 187         return -EIO;
 188 }
 189
 190 int ext4_read_bh_lock(struct buffer_head *bh, int op_flags, bool wait)
 191 {
 192         if (trylock_buffer(bh)) {
 193                 if (wait)
 194                         return ext4_read_bh(bh, op_flags, NULL);
 195                 ext4_read_bh_nowait(bh, op_flags, NULL);
 196                 return 0;
 197         }
 198         if (wait) {
 199                 wait_on_buffer(bh);
 200                 if (buffer_uptodate(bh))
 201                         return 0;
 202                 return -EIO;
 203         }
 204         return 0;
 205 }
 206
 207 /*
 208  * This works like __bread_gfp() except it uses ERR_PTR for error
 209  * returns.  Currently with sb_bread it's impossible to distinguish
 210  * between ENOMEM and EIO situations (since both result in a NULL
 211  * return.
 212  */
 213 static struct buffer_head *__ext4_sb_bread_gfp(struct super_block *sb,
 214                                                sector_t block, int op_flags,
 215                                                gfp_t gfp)
 216 {
 217         struct buffer_head *bh;
 218         int ret;
 219
 220         bh = sb_getblk_gfp(sb, block, gfp);
 221         if (bh == NULL)
 222                 return ERR_PTR(-ENOMEM);
 223         if (ext4_buffer_uptodate(bh))
 224                 return bh;
 225
 226         ret = ext4_read_bh_lock(bh, REQ_META | op_flags, true);
 227         if (ret) {
 228                 put_bh(bh);
 229                 return ERR_PTR(ret);
 230         }
 231         return bh;
 232 }
 233
 234 struct buffer_head *ext4_sb_bread(struct super_block *sb, sector_t block,
 235                                    int op_flags)
 236 {
 237         return __ext4_sb_bread_gfp(sb, block, op_flags, __GFP_MOVABLE);
 238 }
 239
 240 struct buffer_head *ext4_sb_bread_unmovable(struct super_block *sb,
 241                                             sector_t block)
 242 {
 243         return __ext4_sb_bread_gfp(sb, block, 0, 0);
 244 }
 245
 246 void ext4_sb_breadahead_unmovable(struct super_block *sb, sector_t block)
 247 {
 248         struct buffer_head *bh = sb_getblk_gfp(sb, block, 0);
 249
 250         if (likely(bh)) {
 251                 ext4_read_bh_lock(bh, REQ_RAHEAD, false);
 252                 brelse(bh);
 253         }
 254 }
 255
 256 static int ext4_verify_csum_type(struct super_block *sb,
 257                                  struct ext4_super_block *es)
 258 {
 259         if (!ext4_has_feature_metadata_csum(sb))
 260                 return 1;
 261
 262         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 263 }
 264
 265 static __le32 ext4_superblock_csum(struct super_block *sb,
 266                                    struct ext4_super_block *es)
 267 {
 268         struct ext4_sb_info *sbi = EXT4_SB(sb);
 269         int offset = offsetof(struct ext4_super_block, s_checksum);
 270         __u32 csum;
 271
 272         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 273
 274         return cpu_to_le32(csum);
 275 }
 276
 277 static int ext4_superblock_csum_verify(struct super_block *sb,
 278                                        struct ext4_super_block *es)
 279 {
 280         if (!ext4_has_metadata_csum(sb))
 281                 return 1;
 282
 283         return es->s_checksum == ext4_superblock_csum(sb, es);
 284 }
 285
 286 void ext4_superblock_csum_set(struct super_block *sb)
 287 {
 288         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 289
 290         if (!ext4_has_metadata_csum(sb))
 291                 return;
 292
 293         es->s_checksum = ext4_superblock_csum(sb, es);
 294 }
 295
 296 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 297                                struct ext4_group_desc *bg)
 298 {
 299         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 300                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 301                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 302 }
 303
 304 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 305                                struct ext4_group_desc *bg)
 306 {
 307         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 308                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 309                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 310 }
 311
 312 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 313                               struct ext4_group_desc *bg)
 314 {
 315         return le32_to_cpu(bg->bg_inode_table_lo) |
 316                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 317                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 318 }
 319
 320 __u32 ext4_free_group_clusters(struct super_block *sb,
 321                                struct ext4_group_desc *bg)
 322 {
 323         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 324                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 325                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 326 }
 327
 328 __u32 ext4_free_inodes_count(struct super_block *sb,
 329                               struct ext4_group_desc *bg)
 330 {
 331         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 332                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 333                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 334 }
 335
 336 __u32 ext4_used_dirs_count(struct super_block *sb,
 337                               struct ext4_group_desc *bg)
 338 {
 339         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 340                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 341                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 342 }
 343
 344 __u32 ext4_itable_unused_count(struct super_block *sb,
 345                               struct ext4_group_desc *bg)
 346 {
 347         return le16_to_cpu(bg->bg_itable_unused_lo) |
 348                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 349                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 350 }
 351
 352 void ext4_block_bitmap_set(struct super_block *sb,
 353                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 354 {
 355         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 356         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 357                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 358 }
 359
 360 void ext4_inode_bitmap_set(struct super_block *sb,
 361                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 362 {
 363         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 364         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 365                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 366 }
 367
 368 void ext4_inode_table_set(struct super_block *sb,
 369                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 370 {
 371         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 372         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 373                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 374 }
 375
 376 void ext4_free_group_clusters_set(struct super_block *sb,
 377                                   struct ext4_group_desc *bg, __u32 count)
 378 {
 379         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 380         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 381                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 382 }
 383
 384 void ext4_free_inodes_set(struct super_block *sb,
 385                           struct ext4_group_desc *bg, __u32 count)
 386 {
 387         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 388         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 389                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 390 }
 391
 392 void ext4_used_dirs_set(struct super_block *sb,
 393                           struct ext4_group_desc *bg, __u32 count)
 394 {
 395         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 396         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 397                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 398 }
 399
 400 void ext4_itable_unused_set(struct super_block *sb,
 401                           struct ext4_group_desc *bg, __u32 count)
 402 {
 403         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 404         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 405                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 406 }
 407
 408 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi, time64_t now)
 409 {
 410         now = clamp_val(now, 0, (1ull << 40) - 1);
 411
 412         *lo = cpu_to_le32(lower_32_bits(now));
 413         *hi = upper_32_bits(now);
 414 }
 415
 416 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi)
 417 {
 418         return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo);
 419 }
 420 #define ext4_update_tstamp(es, tstamp) \
 421         __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi, \
 422                              ktime_get_real_seconds())
 423 #define ext4_get_tstamp(es, tstamp) \
 424         __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
 425
 426 /*
 427  * The del_gendisk() function uninitializes the disk-specific data
 428  * structures, including the bdi structure, without telling anyone
 429  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 430  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 431  * This is a kludge to prevent these oops until we can put in a proper
 432  * hook in del_gendisk() to inform the VFS and file system layers.
 433  */
 434 static int block_device_ejected(struct super_block *sb)
 435 {
 436         struct inode *bd_inode = sb->s_bdev->bd_inode;
 437         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 438
 439         return bdi->dev == NULL;
 440 }
 441
 442 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 443 {
 444         struct super_block              *sb = journal->j_private;
 445         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 446         int                             error = is_journal_aborted(journal);
 447         struct ext4_journal_cb_entry    *jce;
 448
 449         BUG_ON(txn->t_state == T_FINISHED);
 450
 451         ext4_process_freed_data(sb, txn->t_tid);
 452
 453         spin_lock(&sbi->s_md_lock);
 454         while (!list_empty(&txn->t_private_list)) {
 455                 jce = list_entry(txn->t_private_list.next,
 456                                  struct ext4_journal_cb_entry, jce_list);
 457                 list_del_init(&jce->jce_list);
 458                 spin_unlock(&sbi->s_md_lock);
 459                 jce->jce_func(sb, jce, error);
 460                 spin_lock(&sbi->s_md_lock);
 461         }
 462         spin_unlock(&sbi->s_md_lock);
 463 }
 464
 465 /*
 466  * This writepage callback for write_cache_pages()
 467  * takes care of a few cases after page cleaning.
 468  *
 469  * write_cache_pages() already checks for dirty pages
 470  * and calls clear_page_dirty_for_io(), which we want,
 471  * to write protect the pages.
 472  *
 473  * However, we may have to redirty a page (see below.)
 474  */
 475 static int ext4_journalled_writepage_callback(struct page *page,
 476                                               struct writeback_control *wbc,
 477                                               void *data)
 478 {
 479         transaction_t *transaction = (transaction_t *) data;
 480         struct buffer_head *bh, *head;
 481         struct journal_head *jh;
 482
 483         bh = head = page_buffers(page);
 484         do {
 485                 /*
 486                  * We have to redirty a page in these cases:
 487                  * 1) If buffer is dirty, it means the page was dirty because it
 488                  * contains a buffer that needs checkpointing. So the dirty bit
 489                  * needs to be preserved so that checkpointing writes the buffer
 490                  * properly.
 491                  * 2) If buffer is not part of the committing transaction
 492                  * (we may have just accidentally come across this buffer because
 493                  * inode range tracking is not exact) or if the currently running
 494                  * transaction already contains this buffer as well, dirty bit
 495                  * needs to be preserved so that the buffer gets writeprotected
 496                  * properly on running transaction's commit.
 497                  */
 498                 jh = bh2jh(bh);
 499                 if (buffer_dirty(bh) ||
 500                     (jh && (jh->b_transaction != transaction ||
 501                             jh->b_next_transaction))) {
 502                         redirty_page_for_writepage(wbc, page);
 503                         goto out;
 504                 }
 505         } while ((bh = bh->b_this_page) != head);
 506
 507 out:
 508         return AOP_WRITEPAGE_ACTIVATE;
 509 }
 510
 511 static int ext4_journalled_submit_inode_data_buffers(struct jbd2_inode *jinode)
 512 {
 513         struct address_space *mapping = jinode->i_vfs_inode->i_mapping;
 514         struct writeback_control wbc = {
 515                 .sync_mode =  WB_SYNC_ALL,
 516                 .nr_to_write = LONG_MAX,
 517                 .range_start = jinode->i_dirty_start,
 518                 .range_end = jinode->i_dirty_end,
 519         };
 520
 521         return write_cache_pages(mapping, &wbc,
 522                                  ext4_journalled_writepage_callback,
 523                                  jinode->i_transaction);
 524 }
 525
 526 static int ext4_journal_submit_inode_data_buffers(struct jbd2_inode *jinode)
 527 {
 528         int ret;
 529
 530         if (ext4_should_journal_data(jinode->i_vfs_inode))
 531                 ret = ext4_journalled_submit_inode_data_buffers(jinode);
 532         else
 533                 ret = jbd2_journal_submit_inode_data_buffers(jinode);
 534
 535         return ret;
 536 }
 537
 538 static int ext4_journal_finish_inode_data_buffers(struct jbd2_inode *jinode)
 539 {
 540         int ret = 0;
 541
 542         if (!ext4_should_journal_data(jinode->i_vfs_inode))
 543                 ret = jbd2_journal_finish_inode_data_buffers(jinode);
 544
 545         return ret;
 546 }
 547
 548 static bool system_going_down(void)
 549 {
 550         return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF
 551                 || system_state == SYSTEM_RESTART;
 552 }
 553
 554 struct ext4_err_translation {
 555         int code;
 556         int errno;
 557 };
 558
 559 #define EXT4_ERR_TRANSLATE(err) { .code = EXT4_ERR_##err, .errno = err }
 560
 561 static struct ext4_err_translation err_translation[] = {
 562         EXT4_ERR_TRANSLATE(EIO),
 563         EXT4_ERR_TRANSLATE(ENOMEM),
 564         EXT4_ERR_TRANSLATE(EFSBADCRC),
 565         EXT4_ERR_TRANSLATE(EFSCORRUPTED),
 566         EXT4_ERR_TRANSLATE(ENOSPC),
 567         EXT4_ERR_TRANSLATE(ENOKEY),
 568         EXT4_ERR_TRANSLATE(EROFS),
 569         EXT4_ERR_TRANSLATE(EFBIG),
 570         EXT4_ERR_TRANSLATE(EEXIST),
 571         EXT4_ERR_TRANSLATE(ERANGE),
 572         EXT4_ERR_TRANSLATE(EOVERFLOW),
 573         EXT4_ERR_TRANSLATE(EBUSY),
 574         EXT4_ERR_TRANSLATE(ENOTDIR),
 575         EXT4_ERR_TRANSLATE(ENOTEMPTY),
 576         EXT4_ERR_TRANSLATE(ESHUTDOWN),
 577         EXT4_ERR_TRANSLATE(EFAULT),
 578 };
 579
 580 static int ext4_errno_to_code(int errno)
 581 {
 582         int i;
 583
 584         for (i = 0; i < ARRAY_SIZE(err_translation); i++)
 585                 if (err_translation[i].errno == errno)
 586                         return err_translation[i].code;
 587         return EXT4_ERR_UNKNOWN;
 588 }
 589
 590 static void save_error_info(struct super_block *sb, int error,
 591                             __u32 ino, __u64 block,
 592                             const char *func, unsigned int line)
 593 {
 594         struct ext4_sb_info *sbi = EXT4_SB(sb);
 595
 596         /* We default to EFSCORRUPTED error... */
 597         if (error == 0)
 598                 error = EFSCORRUPTED;
 599
 600         spin_lock(&sbi->s_error_lock);
 601         sbi->s_add_error_count++;
 602         sbi->s_last_error_code = error;
 603         sbi->s_last_error_line = line;
 604         sbi->s_last_error_ino = ino;
 605         sbi->s_last_error_block = block;
 606         sbi->s_last_error_func = func;
 607         sbi->s_last_error_time = ktime_get_real_seconds();
 608         if (!sbi->s_first_error_time) {
 609                 sbi->s_first_error_code = error;
 610                 sbi->s_first_error_line = line;
 611                 sbi->s_first_error_ino = ino;
 612                 sbi->s_first_error_block = block;
 613                 sbi->s_first_error_func = func;
 614                 sbi->s_first_error_time = sbi->s_last_error_time;
 615         }
 616         spin_unlock(&sbi->s_error_lock);
 617 }
 618
 619 /* Deal with the reporting of failure conditions on a filesystem such as
 620  * inconsistencies detected or read IO failures.
 621  *
 622  * On ext2, we can store the error state of the filesystem in the
 623  * superblock.  That is not possible on ext4, because we may have other
 624  * write ordering constraints on the superblock which prevent us from
 625  * writing it out straight away; and given that the journal is about to
 626  * be aborted, we can't rely on the current, or future, transactions to
 627  * write out the superblock safely.
 628  *
 629  * We'll just use the jbd2_journal_abort() error code to record an error in
 630  * the journal instead.  On recovery, the journal will complain about
 631  * that error until we've noted it down and cleared it.
 632  *
 633  * If force_ro is set, we unconditionally force the filesystem into an
 634  * ABORT|READONLY state, unless the error response on the fs has been set to
 635  * panic in which case we take the easy way out and panic immediately. This is
 636  * used to deal with unrecoverable failures such as journal IO errors or ENOMEM
 637  * at a critical moment in log management.
 638  */
 639 static void ext4_handle_error(struct super_block *sb, bool force_ro, int error,
 640                               __u32 ino, __u64 block,
 641                               const char *func, unsigned int line)
 642 {
 643         journal_t *journal = EXT4_SB(sb)->s_journal;
 644         bool continue_fs = !force_ro && test_opt(sb, ERRORS_CONT);
 645
 646         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 647         if (test_opt(sb, WARN_ON_ERROR))
 648                 WARN_ON_ONCE(1);
 649
 650         if (!continue_fs && !sb_rdonly(sb)) {
 651                 ext4_set_mount_flag(sb, EXT4_MF_FS_ABORTED);
 652                 if (journal)
 653                         jbd2_journal_abort(journal, -EIO);
 654         }
 655
 656         if (!bdev_read_only(sb->s_bdev)) {
 657                 save_error_info(sb, error, ino, block, func, line);
 658                 /*
 659                  * In case the fs should keep running, we need to writeout
 660                  * superblock through the journal. Due to lock ordering
 661                  * constraints, it may not be safe to do it right here so we
 662                  * defer superblock flushing to a workqueue.
 663                  */
 664                 if (continue_fs)
 665                         schedule_work(&EXT4_SB(sb)->s_error_work);
 666                 else
 667                         ext4_commit_super(sb);
 668         }
 669
 670         /*
 671          * We force ERRORS_RO behavior when system is rebooting. Otherwise we
 672          * could panic during 'reboot -f' as the underlying device got already
 673          * disabled.
 674          */
 675         if (test_opt(sb, ERRORS_PANIC) && !system_going_down()) {
 676                 panic("EXT4-fs (device %s): panic forced after error\n",
 677                         sb->s_id);
 678         }
 679
 680         if (sb_rdonly(sb) || continue_fs)
 681                 return;
 682
 683         ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 684         /*
 685          * Make sure updated value of ->s_mount_flags will be visible before
 686          * ->s_flags update
 687          */
 688         smp_wmb();
 689         sb->s_flags |= SB_RDONLY;
 690 }
 691
 692 static void flush_stashed_error_work(struct work_struct *work)
 693 {
 694         struct ext4_sb_info *sbi = container_of(work, struct ext4_sb_info,
 695                                                 s_error_work);
 696         journal_t *journal = sbi->s_journal;
 697         handle_t *handle;
 698
 699         /*
 700          * If the journal is still running, we have to write out superblock
 701          * through the journal to avoid collisions of other journalled sb
 702          * updates.
 703          *
 704          * We use directly jbd2 functions here to avoid recursing back into
 705          * ext4 error handling code during handling of previous errors.
 706          */
 707         if (!sb_rdonly(sbi->s_sb) && journal) {
 708                 struct buffer_head *sbh = sbi->s_sbh;
 709                 handle = jbd2_journal_start(journal, 1);
 710                 if (IS_ERR(handle))
 711                         goto write_directly;
 712                 if (jbd2_journal_get_write_access(handle, sbh)) {
 713                         jbd2_journal_stop(handle);
 714                         goto write_directly;
 715                 }
 716                 ext4_update_super(sbi->s_sb);
 717                 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
 718                         ext4_msg(sbi->s_sb, KERN_ERR, "previous I/O error to "
 719                                  "superblock detected");
 720                         clear_buffer_write_io_error(sbh);
 721                         set_buffer_uptodate(sbh);
 722                 }
 723
 724                 if (jbd2_journal_dirty_metadata(handle, sbh)) {
 725                         jbd2_journal_stop(handle);
 726                         goto write_directly;
 727                 }
 728                 jbd2_journal_stop(handle);
 729                 ext4_notify_error_sysfs(sbi);
 730                 return;
 731         }
 732 write_directly:
 733         /*
 734          * Write through journal failed. Write sb directly to get error info
 735          * out and hope for the best.
 736          */
 737         ext4_commit_super(sbi->s_sb);
 738         ext4_notify_error_sysfs(sbi);
 739 }
 740
 741 #define ext4_error_ratelimit(sb)                                        \
 742                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 743                              "EXT4-fs error")
 744
 745 void __ext4_error(struct super_block *sb, const char *function,
 746                   unsigned int line, bool force_ro, int error, __u64 block,
 747                   const char *fmt, ...)
 748 {
 749         struct va_format vaf;
 750         va_list args;
 751
 752         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 753                 return;
 754
 755         trace_ext4_error(sb, function, line);
 756         if (ext4_error_ratelimit(sb)) {
 757                 va_start(args, fmt);
 758                 vaf.fmt = fmt;
 759                 vaf.va = &args;
 760                 printk(KERN_CRIT
 761                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 762                        sb->s_id, function, line, current->comm, &vaf);
 763                 va_end(args);
 764         }
 765         ext4_handle_error(sb, force_ro, error, 0, block, function, line);
 766 }
 767
 768 void __ext4_error_inode(struct inode *inode, const char *function,
 769                         unsigned int line, ext4_fsblk_t block, int error,
 770                         const char *fmt, ...)
 771 {
 772         va_list args;
 773         struct va_format vaf;
 774
 775         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 776                 return;
 777
 778         trace_ext4_error(inode->i_sb, function, line);
 779         if (ext4_error_ratelimit(inode->i_sb)) {
 780                 va_start(args, fmt);
 781                 vaf.fmt = fmt;
 782                 vaf.va = &args;
 783                 if (block)
 784                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 785                                "inode #%lu: block %llu: comm %s: %pV\n",
 786                                inode->i_sb->s_id, function, line, inode->i_ino,
 787                                block, current->comm, &vaf);
 788                 else
 789                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 790                                "inode #%lu: comm %s: %pV\n",
 791                                inode->i_sb->s_id, function, line, inode->i_ino,
 792                                current->comm, &vaf);
 793                 va_end(args);
 794         }
 795         ext4_handle_error(inode->i_sb, false, error, inode->i_ino, block,
 796                           function, line);
 797 }
 798
 799 void __ext4_error_file(struct file *file, const char *function,
 800                        unsigned int line, ext4_fsblk_t block,
 801                        const char *fmt, ...)
 802 {
 803         va_list args;
 804         struct va_format vaf;
 805         struct inode *inode = file_inode(file);
 806         char pathname[80], *path;
 807
 808         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 809                 return;
 810
 811         trace_ext4_error(inode->i_sb, function, line);
 812         if (ext4_error_ratelimit(inode->i_sb)) {
 813                 path = file_path(file, pathname, sizeof(pathname));
 814                 if (IS_ERR(path))
 815                         path = "(unknown)";
 816                 va_start(args, fmt);
 817                 vaf.fmt = fmt;
 818                 vaf.va = &args;
 819                 if (block)
 820                         printk(KERN_CRIT
 821                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 822                                "block %llu: comm %s: path %s: %pV\n",
 823                                inode->i_sb->s_id, function, line, inode->i_ino,
 824                                block, current->comm, path, &vaf);
 825                 else
 826                         printk(KERN_CRIT
 827                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 828                                "comm %s: path %s: %pV\n",
 829                                inode->i_sb->s_id, function, line, inode->i_ino,
 830                                current->comm, path, &vaf);
 831                 va_end(args);
 832         }
 833         ext4_handle_error(inode->i_sb, false, EFSCORRUPTED, inode->i_ino, block,
 834                           function, line);
 835 }
 836
 837 const char *ext4_decode_error(struct super_block *sb, int errno,
 838                               char nbuf[16])
 839 {
 840         char *errstr = NULL;
 841
 842         switch (errno) {
 843         case -EFSCORRUPTED:
 844                 errstr = "Corrupt filesystem";
 845                 break;
 846         case -EFSBADCRC:
 847                 errstr = "Filesystem failed CRC";
 848                 break;
 849         case -EIO:
 850                 errstr = "IO failure";
 851                 break;
 852         case -ENOMEM:
 853                 errstr = "Out of memory";
 854                 break;
 855         case -EROFS:
 856                 if (!sb || (EXT4_SB(sb)->s_journal &&
 857                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 858                         errstr = "Journal has aborted";
 859                 else
 860                         errstr = "Readonly filesystem";
 861                 break;
 862         default:
 863                 /* If the caller passed in an extra buffer for unknown
 864                  * errors, textualise them now.  Else we just return
 865                  * NULL. */
 866                 if (nbuf) {
 867                         /* Check for truncated error codes... */
 868                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 869                                 errstr = nbuf;
 870                 }
 871                 break;
 872         }
 873
 874         return errstr;
 875 }
 876
 877 /* __ext4_std_error decodes expected errors from journaling functions
 878  * automatically and invokes the appropriate error response.  */
 879
 880 void __ext4_std_error(struct super_block *sb, const char *function,
 881                       unsigned int line, int errno)
 882 {
 883         char nbuf[16];
 884         const char *errstr;
 885
 886         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 887                 return;
 888
 889         /* Special case: if the error is EROFS, and we're not already
 890          * inside a transaction, then there's really no point in logging
 891          * an error. */
 892         if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb))
 893                 return;
 894
 895         if (ext4_error_ratelimit(sb)) {
 896                 errstr = ext4_decode_error(sb, errno, nbuf);
 897                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 898                        sb->s_id, function, line, errstr);
 899         }
 900
 901         ext4_handle_error(sb, false, -errno, 0, 0, function, line);
 902 }
 903
 904 void __ext4_msg(struct super_block *sb,
 905                 const char *prefix, const char *fmt, ...)
 906 {
 907         struct va_format vaf;
 908         va_list args;
 909
 910         atomic_inc(&EXT4_SB(sb)->s_msg_count);
 911         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 912                 return;
 913
 914         va_start(args, fmt);
 915         vaf.fmt = fmt;
 916         vaf.va = &args;
 917         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 918         va_end(args);
 919 }
 920
 921 static int ext4_warning_ratelimit(struct super_block *sb)
 922 {
 923         atomic_inc(&EXT4_SB(sb)->s_warning_count);
 924         return ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state),
 925                             "EXT4-fs warning");
 926 }
 927
 928 void __ext4_warning(struct super_block *sb, const char *function,
 929                     unsigned int line, const char *fmt, ...)
 930 {
 931         struct va_format vaf;
 932         va_list args;
 933
 934         if (!ext4_warning_ratelimit(sb))
 935                 return;
 936
 937         va_start(args, fmt);
 938         vaf.fmt = fmt;
 939         vaf.va = &args;
 940         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 941                sb->s_id, function, line, &vaf);
 942         va_end(args);
 943 }
 944
 945 void __ext4_warning_inode(const struct inode *inode, const char *function,
 946                           unsigned int line, const char *fmt, ...)
 947 {
 948         struct va_format vaf;
 949         va_list args;
 950
 951         if (!ext4_warning_ratelimit(inode->i_sb))
 952                 return;
 953
 954         va_start(args, fmt);
 955         vaf.fmt = fmt;
 956         vaf.va = &args;
 957         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 958                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 959                function, line, inode->i_ino, current->comm, &vaf);
 960         va_end(args);
 961 }
 962
 963 void __ext4_grp_locked_error(const char *function, unsigned int line,
 964                              struct super_block *sb, ext4_group_t grp,
 965                              unsigned long ino, ext4_fsblk_t block,
 966                              const char *fmt, ...)
 967 __releases(bitlock)
 968 __acquires(bitlock)
 969 {
 970         struct va_format vaf;
 971         va_list args;
 972
 973         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 974                 return;
 975
 976         trace_ext4_error(sb, function, line);
 977         if (ext4_error_ratelimit(sb)) {
 978                 va_start(args, fmt);
 979                 vaf.fmt = fmt;
 980                 vaf.va = &args;
 981                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 982                        sb->s_id, function, line, grp);
 983                 if (ino)
 984                         printk(KERN_CONT "inode %lu: ", ino);
 985                 if (block)
 986                         printk(KERN_CONT "block %llu:",
 987                                (unsigned long long) block);
 988                 printk(KERN_CONT "%pV\n", &vaf);
 989                 va_end(args);
 990         }
 991
 992         if (test_opt(sb, ERRORS_CONT)) {
 993                 if (test_opt(sb, WARN_ON_ERROR))
 994                         WARN_ON_ONCE(1);
 995                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 996                 if (!bdev_read_only(sb->s_bdev)) {
 997                         save_error_info(sb, EFSCORRUPTED, ino, block, function,
 998                                         line);
 999                         schedule_work(&EXT4_SB(sb)->s_error_work);
1000                 }
1001                 return;
1002         }
1003         ext4_unlock_group(sb, grp);
1004         ext4_handle_error(sb, false, EFSCORRUPTED, ino, block, function, line);
1005         /*
1006          * We only get here in the ERRORS_RO case; relocking the group
1007          * may be dangerous, but nothing bad will happen since the
1008          * filesystem will have already been marked read/only and the
1009          * journal has been aborted.  We return 1 as a hint to callers
1010          * who might what to use the return value from
1011          * ext4_grp_locked_error() to distinguish between the
1012          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
1013          * aggressively from the ext4 function in question, with a
1014          * more appropriate error code.
1015          */
1016         ext4_lock_group(sb, grp);
1017         return;
1018 }
1019
1020 void ext4_mark_group_bitmap_corrupted(struct super_block *sb,
1021                                      ext4_group_t group,
1022                                      unsigned int flags)
1023 {
1024         struct ext4_sb_info *sbi = EXT4_SB(sb);
1025         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1026         struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
1027         int ret;
1028
1029         if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) {
1030                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT,
1031                                             &grp->bb_state);
1032                 if (!ret)
1033                         percpu_counter_sub(&sbi->s_freeclusters_counter,
1034                                            grp->bb_free);
1035         }
1036
1037         if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) {
1038                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT,
1039                                             &grp->bb_state);
1040                 if (!ret && gdp) {
1041                         int count;
1042
1043                         count = ext4_free_inodes_count(sb, gdp);
1044                         percpu_counter_sub(&sbi->s_freeinodes_counter,
1045                                            count);
1046                 }
1047         }
1048 }
1049
1050 void ext4_update_dynamic_rev(struct super_block *sb)
1051 {
1052         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1053
1054         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
1055                 return;
1056
1057         ext4_warning(sb,
1058                      "updating to rev %d because of new feature flag, "
1059                      "running e2fsck is recommended",
1060                      EXT4_DYNAMIC_REV);
1061
1062         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
1063         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
1064         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
1065         /* leave es->s_feature_*compat flags alone */
1066         /* es->s_uuid will be set by e2fsck if empty */
1067
1068         /*
1069          * The rest of the superblock fields should be zero, and if not it
1070          * means they are likely already in use, so leave them alone.  We
1071          * can leave it up to e2fsck to clean up any inconsistencies there.
1072          */
1073 }
1074
1075 /*
1076  * Open the external journal device
1077  */
1078 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
1079 {
1080         struct block_device *bdev;
1081
1082         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
1083         if (IS_ERR(bdev))
1084                 goto fail;
1085         return bdev;
1086
1087 fail:
1088         ext4_msg(sb, KERN_ERR,
1089                  "failed to open journal device unknown-block(%u,%u) %ld",
1090                  MAJOR(dev), MINOR(dev), PTR_ERR(bdev));
1091         return NULL;
1092 }
1093
1094 /*
1095  * Release the journal device
1096  */
1097 static void ext4_blkdev_put(struct block_device *bdev)
1098 {
1099         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1100 }
1101
1102 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
1103 {
1104         struct block_device *bdev;
1105         bdev = sbi->s_journal_bdev;
1106         if (bdev) {
1107                 ext4_blkdev_put(bdev);
1108                 sbi->s_journal_bdev = NULL;
1109         }
1110 }
1111
1112 static inline struct inode *orphan_list_entry(struct list_head *l)
1113 {
1114         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
1115 }
1116
1117 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
1118 {
1119         struct list_head *l;
1120
1121         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
1122                  le32_to_cpu(sbi->s_es->s_last_orphan));
1123
1124         printk(KERN_ERR "sb_info orphan list:\n");
1125         list_for_each(l, &sbi->s_orphan) {
1126                 struct inode *inode = orphan_list_entry(l);
1127                 printk(KERN_ERR "  "
1128                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
1129                        inode->i_sb->s_id, inode->i_ino, inode,
1130                        inode->i_mode, inode->i_nlink,
1131                        NEXT_ORPHAN(inode));
1132         }
1133 }
1134
1135 #ifdef CONFIG_QUOTA
1136 static int ext4_quota_off(struct super_block *sb, int type);
1137
1138 static inline void ext4_quota_off_umount(struct super_block *sb)
1139 {
1140         int type;
1141
1142         /* Use our quota_off function to clear inode flags etc. */
1143         for (type = 0; type < EXT4_MAXQUOTAS; type++)
1144                 ext4_quota_off(sb, type);
1145 }
1146
1147 /*
1148  * This is a helper function which is used in the mount/remount
1149  * codepaths (which holds s_umount) to fetch the quota file name.
1150  */
1151 static inline char *get_qf_name(struct super_block *sb,
1152                                 struct ext4_sb_info *sbi,
1153                                 int type)
1154 {
1155         return rcu_dereference_protected(sbi->s_qf_names[type],
1156                                          lockdep_is_held(&sb->s_umount));
1157 }
1158 #else
1159 static inline void ext4_quota_off_umount(struct super_block *sb)
1160 {
1161 }
1162 #endif
1163
1164 static void ext4_put_super(struct super_block *sb)
1165 {
1166         struct ext4_sb_info *sbi = EXT4_SB(sb);
1167         struct ext4_super_block *es = sbi->s_es;
1168         struct buffer_head **group_desc;
1169         struct flex_groups **flex_groups;
1170         int aborted = 0;
1171         int i, err;
1172
1173         ext4_unregister_li_request(sb);
1174         ext4_quota_off_umount(sb);
1175
1176         flush_work(&sbi->s_error_work);
1177         destroy_workqueue(sbi->rsv_conversion_wq);
1178
1179         /*
1180          * Unregister sysfs before destroying jbd2 journal.
1181          * Since we could still access attr_journal_task attribute via sysfs
1182          * path which could have sbi->s_journal->j_task as NULL
1183          */
1184         ext4_unregister_sysfs(sb);
1185
1186         if (sbi->s_journal) {
1187                 aborted = is_journal_aborted(sbi->s_journal);
1188                 err = jbd2_journal_destroy(sbi->s_journal);
1189                 sbi->s_journal = NULL;
1190                 if ((err < 0) && !aborted) {
1191                         ext4_abort(sb, -err, "Couldn't clean up the journal");
1192                 }
1193         }
1194
1195         ext4_es_unregister_shrinker(sbi);
1196         del_timer_sync(&sbi->s_err_report);
1197         ext4_release_system_zone(sb);
1198         ext4_mb_release(sb);
1199         ext4_ext_release(sb);
1200
1201         if (!sb_rdonly(sb) && !aborted) {
1202                 ext4_clear_feature_journal_needs_recovery(sb);
1203                 es->s_state = cpu_to_le16(sbi->s_mount_state);
1204         }
1205         if (!sb_rdonly(sb))
1206                 ext4_commit_super(sb);
1207
1208         rcu_read_lock();
1209         group_desc = rcu_dereference(sbi->s_group_desc);
1210         for (i = 0; i < sbi->s_gdb_count; i++)
1211                 brelse(group_desc[i]);
1212         kvfree(group_desc);
1213         flex_groups = rcu_dereference(sbi->s_flex_groups);
1214         if (flex_groups) {
1215                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
1216                         kvfree(flex_groups[i]);
1217                 kvfree(flex_groups);
1218         }
1219         rcu_read_unlock();
1220         percpu_counter_destroy(&sbi->s_freeclusters_counter);
1221         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1222         percpu_counter_destroy(&sbi->s_dirs_counter);
1223         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
1224         percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit);
1225         percpu_free_rwsem(&sbi->s_writepages_rwsem);
1226 #ifdef CONFIG_QUOTA
1227         for (i = 0; i < EXT4_MAXQUOTAS; i++)
1228                 kfree(get_qf_name(sb, sbi, i));
1229 #endif
1230
1231         /* Debugging code just in case the in-memory inode orphan list
1232          * isn't empty.  The on-disk one can be non-empty if we've
1233          * detected an error and taken the fs readonly, but the
1234          * in-memory list had better be clean by this point. */
1235         if (!list_empty(&sbi->s_orphan))
1236                 dump_orphan_list(sb, sbi);
1237         ASSERT(list_empty(&sbi->s_orphan));
1238
1239         sync_blockdev(sb->s_bdev);
1240         invalidate_bdev(sb->s_bdev);
1241         if (sbi->s_journal_bdev && sbi->s_journal_bdev != sb->s_bdev) {
1242                 /*
1243                  * Invalidate the journal device's buffers.  We don't want them
1244                  * floating about in memory - the physical journal device may
1245                  * hotswapped, and it breaks the `ro-after' testing code.
1246                  */
1247                 sync_blockdev(sbi->s_journal_bdev);
1248                 invalidate_bdev(sbi->s_journal_bdev);
1249                 ext4_blkdev_remove(sbi);
1250         }
1251
1252         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
1253         sbi->s_ea_inode_cache = NULL;
1254
1255         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
1256         sbi->s_ea_block_cache = NULL;
1257
1258         ext4_stop_mmpd(sbi);
1259
1260         brelse(sbi->s_sbh);
1261         sb->s_fs_info = NULL;
1262         /*
1263          * Now that we are completely done shutting down the
1264          * superblock, we need to actually destroy the kobject.
1265          */
1266         kobject_put(&sbi->s_kobj);
1267         wait_for_completion(&sbi->s_kobj_unregister);
1268         if (sbi->s_chksum_driver)
1269                 crypto_free_shash(sbi->s_chksum_driver);
1270         kfree(sbi->s_blockgroup_lock);
1271         fs_put_dax(sbi->s_daxdev);
1272         fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy);
1273 #ifdef CONFIG_UNICODE
1274         utf8_unload(sb->s_encoding);
1275 #endif
1276         kfree(sbi);
1277 }
1278
1279 static struct kmem_cache *ext4_inode_cachep;
1280
1281 /*
1282  * Called inside transaction, so use GFP_NOFS
1283  */
1284 static struct inode *ext4_alloc_inode(struct super_block *sb)
1285 {
1286         struct ext4_inode_info *ei;
1287
1288         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
1289         if (!ei)
1290                 return NULL;
1291
1292         inode_set_iversion(&ei->vfs_inode, 1);
1293         spin_lock_init(&ei->i_raw_lock);
1294         INIT_LIST_HEAD(&ei->i_prealloc_list);
1295         atomic_set(&ei->i_prealloc_active, 0);
1296         spin_lock_init(&ei->i_prealloc_lock);
1297         ext4_es_init_tree(&ei->i_es_tree);
1298         rwlock_init(&ei->i_es_lock);
1299         INIT_LIST_HEAD(&ei->i_es_list);
1300         ei->i_es_all_nr = 0;
1301         ei->i_es_shk_nr = 0;
1302         ei->i_es_shrink_lblk = 0;
1303         ei->i_reserved_data_blocks = 0;
1304         spin_lock_init(&(ei->i_block_reservation_lock));
1305         ext4_init_pending_tree(&ei->i_pending_tree);
1306 #ifdef CONFIG_QUOTA
1307         ei->i_reserved_quota = 0;
1308         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
1309 #endif
1310         ei->jinode = NULL;
1311         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
1312         spin_lock_init(&ei->i_completed_io_lock);
1313         ei->i_sync_tid = 0;
1314         ei->i_datasync_tid = 0;
1315         atomic_set(&ei->i_unwritten, 0);
1316         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
1317         ext4_fc_init_inode(&ei->vfs_inode);
1318         mutex_init(&ei->i_fc_lock);
1319         return &ei->vfs_inode;
1320 }
1321
1322 static int ext4_drop_inode(struct inode *inode)
1323 {
1324         int drop = generic_drop_inode(inode);
1325
1326         if (!drop)
1327                 drop = fscrypt_drop_inode(inode);
1328
1329         trace_ext4_drop_inode(inode, drop);
1330         return drop;
1331 }
1332
1333 static void ext4_free_in_core_inode(struct inode *inode)
1334 {
1335         fscrypt_free_inode(inode);
1336         if (!list_empty(&(EXT4_I(inode)->i_fc_list))) {
1337                 pr_warn("%s: inode %ld still in fc list",
1338                         __func__, inode->i_ino);
1339         }
1340         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1341 }
1342
1343 static void ext4_destroy_inode(struct inode *inode)
1344 {
1345         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1346                 ext4_msg(inode->i_sb, KERN_ERR,
1347                          "Inode %lu (%p): orphan list check failed!",
1348                          inode->i_ino, EXT4_I(inode));
1349                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1350                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1351                                 true);
1352                 dump_stack();
1353         }
1354 }
1355
1356 static void init_once(void *foo)
1357 {
1358         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1359
1360         INIT_LIST_HEAD(&ei->i_orphan);
1361         init_rwsem(&ei->xattr_sem);
1362         init_rwsem(&ei->i_data_sem);
1363         init_rwsem(&ei->i_mmap_sem);
1364         inode_init_once(&ei->vfs_inode);
1365         ext4_fc_init_inode(&ei->vfs_inode);
1366 }
1367
1368 static int __init init_inodecache(void)
1369 {
1370         ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache",
1371                                 sizeof(struct ext4_inode_info), 0,
1372                                 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
1373                                         SLAB_ACCOUNT),
1374                                 offsetof(struct ext4_inode_info, i_data),
1375                                 sizeof_field(struct ext4_inode_info, i_data),
1376                                 init_once);
1377         if (ext4_inode_cachep == NULL)
1378                 return -ENOMEM;
1379         return 0;
1380 }
1381
1382 static void destroy_inodecache(void)
1383 {
1384         /*
1385          * Make sure all delayed rcu free inodes are flushed before we
1386          * destroy cache.
1387          */
1388         rcu_barrier();
1389         kmem_cache_destroy(ext4_inode_cachep);
1390 }
1391
1392 void ext4_clear_inode(struct inode *inode)
1393 {
1394         ext4_fc_del(inode);
1395         invalidate_inode_buffers(inode);
1396         clear_inode(inode);
1397         ext4_discard_preallocations(inode, 0);
1398         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1399         dquot_drop(inode);
1400         if (EXT4_I(inode)->jinode) {
1401                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1402                                                EXT4_I(inode)->jinode);
1403                 jbd2_free_inode(EXT4_I(inode)->jinode);
1404                 EXT4_I(inode)->jinode = NULL;
1405         }
1406         fscrypt_put_encryption_info(inode);
1407         fsverity_cleanup_inode(inode);
1408 }
1409
1410 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1411                                         u64 ino, u32 generation)
1412 {
1413         struct inode *inode;
1414
1415         /*
1416          * Currently we don't know the generation for parent directory, so
1417          * a generation of 0 means "accept any"
1418          */
1419         inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE);
1420         if (IS_ERR(inode))
1421                 return ERR_CAST(inode);
1422         if (generation && inode->i_generation != generation) {
1423                 iput(inode);
1424                 return ERR_PTR(-ESTALE);
1425         }
1426
1427         return inode;
1428 }
1429
1430 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1431                                         int fh_len, int fh_type)
1432 {
1433         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1434                                     ext4_nfs_get_inode);
1435 }
1436
1437 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1438                                         int fh_len, int fh_type)
1439 {
1440         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1441                                     ext4_nfs_get_inode);
1442 }
1443
1444 static int ext4_nfs_commit_metadata(struct inode *inode)
1445 {
1446         struct writeback_control wbc = {
1447                 .sync_mode = WB_SYNC_ALL
1448         };
1449
1450         trace_ext4_nfs_commit_metadata(inode);
1451         return ext4_write_inode(inode, &wbc);
1452 }
1453
1454 #ifdef CONFIG_FS_ENCRYPTION
1455 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1456 {
1457         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1458                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1459 }
1460
1461 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1462                                                         void *fs_data)
1463 {
1464         handle_t *handle = fs_data;
1465         int res, res2, credits, retries = 0;
1466
1467         /*
1468          * Encrypting the root directory is not allowed because e2fsck expects
1469          * lost+found to exist and be unencrypted, and encrypting the root
1470          * directory would imply encrypting the lost+found directory as well as
1471          * the filename "lost+found" itself.
1472          */
1473         if (inode->i_ino == EXT4_ROOT_INO)
1474                 return -EPERM;
1475
1476         if (WARN_ON_ONCE(IS_DAX(inode) && i_size_read(inode)))
1477                 return -EINVAL;
1478
1479         if (ext4_test_inode_flag(inode, EXT4_INODE_DAX))
1480                 return -EOPNOTSUPP;
1481
1482         res = ext4_convert_inline_data(inode);
1483         if (res)
1484                 return res;
1485
1486         /*
1487          * If a journal handle was specified, then the encryption context is
1488          * being set on a new inode via inheritance and is part of a larger
1489          * transaction to create the inode.  Otherwise the encryption context is
1490          * being set on an existing inode in its own transaction.  Only in the
1491          * latter case should the "retry on ENOSPC" logic be used.
1492          */
1493
1494         if (handle) {
1495                 res = ext4_xattr_set_handle(handle, inode,
1496                                             EXT4_XATTR_INDEX_ENCRYPTION,
1497                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1498                                             ctx, len, 0);
1499                 if (!res) {
1500                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1501                         ext4_clear_inode_state(inode,
1502                                         EXT4_STATE_MAY_INLINE_DATA);
1503                         /*
1504                          * Update inode->i_flags - S_ENCRYPTED will be enabled,
1505                          * S_DAX may be disabled
1506                          */
1507                         ext4_set_inode_flags(inode, false);
1508                 }
1509                 return res;
1510         }
1511
1512         res = dquot_initialize(inode);
1513         if (res)
1514                 return res;
1515 retry:
1516         res = ext4_xattr_set_credits(inode, len, false /* is_create */,
1517                                      &credits);
1518         if (res)
1519                 return res;
1520
1521         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
1522         if (IS_ERR(handle))
1523                 return PTR_ERR(handle);
1524
1525         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1526                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1527                                     ctx, len, 0);
1528         if (!res) {
1529                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1530                 /*
1531                  * Update inode->i_flags - S_ENCRYPTED will be enabled,
1532                  * S_DAX may be disabled
1533                  */
1534                 ext4_set_inode_flags(inode, false);
1535                 res = ext4_mark_inode_dirty(handle, inode);
1536                 if (res)
1537                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1538         }
1539         res2 = ext4_journal_stop(handle);
1540
1541         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1542                 goto retry;
1543         if (!res)
1544                 res = res2;
1545         return res;
1546 }
1547
1548 static const union fscrypt_policy *ext4_get_dummy_policy(struct super_block *sb)
1549 {
1550         return EXT4_SB(sb)->s_dummy_enc_policy.policy;
1551 }
1552
1553 static bool ext4_has_stable_inodes(struct super_block *sb)
1554 {
1555         return ext4_has_feature_stable_inodes(sb);
1556 }
1557
1558 static void ext4_get_ino_and_lblk_bits(struct super_block *sb,
1559                                        int *ino_bits_ret, int *lblk_bits_ret)
1560 {
1561         *ino_bits_ret = 8 * sizeof(EXT4_SB(sb)->s_es->s_inodes_count);
1562         *lblk_bits_ret = 8 * sizeof(ext4_lblk_t);
1563 }
1564
1565 static const struct fscrypt_operations ext4_cryptops = {
1566         .key_prefix             = "ext4:",
1567         .get_context            = ext4_get_context,
1568         .set_context            = ext4_set_context,
1569         .get_dummy_policy       = ext4_get_dummy_policy,
1570         .empty_dir              = ext4_empty_dir,
1571         .max_namelen            = EXT4_NAME_LEN,
1572         .has_stable_inodes      = ext4_has_stable_inodes,
1573         .get_ino_and_lblk_bits  = ext4_get_ino_and_lblk_bits,
1574 };
1575 #endif
1576
1577 #ifdef CONFIG_QUOTA
1578 static const char * const quotatypes[] = INITQFNAMES;
1579 #define QTYPE2NAME(t) (quotatypes[t])
1580
1581 static int ext4_write_dquot(struct dquot *dquot);
1582 static int ext4_acquire_dquot(struct dquot *dquot);
1583 static int ext4_release_dquot(struct dquot *dquot);
1584 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1585 static int ext4_write_info(struct super_block *sb, int type);
1586 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1587                          const struct path *path);
1588 static int ext4_quota_on_mount(struct super_block *sb, int type);
1589 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1590                                size_t len, loff_t off);
1591 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1592                                 const char *data, size_t len, loff_t off);
1593 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1594                              unsigned int flags);
1595 static int ext4_enable_quotas(struct super_block *sb);
1596
1597 static struct dquot **ext4_get_dquots(struct inode *inode)
1598 {
1599         return EXT4_I(inode)->i_dquot;
1600 }
1601
1602 static const struct dquot_operations ext4_quota_operations = {
1603         .get_reserved_space     = ext4_get_reserved_space,
1604         .write_dquot            = ext4_write_dquot,
1605         .acquire_dquot          = ext4_acquire_dquot,
1606         .release_dquot          = ext4_release_dquot,
1607         .mark_dirty             = ext4_mark_dquot_dirty,
1608         .write_info             = ext4_write_info,
1609         .alloc_dquot            = dquot_alloc,
1610         .destroy_dquot          = dquot_destroy,
1611         .get_projid             = ext4_get_projid,
1612         .get_inode_usage        = ext4_get_inode_usage,
1613         .get_next_id            = dquot_get_next_id,
1614 };
1615
1616 static const struct quotactl_ops ext4_qctl_operations = {
1617         .quota_on       = ext4_quota_on,
1618         .quota_off      = ext4_quota_off,
1619         .quota_sync     = dquot_quota_sync,
1620         .get_state      = dquot_get_state,
1621         .set_info       = dquot_set_dqinfo,
1622         .get_dqblk      = dquot_get_dqblk,
1623         .set_dqblk      = dquot_set_dqblk,
1624         .get_nextdqblk  = dquot_get_next_dqblk,
1625 };
1626 #endif
1627
1628 static const struct super_operations ext4_sops = {
1629         .alloc_inode    = ext4_alloc_inode,
1630         .free_inode     = ext4_free_in_core_inode,
1631         .destroy_inode  = ext4_destroy_inode,
1632         .write_inode    = ext4_write_inode,
1633         .dirty_inode    = ext4_dirty_inode,
1634         .drop_inode     = ext4_drop_inode,
1635         .evict_inode    = ext4_evict_inode,
1636         .put_super      = ext4_put_super,
1637         .sync_fs        = ext4_sync_fs,
1638         .freeze_fs      = ext4_freeze,
1639         .unfreeze_fs    = ext4_unfreeze,
1640         .statfs         = ext4_statfs,
1641         .remount_fs     = ext4_remount,
1642         .show_options   = ext4_show_options,
1643 #ifdef CONFIG_QUOTA
1644         .quota_read     = ext4_quota_read,
1645         .quota_write    = ext4_quota_write,
1646         .get_dquots     = ext4_get_dquots,
1647 #endif
1648 };
1649
1650 static const struct export_operations ext4_export_ops = {
1651         .fh_to_dentry = ext4_fh_to_dentry,
1652         .fh_to_parent = ext4_fh_to_parent,
1653         .get_parent = ext4_get_parent,
1654         .commit_metadata = ext4_nfs_commit_metadata,
1655 };
1656
1657 enum {
1658         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1659         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1660         Opt_nouid32, Opt_debug, Opt_removed,
1661         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1662         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1663         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1664         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1665         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1666         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1667         Opt_inlinecrypt,
1668         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1669         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1670         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1671         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version,
1672         Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never,
1673         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
1674         Opt_nowarn_on_error, Opt_mblk_io_submit,
1675         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1676         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1677         Opt_inode_readahead_blks, Opt_journal_ioprio,
1678         Opt_dioread_nolock, Opt_dioread_lock,
1679         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1680         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1681         Opt_no_prefetch_block_bitmaps, Opt_mb_optimize_scan,
1682 #ifdef CONFIG_EXT4_DEBUG
1683         Opt_fc_debug_max_replay, Opt_fc_debug_force
1684 #endif
1685 };
1686
1687 static const match_table_t tokens = {
1688         {Opt_bsd_df, "bsddf"},
1689         {Opt_minix_df, "minixdf"},
1690         {Opt_grpid, "grpid"},
1691         {Opt_grpid, "bsdgroups"},
1692         {Opt_nogrpid, "nogrpid"},
1693         {Opt_nogrpid, "sysvgroups"},
1694         {Opt_resgid, "resgid=%u"},
1695         {Opt_resuid, "resuid=%u"},
1696         {Opt_sb, "sb=%u"},
1697         {Opt_err_cont, "errors=continue"},
1698         {Opt_err_panic, "errors=panic"},
1699         {Opt_err_ro, "errors=remount-ro"},
1700         {Opt_nouid32, "nouid32"},
1701         {Opt_debug, "debug"},
1702         {Opt_removed, "oldalloc"},
1703         {Opt_removed, "orlov"},
1704         {Opt_user_xattr, "user_xattr"},
1705         {Opt_nouser_xattr, "nouser_xattr"},
1706         {Opt_acl, "acl"},
1707         {Opt_noacl, "noacl"},
1708         {Opt_noload, "norecovery"},
1709         {Opt_noload, "noload"},
1710         {Opt_removed, "nobh"},
1711         {Opt_removed, "bh"},
1712         {Opt_commit, "commit=%u"},
1713         {Opt_min_batch_time, "min_batch_time=%u"},
1714         {Opt_max_batch_time, "max_batch_time=%u"},
1715         {Opt_journal_dev, "journal_dev=%u"},
1716         {Opt_journal_path, "journal_path=%s"},
1717         {Opt_journal_checksum, "journal_checksum"},
1718         {Opt_nojournal_checksum, "nojournal_checksum"},
1719         {Opt_journal_async_commit, "journal_async_commit"},
1720         {Opt_abort, "abort"},
1721         {Opt_data_journal, "data=journal"},
1722         {Opt_data_ordered, "data=ordered"},
1723         {Opt_data_writeback, "data=writeback"},
1724         {Opt_data_err_abort, "data_err=abort"},
1725         {Opt_data_err_ignore, "data_err=ignore"},
1726         {Opt_offusrjquota, "usrjquota="},
1727         {Opt_usrjquota, "usrjquota=%s"},
1728         {Opt_offgrpjquota, "grpjquota="},
1729         {Opt_grpjquota, "grpjquota=%s"},
1730         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1731         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1732         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1733         {Opt_grpquota, "grpquota"},
1734         {Opt_noquota, "noquota"},
1735         {Opt_quota, "quota"},
1736         {Opt_usrquota, "usrquota"},
1737         {Opt_prjquota, "prjquota"},
1738         {Opt_barrier, "barrier=%u"},
1739         {Opt_barrier, "barrier"},
1740         {Opt_nobarrier, "nobarrier"},
1741         {Opt_i_version, "i_version"},
1742         {Opt_dax, "dax"},
1743         {Opt_dax_always, "dax=always"},
1744         {Opt_dax_inode, "dax=inode"},
1745         {Opt_dax_never, "dax=never"},
1746         {Opt_stripe, "stripe=%u"},
1747         {Opt_delalloc, "delalloc"},
1748         {Opt_warn_on_error, "warn_on_error"},
1749         {Opt_nowarn_on_error, "nowarn_on_error"},
1750         {Opt_lazytime, "lazytime"},
1751         {Opt_nolazytime, "nolazytime"},
1752         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1753         {Opt_nodelalloc, "nodelalloc"},
1754         {Opt_removed, "mblk_io_submit"},
1755         {Opt_removed, "nomblk_io_submit"},
1756         {Opt_block_validity, "block_validity"},
1757         {Opt_noblock_validity, "noblock_validity"},
1758         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1759         {Opt_journal_ioprio, "journal_ioprio=%u"},
1760         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1761         {Opt_auto_da_alloc, "auto_da_alloc"},
1762         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1763         {Opt_dioread_nolock, "dioread_nolock"},
1764         {Opt_dioread_lock, "nodioread_nolock"},
1765         {Opt_dioread_lock, "dioread_lock"},
1766         {Opt_discard, "discard"},
1767         {Opt_nodiscard, "nodiscard"},
1768         {Opt_init_itable, "init_itable=%u"},
1769         {Opt_init_itable, "init_itable"},
1770         {Opt_noinit_itable, "noinit_itable"},
1771 #ifdef CONFIG_EXT4_DEBUG
1772         {Opt_fc_debug_force, "fc_debug_force"},
1773         {Opt_fc_debug_max_replay, "fc_debug_max_replay=%u"},
1774 #endif
1775         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1776         {Opt_test_dummy_encryption, "test_dummy_encryption=%s"},
1777         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1778         {Opt_inlinecrypt, "inlinecrypt"},
1779         {Opt_nombcache, "nombcache"},
1780         {Opt_nombcache, "no_mbcache"},  /* for backward compatibility */
1781         {Opt_removed, "prefetch_block_bitmaps"},
1782         {Opt_no_prefetch_block_bitmaps, "no_prefetch_block_bitmaps"},
1783         {Opt_mb_optimize_scan, "mb_optimize_scan=%d"},
1784         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1785         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1786         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1787         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1788         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1789         {Opt_err, NULL},
1790 };
1791
1792 static ext4_fsblk_t get_sb_block(void **data)
1793 {
1794         ext4_fsblk_t    sb_block;
1795         char            *options = (char *) *data;
1796
1797         if (!options || strncmp(options, "sb=", 3) != 0)
1798                 return 1;       /* Default location */
1799
1800         options += 3;
1801         /* TODO: use simple_strtoll with >32bit ext4 */
1802         sb_block = simple_strtoul(options, &options, 0);
1803         if (*options && *options != ',') {
1804                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1805                        (char *) *data);
1806                 return 1;
1807         }
1808         if (*options == ',')
1809                 options++;
1810         *data = (void *) options;
1811
1812         return sb_block;
1813 }
1814
1815 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1816 #define DEFAULT_MB_OPTIMIZE_SCAN        (-1)
1817
1818 static const char deprecated_msg[] =
1819         "Mount option \"%s\" will be removed by %s\n"
1820         "Contact [email protected] if you think we should keep it.\n";
1821
1822 #ifdef CONFIG_QUOTA
1823 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1824 {
1825         struct ext4_sb_info *sbi = EXT4_SB(sb);
1826         char *qname, *old_qname = get_qf_name(sb, sbi, qtype);
1827         int ret = -1;
1828
1829         if (sb_any_quota_loaded(sb) && !old_qname) {
1830                 ext4_msg(sb, KERN_ERR,
1831                         "Cannot change journaled "
1832                         "quota options when quota turned on");
1833                 return -1;
1834         }
1835         if (ext4_has_feature_quota(sb)) {
1836                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1837                          "ignored when QUOTA feature is enabled");
1838                 return 1;
1839         }
1840         qname = match_strdup(args);
1841         if (!qname) {
1842                 ext4_msg(sb, KERN_ERR,
1843                         "Not enough memory for storing quotafile name");
1844                 return -1;
1845         }
1846         if (old_qname) {
1847                 if (strcmp(old_qname, qname) == 0)
1848                         ret = 1;
1849                 else
1850                         ext4_msg(sb, KERN_ERR,
1851                                  "%s quota file already specified",
1852                                  QTYPE2NAME(qtype));
1853                 goto errout;
1854         }
1855         if (strchr(qname, '/')) {
1856                 ext4_msg(sb, KERN_ERR,
1857                         "quotafile must be on filesystem root");
1858                 goto errout;
1859         }
1860         rcu_assign_pointer(sbi->s_qf_names[qtype], qname);
1861         set_opt(sb, QUOTA);
1862         return 1;
1863 errout:
1864         kfree(qname);
1865         return ret;
1866 }
1867
1868 static int clear_qf_name(struct super_block *sb, int qtype)
1869 {
1870
1871         struct ext4_sb_info *sbi = EXT4_SB(sb);
1872         char *old_qname = get_qf_name(sb, sbi, qtype);
1873
1874         if (sb_any_quota_loaded(sb) && old_qname) {
1875                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1876                         " when quota turned on");
1877                 return -1;
1878         }
1879         rcu_assign_pointer(sbi->s_qf_names[qtype], NULL);
1880         synchronize_rcu();
1881         kfree(old_qname);
1882         return 1;
1883 }
1884 #endif
1885
1886 #define MOPT_SET        0x0001
1887 #define MOPT_CLEAR      0x0002
1888 #define MOPT_NOSUPPORT  0x0004
1889 #define MOPT_EXPLICIT   0x0008
1890 #define MOPT_CLEAR_ERR  0x0010
1891 #define MOPT_GTE0       0x0020
1892 #ifdef CONFIG_QUOTA
1893 #define MOPT_Q          0
1894 #define MOPT_QFMT       0x0040
1895 #else
1896 #define MOPT_Q          MOPT_NOSUPPORT
1897 #define MOPT_QFMT       MOPT_NOSUPPORT
1898 #endif
1899 #define MOPT_DATAJ      0x0080
1900 #define MOPT_NO_EXT2    0x0100
1901 #define MOPT_NO_EXT3    0x0200
1902 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1903 #define MOPT_STRING     0x0400
1904 #define MOPT_SKIP       0x0800
1905 #define MOPT_2          0x1000
1906
1907 static const struct mount_opts {
1908         int     token;
1909         int     mount_opt;
1910         int     flags;
1911 } ext4_mount_opts[] = {
1912         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1913         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1914         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1915         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1916         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1917         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1918         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1919          MOPT_EXT4_ONLY | MOPT_SET},
1920         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1921          MOPT_EXT4_ONLY | MOPT_CLEAR},
1922         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1923         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1924         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1925          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1926         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1927          MOPT_EXT4_ONLY | MOPT_CLEAR},
1928         {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET},
1929         {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR},
1930         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1931          MOPT_EXT4_ONLY | MOPT_CLEAR},
1932         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1933          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1934         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1935                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1936          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1937         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1938         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1939         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1940         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1941         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1942          MOPT_NO_EXT2},
1943         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1944          MOPT_NO_EXT2},
1945         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1946         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1947         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1948         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1949         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1950         {Opt_commit, 0, MOPT_GTE0},
1951         {Opt_max_batch_time, 0, MOPT_GTE0},
1952         {Opt_min_batch_time, 0, MOPT_GTE0},
1953         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1954         {Opt_init_itable, 0, MOPT_GTE0},
1955         {Opt_dax, EXT4_MOUNT_DAX_ALWAYS, MOPT_SET | MOPT_SKIP},
1956         {Opt_dax_always, EXT4_MOUNT_DAX_ALWAYS,
1957                 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP},
1958         {Opt_dax_inode, EXT4_MOUNT2_DAX_INODE,
1959                 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP},
1960         {Opt_dax_never, EXT4_MOUNT2_DAX_NEVER,
1961                 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP},
1962         {Opt_stripe, 0, MOPT_GTE0},
1963         {Opt_resuid, 0, MOPT_GTE0},
1964         {Opt_resgid, 0, MOPT_GTE0},
1965         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1966         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1967         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1968         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1969         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1970         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1971          MOPT_NO_EXT2 | MOPT_DATAJ},
1972         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1973         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1974 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1975         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1976         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1977 #else
1978         {Opt_acl, 0, MOPT_NOSUPPORT},
1979         {Opt_noacl, 0, MOPT_NOSUPPORT},
1980 #endif
1981         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1982         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1983         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1984         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1985         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1986                                                         MOPT_SET | MOPT_Q},
1987         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1988                                                         MOPT_SET | MOPT_Q},
1989         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1990                                                         MOPT_SET | MOPT_Q},
1991         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1992                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1993                                                         MOPT_CLEAR | MOPT_Q},
1994         {Opt_usrjquota, 0, MOPT_Q | MOPT_STRING},
1995         {Opt_grpjquota, 0, MOPT_Q | MOPT_STRING},
1996         {Opt_offusrjquota, 0, MOPT_Q},
1997         {Opt_offgrpjquota, 0, MOPT_Q},
1998         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1999         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
2000         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
2001         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
2002         {Opt_test_dummy_encryption, 0, MOPT_STRING},
2003         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
2004         {Opt_no_prefetch_block_bitmaps, EXT4_MOUNT_NO_PREFETCH_BLOCK_BITMAPS,
2005          MOPT_SET},
2006         {Opt_mb_optimize_scan, EXT4_MOUNT2_MB_OPTIMIZE_SCAN, MOPT_GTE0},
2007 #ifdef CONFIG_EXT4_DEBUG
2008         {Opt_fc_debug_force, EXT4_MOUNT2_JOURNAL_FAST_COMMIT,
2009          MOPT_SET | MOPT_2 | MOPT_EXT4_ONLY},
2010         {Opt_fc_debug_max_replay, 0, MOPT_GTE0},
2011 #endif
2012         {Opt_err, 0, 0}
2013 };
2014
2015 #ifdef CONFIG_UNICODE
2016 static const struct ext4_sb_encodings {
2017         __u16 magic;
2018         char *name;
2019         char *version;
2020 } ext4_sb_encoding_map[] = {
2021         {EXT4_ENC_UTF8_12_1, "utf8", "12.1.0"},
2022 };
2023
2024 static int ext4_sb_read_encoding(const struct ext4_super_block *es,
2025                                  const struct ext4_sb_encodings **encoding,
2026                                  __u16 *flags)
2027 {
2028         __u16 magic = le16_to_cpu(es->s_encoding);
2029         int i;
2030
2031         for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++)
2032                 if (magic == ext4_sb_encoding_map[i].magic)
2033                         break;
2034
2035         if (i >= ARRAY_SIZE(ext4_sb_encoding_map))
2036                 return -EINVAL;
2037
2038         *encoding = &ext4_sb_encoding_map[i];
2039         *flags = le16_to_cpu(es->s_encoding_flags);
2040
2041         return 0;
2042 }
2043 #endif
2044
2045 static int ext4_set_test_dummy_encryption(struct super_block *sb,
2046                                           const char *opt,
2047                                           const substring_t *arg,
2048                                           bool is_remount)
2049 {
2050 #ifdef CONFIG_FS_ENCRYPTION
2051         struct ext4_sb_info *sbi = EXT4_SB(sb);
2052         int err;
2053
2054         /*
2055          * This mount option is just for testing, and it's not worthwhile to
2056          * implement the extra complexity (e.g. RCU protection) that would be
2057          * needed to allow it to be set or changed during remount.  We do allow
2058          * it to be specified during remount, but only if there is no change.
2059          */
2060         if (is_remount && !sbi->s_dummy_enc_policy.policy) {
2061                 ext4_msg(sb, KERN_WARNING,
2062                          "Can't set test_dummy_encryption on remount");
2063                 return -1;
2064         }
2065         err = fscrypt_set_test_dummy_encryption(sb, arg->from,
2066                                                 &sbi->s_dummy_enc_policy);
2067         if (err) {
2068                 if (err == -EEXIST)
2069                         ext4_msg(sb, KERN_WARNING,
2070                                  "Can't change test_dummy_encryption on remount");
2071                 else if (err == -EINVAL)
2072                         ext4_msg(sb, KERN_WARNING,
2073                                  "Value of option \"%s\" is unrecognized", opt);
2074                 else
2075                         ext4_msg(sb, KERN_WARNING,
2076                                  "Error processing option \"%s\" [%d]",
2077                                  opt, err);
2078                 return -1;
2079         }
2080         ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled");
2081 #else
2082         ext4_msg(sb, KERN_WARNING,
2083                  "Test dummy encryption mount option ignored");
2084 #endif
2085         return 1;
2086 }
2087
2088 struct ext4_parsed_options {
2089         unsigned long journal_devnum;
2090         unsigned int journal_ioprio;
2091         int mb_optimize_scan;
2092 };
2093
2094 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
2095                             substring_t *args, struct ext4_parsed_options *parsed_opts,
2096                             int is_remount)
2097 {
2098         struct ext4_sb_info *sbi = EXT4_SB(sb);
2099         const struct mount_opts *m;
2100         kuid_t uid;
2101         kgid_t gid;
2102         int arg = 0;
2103
2104 #ifdef CONFIG_QUOTA
2105         if (token == Opt_usrjquota)
2106                 return set_qf_name(sb, USRQUOTA, &args[0]);
2107         else if (token == Opt_grpjquota)
2108                 return set_qf_name(sb, GRPQUOTA, &args[0]);
2109         else if (token == Opt_offusrjquota)
2110                 return clear_qf_name(sb, USRQUOTA);
2111         else if (token == Opt_offgrpjquota)
2112                 return clear_qf_name(sb, GRPQUOTA);
2113 #endif
2114         switch (token) {
2115         case Opt_noacl:
2116         case Opt_nouser_xattr:
2117                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
2118                 break;
2119         case Opt_sb:
2120                 return 1;       /* handled by get_sb_block() */
2121         case Opt_removed:
2122                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
2123                 return 1;
2124         case Opt_abort:
2125                 ext4_set_mount_flag(sb, EXT4_MF_FS_ABORTED);
2126                 return 1;
2127         case Opt_i_version:
2128                 sb->s_flags |= SB_I_VERSION;
2129                 return 1;
2130         case Opt_lazytime:
2131                 sb->s_flags |= SB_LAZYTIME;
2132                 return 1;
2133         case Opt_nolazytime:
2134                 sb->s_flags &= ~SB_LAZYTIME;
2135                 return 1;
2136         case Opt_inlinecrypt:
2137 #ifdef CONFIG_FS_ENCRYPTION_INLINE_CRYPT
2138                 sb->s_flags |= SB_INLINECRYPT;
2139 #else
2140                 ext4_msg(sb, KERN_ERR, "inline encryption not supported");
2141 #endif
2142                 return 1;
2143         }
2144
2145         for (m = ext4_mount_opts; m->token != Opt_err; m++)
2146                 if (token == m->token)
2147                         break;
2148
2149         if (m->token == Opt_err) {
2150                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
2151                          "or missing value", opt);
2152                 return -1;
2153         }
2154
2155         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
2156                 ext4_msg(sb, KERN_ERR,
2157                          "Mount option \"%s\" incompatible with ext2", opt);
2158                 return -1;
2159         }
2160         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
2161                 ext4_msg(sb, KERN_ERR,
2162                          "Mount option \"%s\" incompatible with ext3", opt);
2163                 return -1;
2164         }
2165
2166         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
2167                 return -1;
2168         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
2169                 return -1;
2170         if (m->flags & MOPT_EXPLICIT) {
2171                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
2172                         set_opt2(sb, EXPLICIT_DELALLOC);
2173                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
2174                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
2175                 } else
2176                         return -1;
2177         }
2178         if (m->flags & MOPT_CLEAR_ERR)
2179                 clear_opt(sb, ERRORS_MASK);
2180         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
2181                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
2182                          "options when quota turned on");
2183                 return -1;
2184         }
2185
2186         if (m->flags & MOPT_NOSUPPORT) {
2187                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
2188         } else if (token == Opt_commit) {
2189                 if (arg == 0)
2190                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
2191                 else if (arg > INT_MAX / HZ) {
2192                         ext4_msg(sb, KERN_ERR,
2193                                  "Invalid commit interval %d, "
2194                                  "must be smaller than %d",
2195                                  arg, INT_MAX / HZ);
2196                         return -1;
2197                 }
2198                 sbi->s_commit_interval = HZ * arg;
2199         } else if (token == Opt_debug_want_extra_isize) {
2200                 if ((arg & 1) ||
2201                     (arg < 4) ||
2202                     (arg > (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE))) {
2203                         ext4_msg(sb, KERN_ERR,
2204                                  "Invalid want_extra_isize %d", arg);
2205                         return -1;
2206                 }
2207                 sbi->s_want_extra_isize = arg;
2208         } else if (token == Opt_max_batch_time) {
2209                 sbi->s_max_batch_time = arg;
2210         } else if (token == Opt_min_batch_time) {
2211                 sbi->s_min_batch_time = arg;
2212         } else if (token == Opt_inode_readahead_blks) {
2213                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
2214                         ext4_msg(sb, KERN_ERR,
2215                                  "EXT4-fs: inode_readahead_blks must be "
2216                                  "0 or a power of 2 smaller than 2^31");
2217                         return -1;
2218                 }
2219                 sbi->s_inode_readahead_blks = arg;
2220         } else if (token == Opt_init_itable) {
2221                 set_opt(sb, INIT_INODE_TABLE);
2222                 if (!args->from)
2223                         arg = EXT4_DEF_LI_WAIT_MULT;
2224                 sbi->s_li_wait_mult = arg;
2225         } else if (token == Opt_max_dir_size_kb) {
2226                 sbi->s_max_dir_size_kb = arg;
2227 #ifdef CONFIG_EXT4_DEBUG
2228         } else if (token == Opt_fc_debug_max_replay) {
2229                 sbi->s_fc_debug_max_replay = arg;
2230 #endif
2231         } else if (token == Opt_stripe) {
2232                 sbi->s_stripe = arg;
2233         } else if (token == Opt_resuid) {
2234                 uid = make_kuid(current_user_ns(), arg);
2235                 if (!uid_valid(uid)) {
2236                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
2237                         return -1;
2238                 }
2239                 sbi->s_resuid = uid;
2240         } else if (token == Opt_resgid) {
2241                 gid = make_kgid(current_user_ns(), arg);
2242                 if (!gid_valid(gid)) {
2243                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
2244                         return -1;
2245                 }
2246                 sbi->s_resgid = gid;
2247         } else if (token == Opt_journal_dev) {
2248                 if (is_remount) {
2249                         ext4_msg(sb, KERN_ERR,
2250                                  "Cannot specify journal on remount");
2251                         return -1;
2252                 }
2253                 parsed_opts->journal_devnum = arg;
2254         } else if (token == Opt_journal_path) {
2255                 char *journal_path;
2256                 struct inode *journal_inode;
2257                 struct path path;
2258                 int error;
2259
2260                 if (is_remount) {
2261                         ext4_msg(sb, KERN_ERR,
2262                                  "Cannot specify journal on remount");
2263                         return -1;
2264                 }
2265                 journal_path = match_strdup(&args[0]);
2266                 if (!journal_path) {
2267                         ext4_msg(sb, KERN_ERR, "error: could not dup "
2268                                 "journal device string");
2269                         return -1;
2270                 }
2271
2272                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
2273                 if (error) {
2274                         ext4_msg(sb, KERN_ERR, "error: could not find "
2275                                 "journal device path: error %d", error);
2276                         kfree(journal_path);
2277                         return -1;
2278                 }
2279
2280                 journal_inode = d_inode(path.dentry);
2281                 if (!S_ISBLK(journal_inode->i_mode)) {
2282                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
2283                                 "is not a block device", journal_path);
2284                         path_put(&path);
2285                         kfree(journal_path);
2286                         return -1;
2287                 }
2288
2289                 parsed_opts->journal_devnum = new_encode_dev(journal_inode->i_rdev);
2290                 path_put(&path);
2291                 kfree(journal_path);
2292         } else if (token == Opt_journal_ioprio) {
2293                 if (arg > 7) {
2294                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
2295                                  " (must be 0-7)");
2296                         return -1;
2297                 }
2298                 parsed_opts->journal_ioprio =
2299                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
2300         } else if (token == Opt_test_dummy_encryption) {
2301                 return ext4_set_test_dummy_encryption(sb, opt, &args[0],
2302                                                       is_remount);
2303         } else if (m->flags & MOPT_DATAJ) {
2304                 if (is_remount) {
2305                         if (!sbi->s_journal)
2306                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
2307                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
2308                                 ext4_msg(sb, KERN_ERR,
2309                                          "Cannot change data mode on remount");
2310                                 return -1;
2311                         }
2312                 } else {
2313                         clear_opt(sb, DATA_FLAGS);
2314                         sbi->s_mount_opt |= m->mount_opt;
2315                 }
2316 #ifdef CONFIG_QUOTA
2317         } else if (m->flags & MOPT_QFMT) {
2318                 if (sb_any_quota_loaded(sb) &&
2319                     sbi->s_jquota_fmt != m->mount_opt) {
2320                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
2321                                  "quota options when quota turned on");
2322                         return -1;
2323                 }
2324                 if (ext4_has_feature_quota(sb)) {
2325                         ext4_msg(sb, KERN_INFO,
2326                                  "Quota format mount options ignored "
2327                                  "when QUOTA feature is enabled");
2328                         return 1;
2329                 }
2330                 sbi->s_jquota_fmt = m->mount_opt;
2331 #endif
2332         } else if (token == Opt_dax || token == Opt_dax_always ||
2333                    token == Opt_dax_inode || token == Opt_dax_never) {
2334 #ifdef CONFIG_FS_DAX
2335                 switch (token) {
2336                 case Opt_dax:
2337                 case Opt_dax_always:
2338                         if (is_remount &&
2339                             (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2340                              (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) {
2341                         fail_dax_change_remount:
2342                                 ext4_msg(sb, KERN_ERR, "can't change "
2343                                          "dax mount option while remounting");
2344                                 return -1;
2345                         }
2346                         if (is_remount &&
2347                             (test_opt(sb, DATA_FLAGS) ==
2348                              EXT4_MOUNT_JOURNAL_DATA)) {
2349                                     ext4_msg(sb, KERN_ERR, "can't mount with "
2350                                              "both data=journal and dax");
2351                                     return -1;
2352                         }
2353                         ext4_msg(sb, KERN_WARNING,
2354                                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
2355                         sbi->s_mount_opt |= EXT4_MOUNT_DAX_ALWAYS;
2356                         sbi->s_mount_opt2 &= ~EXT4_MOUNT2_DAX_NEVER;
2357                         break;
2358                 case Opt_dax_never:
2359                         if (is_remount &&
2360                             (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2361                              (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS)))
2362                                 goto fail_dax_change_remount;
2363                         sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_NEVER;
2364                         sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS;
2365                         break;
2366                 case Opt_dax_inode:
2367                         if (is_remount &&
2368                             ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2369                              (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2370                              !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE)))
2371                                 goto fail_dax_change_remount;
2372                         sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS;
2373                         sbi->s_mount_opt2 &= ~EXT4_MOUNT2_DAX_NEVER;
2374                         /* Strictly for printing options */
2375                         sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_INODE;
2376                         break;
2377                 }
2378 #else
2379                 ext4_msg(sb, KERN_INFO, "dax option not supported");
2380                 sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_NEVER;
2381                 sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS;
2382                 return -1;
2383 #endif
2384         } else if (token == Opt_data_err_abort) {
2385                 sbi->s_mount_opt |= m->mount_opt;
2386         } else if (token == Opt_data_err_ignore) {
2387                 sbi->s_mount_opt &= ~m->mount_opt;
2388         } else if (token == Opt_mb_optimize_scan) {
2389                 if (arg != 0 && arg != 1) {
2390                         ext4_msg(sb, KERN_WARNING,
2391                                  "mb_optimize_scan should be set to 0 or 1.");
2392                         return -1;
2393                 }
2394                 parsed_opts->mb_optimize_scan = arg;
2395         } else {
2396                 if (!args->from)
2397                         arg = 1;
2398                 if (m->flags & MOPT_CLEAR)
2399                         arg = !arg;
2400                 else if (unlikely(!(m->flags & MOPT_SET))) {
2401                         ext4_msg(sb, KERN_WARNING,
2402                                  "buggy handling of option %s", opt);
2403                         WARN_ON(1);
2404                         return -1;
2405                 }
2406                 if (m->flags & MOPT_2) {
2407                         if (arg != 0)
2408                                 sbi->s_mount_opt2 |= m->mount_opt;
2409                         else
2410                                 sbi->s_mount_opt2 &= ~m->mount_opt;
2411                 } else {
2412                         if (arg != 0)
2413                                 sbi->s_mount_opt |= m->mount_opt;
2414                         else
2415                                 sbi->s_mount_opt &= ~m->mount_opt;
2416                 }
2417         }
2418         return 1;
2419 }
2420
2421 static int parse_options(char *options, struct super_block *sb,
2422                          struct ext4_parsed_options *ret_opts,
2423                          int is_remount)
2424 {
2425         struct ext4_sb_info __maybe_unused *sbi = EXT4_SB(sb);
2426         char *p, __maybe_unused *usr_qf_name, __maybe_unused *grp_qf_name;
2427         substring_t args[MAX_OPT_ARGS];
2428         int token;
2429
2430         if (!options)
2431                 return 1;
2432
2433         while ((p = strsep(&options, ",")) != NULL) {
2434                 if (!*p)
2435                         continue;
2436                 /*
2437                  * Initialize args struct so we know whether arg was
2438                  * found; some options take optional arguments.
2439                  */
2440                 args[0].to = args[0].from = NULL;
2441                 token = match_token(p, tokens, args);
2442                 if (handle_mount_opt(sb, p, token, args, ret_opts,
2443                                      is_remount) < 0)
2444                         return 0;
2445         }
2446 #ifdef CONFIG_QUOTA
2447         /*
2448          * We do the test below only for project quotas. 'usrquota' and
2449          * 'grpquota' mount options are allowed even without quota feature
2450          * to support legacy quotas in quota files.
2451          */
2452         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
2453                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
2454                          "Cannot enable project quota enforcement.");
2455                 return 0;
2456         }
2457         usr_qf_name = get_qf_name(sb, sbi, USRQUOTA);
2458         grp_qf_name = get_qf_name(sb, sbi, GRPQUOTA);
2459         if (usr_qf_name || grp_qf_name) {
2460                 if (test_opt(sb, USRQUOTA) && usr_qf_name)
2461                         clear_opt(sb, USRQUOTA);
2462
2463                 if (test_opt(sb, GRPQUOTA) && grp_qf_name)
2464                         clear_opt(sb, GRPQUOTA);
2465
2466                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
2467                         ext4_msg(sb, KERN_ERR, "old and new quota "
2468                                         "format mixing");
2469                         return 0;
2470                 }
2471
2472                 if (!sbi->s_jquota_fmt) {
2473                         ext4_msg(sb, KERN_ERR, "journaled quota format "
2474                                         "not specified");
2475                         return 0;
2476                 }
2477         }
2478 #endif
2479         if (test_opt(sb, DIOREAD_NOLOCK)) {
2480                 int blocksize =
2481                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
2482                 if (blocksize < PAGE_SIZE)
2483                         ext4_msg(sb, KERN_WARNING, "Warning: mounting with an "
2484                                  "experimental mount option 'dioread_nolock' "
2485                                  "for blocksize < PAGE_SIZE");
2486         }
2487         return 1;
2488 }
2489
2490 static inline void ext4_show_quota_options(struct seq_file *seq,
2491                                            struct super_block *sb)
2492 {
2493 #if defined(CONFIG_QUOTA)
2494         struct ext4_sb_info *sbi = EXT4_SB(sb);
2495         char *usr_qf_name, *grp_qf_name;
2496
2497         if (sbi->s_jquota_fmt) {
2498                 char *fmtname = "";
2499
2500                 switch (sbi->s_jquota_fmt) {
2501                 case QFMT_VFS_OLD:
2502                         fmtname = "vfsold";
2503                         break;
2504                 case QFMT_VFS_V0:
2505                         fmtname = "vfsv0";
2506                         break;
2507                 case QFMT_VFS_V1:
2508                         fmtname = "vfsv1";
2509                         break;
2510                 }
2511                 seq_printf(seq, ",jqfmt=%s", fmtname);
2512         }
2513
2514         rcu_read_lock();
2515         usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]);
2516         grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]);
2517         if (usr_qf_name)
2518                 seq_show_option(seq, "usrjquota", usr_qf_name);
2519         if (grp_qf_name)
2520                 seq_show_option(seq, "grpjquota", grp_qf_name);
2521         rcu_read_unlock();
2522 #endif
2523 }
2524
2525 static const char *token2str(int token)
2526 {
2527         const struct match_token *t;
2528
2529         for (t = tokens; t->token != Opt_err; t++)
2530                 if (t->token == token && !strchr(t->pattern, '='))
2531                         break;
2532         return t->pattern;
2533 }
2534
2535 /*
2536  * Show an option if
2537  *  - it's set to a non-default value OR
2538  *  - if the per-sb default is different from the global default
2539  */
2540 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2541                               int nodefs)
2542 {
2543         struct ext4_sb_info *sbi = EXT4_SB(sb);
2544         struct ext4_super_block *es = sbi->s_es;
2545         int def_errors, def_mount_opt = sbi->s_def_mount_opt;
2546         const struct mount_opts *m;
2547         char sep = nodefs ? '\n' : ',';
2548
2549 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2550 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2551
2552         if (sbi->s_sb_block != 1)
2553                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2554
2555         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2556                 int want_set = m->flags & MOPT_SET;
2557                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2558                     (m->flags & MOPT_CLEAR_ERR) || m->flags & MOPT_SKIP)
2559                         continue;
2560                 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2561                         continue; /* skip if same as the default */
2562                 if ((want_set &&
2563                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2564                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2565                         continue; /* select Opt_noFoo vs Opt_Foo */
2566                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2567         }
2568
2569         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2570             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2571                 SEQ_OPTS_PRINT("resuid=%u",
2572                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2573         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2574             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2575                 SEQ_OPTS_PRINT("resgid=%u",
2576                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2577         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2578         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2579                 SEQ_OPTS_PUTS("errors=remount-ro");
2580         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2581                 SEQ_OPTS_PUTS("errors=continue");
2582         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2583                 SEQ_OPTS_PUTS("errors=panic");
2584         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2585                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2586         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2587                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2588         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2589                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2590         if (sb->s_flags & SB_I_VERSION)
2591                 SEQ_OPTS_PUTS("i_version");
2592         if (nodefs || sbi->s_stripe)
2593                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2594         if (nodefs || EXT4_MOUNT_DATA_FLAGS &
2595                         (sbi->s_mount_opt ^ def_mount_opt)) {
2596                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2597                         SEQ_OPTS_PUTS("data=journal");
2598                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2599                         SEQ_OPTS_PUTS("data=ordered");
2600                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2601                         SEQ_OPTS_PUTS("data=writeback");
2602         }
2603         if (nodefs ||
2604             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2605                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2606                                sbi->s_inode_readahead_blks);
2607
2608         if (test_opt(sb, INIT_INODE_TABLE) && (nodefs ||
2609                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2610                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2611         if (nodefs || sbi->s_max_dir_size_kb)
2612                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2613         if (test_opt(sb, DATA_ERR_ABORT))
2614                 SEQ_OPTS_PUTS("data_err=abort");
2615
2616         fscrypt_show_test_dummy_encryption(seq, sep, sb);
2617
2618         if (sb->s_flags & SB_INLINECRYPT)
2619                 SEQ_OPTS_PUTS("inlinecrypt");
2620
2621         if (test_opt(sb, DAX_ALWAYS)) {
2622                 if (IS_EXT2_SB(sb))
2623                         SEQ_OPTS_PUTS("dax");
2624                 else
2625                         SEQ_OPTS_PUTS("dax=always");
2626         } else if (test_opt2(sb, DAX_NEVER)) {
2627                 SEQ_OPTS_PUTS("dax=never");
2628         } else if (test_opt2(sb, DAX_INODE)) {
2629                 SEQ_OPTS_PUTS("dax=inode");
2630         }
2631         ext4_show_quota_options(seq, sb);
2632         return 0;
2633 }
2634
2635 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2636 {
2637         return _ext4_show_options(seq, root->d_sb, 0);
2638 }
2639
2640 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2641 {
2642         struct super_block *sb = seq->private;
2643         int rc;
2644
2645         seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
2646         rc = _ext4_show_options(seq, sb, 1);
2647         seq_puts(seq, "\n");
2648         return rc;
2649 }
2650
2651 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2652                             int read_only)
2653 {
2654         struct ext4_sb_info *sbi = EXT4_SB(sb);
2655         int err = 0;
2656
2657         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2658                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2659                          "forcing read-only mode");
2660                 err = -EROFS;
2661                 goto done;
2662         }
2663         if (read_only)
2664                 goto done;
2665         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2666                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2667                          "running e2fsck is recommended");
2668         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2669                 ext4_msg(sb, KERN_WARNING,
2670                          "warning: mounting fs with errors, "
2671                          "running e2fsck is recommended");
2672         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2673                  le16_to_cpu(es->s_mnt_count) >=
2674                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2675                 ext4_msg(sb, KERN_WARNING,
2676                          "warning: maximal mount count reached, "
2677                          "running e2fsck is recommended");
2678         else if (le32_to_cpu(es->s_checkinterval) &&
2679                  (ext4_get_tstamp(es, s_lastcheck) +
2680                   le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds()))
2681                 ext4_msg(sb, KERN_WARNING,
2682                          "warning: checktime reached, "
2683                          "running e2fsck is recommended");
2684         if (!sbi->s_journal)
2685                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2686         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2687                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2688         le16_add_cpu(&es->s_mnt_count, 1);
2689         ext4_update_tstamp(es, s_mtime);
2690         if (sbi->s_journal)
2691                 ext4_set_feature_journal_needs_recovery(sb);
2692
2693         err = ext4_commit_super(sb);
2694 done:
2695         if (test_opt(sb, DEBUG))
2696                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2697                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2698                         sb->s_blocksize,
2699                         sbi->s_groups_count,
2700                         EXT4_BLOCKS_PER_GROUP(sb),
2701                         EXT4_INODES_PER_GROUP(sb),
2702                         sbi->s_mount_opt, sbi->s_mount_opt2);
2703
2704         cleancache_init_fs(sb);
2705         return err;
2706 }
2707
2708 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2709 {
2710         struct ext4_sb_info *sbi = EXT4_SB(sb);
2711         struct flex_groups **old_groups, **new_groups;
2712         int size, i, j;
2713
2714         if (!sbi->s_log_groups_per_flex)
2715                 return 0;
2716
2717         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2718         if (size <= sbi->s_flex_groups_allocated)
2719                 return 0;
2720
2721         new_groups = kvzalloc(roundup_pow_of_two(size *
2722                               sizeof(*sbi->s_flex_groups)), GFP_KERNEL);
2723         if (!new_groups) {
2724                 ext4_msg(sb, KERN_ERR,
2725                          "not enough memory for %d flex group pointers", size);
2726                 return -ENOMEM;
2727         }
2728         for (i = sbi->s_flex_groups_allocated; i < size; i++) {
2729                 new_groups[i] = kvzalloc(roundup_pow_of_two(
2730                                          sizeof(struct flex_groups)),
2731                                          GFP_KERNEL);
2732                 if (!new_groups[i]) {
2733                         for (j = sbi->s_flex_groups_allocated; j < i; j++)
2734                                 kvfree(new_groups[j]);
2735                         kvfree(new_groups);
2736                         ext4_msg(sb, KERN_ERR,
2737                                  "not enough memory for %d flex groups", size);
2738                         return -ENOMEM;
2739                 }
2740         }
2741         rcu_read_lock();
2742         old_groups = rcu_dereference(sbi->s_flex_groups);
2743         if (old_groups)
2744                 memcpy(new_groups, old_groups,
2745                        (sbi->s_flex_groups_allocated *
2746                         sizeof(struct flex_groups *)));
2747         rcu_read_unlock();
2748         rcu_assign_pointer(sbi->s_flex_groups, new_groups);
2749         sbi->s_flex_groups_allocated = size;
2750         if (old_groups)
2751                 ext4_kvfree_array_rcu(old_groups);
2752         return 0;
2753 }
2754
2755 static int ext4_fill_flex_info(struct super_block *sb)
2756 {
2757         struct ext4_sb_info *sbi = EXT4_SB(sb);
2758         struct ext4_group_desc *gdp = NULL;
2759         struct flex_groups *fg;
2760         ext4_group_t flex_group;
2761         int i, err;
2762
2763         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2764         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2765                 sbi->s_log_groups_per_flex = 0;
2766                 return 1;
2767         }
2768
2769         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2770         if (err)
2771                 goto failed;
2772
2773         for (i = 0; i < sbi->s_groups_count; i++) {
2774                 gdp = ext4_get_group_desc(sb, i, NULL);
2775
2776                 flex_group = ext4_flex_group(sbi, i);
2777                 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
2778                 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes);
2779                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2780                              &fg->free_clusters);
2781                 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs);
2782         }
2783
2784         return 1;
2785 failed:
2786         return 0;
2787 }
2788
2789 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2790                                    struct ext4_group_desc *gdp)
2791 {
2792         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2793         __u16 crc = 0;
2794         __le32 le_group = cpu_to_le32(block_group);
2795         struct ext4_sb_info *sbi = EXT4_SB(sb);
2796
2797         if (ext4_has_metadata_csum(sbi->s_sb)) {
2798                 /* Use new metadata_csum algorithm */
2799                 __u32 csum32;
2800                 __u16 dummy_csum = 0;
2801
2802                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2803                                      sizeof(le_group));
2804                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2805                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2806                                      sizeof(dummy_csum));
2807                 offset += sizeof(dummy_csum);
2808                 if (offset < sbi->s_desc_size)
2809                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2810                                              sbi->s_desc_size - offset);
2811
2812                 crc = csum32 & 0xFFFF;
2813                 goto out;
2814         }
2815
2816         /* old crc16 code */
2817         if (!ext4_has_feature_gdt_csum(sb))
2818                 return 0;
2819
2820         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2821         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2822         crc = crc16(crc, (__u8 *)gdp, offset);
2823         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2824         /* for checksum of struct ext4_group_desc do the rest...*/
2825         if (ext4_has_feature_64bit(sb) &&
2826             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2827                 crc = crc16(crc, (__u8 *)gdp + offset,
2828                             le16_to_cpu(sbi->s_es->s_desc_size) -
2829                                 offset);
2830
2831 out:
2832         return cpu_to_le16(crc);
2833 }
2834
2835 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2836                                 struct ext4_group_desc *gdp)
2837 {
2838         if (ext4_has_group_desc_csum(sb) &&
2839             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2840                 return 0;
2841
2842         return 1;
2843 }
2844
2845 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2846                               struct ext4_group_desc *gdp)
2847 {
2848         if (!ext4_has_group_desc_csum(sb))
2849                 return;
2850         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2851 }
2852
2853 /* Called at mount-time, super-block is locked */
2854 static int ext4_check_descriptors(struct super_block *sb,
2855                                   ext4_fsblk_t sb_block,
2856                                   ext4_group_t *first_not_zeroed)
2857 {
2858         struct ext4_sb_info *sbi = EXT4_SB(sb);
2859         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2860         ext4_fsblk_t last_block;
2861         ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0);
2862         ext4_fsblk_t block_bitmap;
2863         ext4_fsblk_t inode_bitmap;
2864         ext4_fsblk_t inode_table;
2865         int flexbg_flag = 0;
2866         ext4_group_t i, grp = sbi->s_groups_count;
2867
2868         if (ext4_has_feature_flex_bg(sb))
2869                 flexbg_flag = 1;
2870
2871         ext4_debug("Checking group descriptors");
2872
2873         for (i = 0; i < sbi->s_groups_count; i++) {
2874                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2875
2876                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2877                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2878                 else
2879                         last_block = first_block +
2880                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2881
2882                 if ((grp == sbi->s_groups_count) &&
2883                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2884                         grp = i;
2885
2886                 block_bitmap = ext4_block_bitmap(sb, gdp);
2887                 if (block_bitmap == sb_block) {
2888                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2889                                  "Block bitmap for group %u overlaps "
2890                                  "superblock", i);
2891                         if (!sb_rdonly(sb))
2892                                 return 0;
2893                 }
2894                 if (block_bitmap >= sb_block + 1 &&
2895                     block_bitmap <= last_bg_block) {
2896                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2897                                  "Block bitmap for group %u overlaps "
2898                                  "block group descriptors", i);
2899                         if (!sb_rdonly(sb))
2900                                 return 0;
2901                 }
2902                 if (block_bitmap < first_block || block_bitmap > last_block) {
2903                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2904                                "Block bitmap for group %u not in group "
2905                                "(block %llu)!", i, block_bitmap);
2906                         return 0;
2907                 }
2908                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2909                 if (inode_bitmap == sb_block) {
2910                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2911                                  "Inode bitmap for group %u overlaps "
2912                                  "superblock", i);
2913                         if (!sb_rdonly(sb))
2914                                 return 0;
2915                 }
2916                 if (inode_bitmap >= sb_block + 1 &&
2917                     inode_bitmap <= last_bg_block) {
2918                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2919                                  "Inode bitmap for group %u overlaps "
2920                                  "block group descriptors", i);
2921                         if (!sb_rdonly(sb))
2922                                 return 0;
2923                 }
2924                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2925                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2926                                "Inode bitmap for group %u not in group "
2927                                "(block %llu)!", i, inode_bitmap);
2928                         return 0;
2929                 }
2930                 inode_table = ext4_inode_table(sb, gdp);
2931                 if (inode_table == sb_block) {
2932                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2933                                  "Inode table for group %u overlaps "
2934                                  "superblock", i);
2935                         if (!sb_rdonly(sb))
2936                                 return 0;
2937                 }
2938                 if (inode_table >= sb_block + 1 &&
2939                     inode_table <= last_bg_block) {
2940                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2941                                  "Inode table for group %u overlaps "
2942                                  "block group descriptors", i);
2943                         if (!sb_rdonly(sb))
2944                                 return 0;
2945                 }
2946                 if (inode_table < first_block ||
2947                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2948                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2949                                "Inode table for group %u not in group "
2950                                "(block %llu)!", i, inode_table);
2951                         return 0;
2952                 }
2953                 ext4_lock_group(sb, i);
2954                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2955                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2956                                  "Checksum for group %u failed (%u!=%u)",
2957                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2958                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2959                         if (!sb_rdonly(sb)) {
2960                                 ext4_unlock_group(sb, i);
2961                                 return 0;
2962                         }
2963                 }
2964                 ext4_unlock_group(sb, i);
2965                 if (!flexbg_flag)
2966                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2967         }
2968         if (NULL != first_not_zeroed)
2969                 *first_not_zeroed = grp;
2970         return 1;
2971 }
2972
2973 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2974  * the superblock) which were deleted from all directories, but held open by
2975  * a process at the time of a crash.  We walk the list and try to delete these
2976  * inodes at recovery time (only with a read-write filesystem).
2977  *
2978  * In order to keep the orphan inode chain consistent during traversal (in
2979  * case of crash during recovery), we link each inode into the superblock
2980  * orphan list_head and handle it the same way as an inode deletion during
2981  * normal operation (which journals the operations for us).
2982  *
2983  * We only do an iget() and an iput() on each inode, which is very safe if we
2984  * accidentally point at an in-use or already deleted inode.  The worst that
2985  * can happen in this case is that we get a "bit already cleared" message from
2986  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2987  * e2fsck was run on this filesystem, and it must have already done the orphan
2988  * inode cleanup for us, so we can safely abort without any further action.
2989  */
2990 static void ext4_orphan_cleanup(struct super_block *sb,
2991                                 struct ext4_super_block *es)
2992 {
2993         unsigned int s_flags = sb->s_flags;
2994         int ret, nr_orphans = 0, nr_truncates = 0;
2995 #ifdef CONFIG_QUOTA
2996         int quota_update = 0;
2997         int i;
2998 #endif
2999         if (!es->s_last_orphan) {
3000                 jbd_debug(4, "no orphan inodes to clean up\n");
3001                 return;
3002         }
3003
3004         if (bdev_read_only(sb->s_bdev)) {
3005                 ext4_msg(sb, KERN_ERR, "write access "
3006                         "unavailable, skipping orphan cleanup");
3007                 return;
3008         }
3009
3010         /* Check if feature set would not allow a r/w mount */
3011         if (!ext4_feature_set_ok(sb, 0)) {
3012                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
3013                          "unknown ROCOMPAT features");
3014                 return;
3015         }
3016
3017         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
3018                 /* don't clear list on RO mount w/ errors */
3019                 if (es->s_last_orphan && !(s_flags & SB_RDONLY)) {
3020                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
3021                                   "clearing orphan list.\n");
3022                         es->s_last_orphan = 0;
3023                 }
3024                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
3025                 return;
3026         }
3027
3028         if (s_flags & SB_RDONLY) {
3029                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
3030                 sb->s_flags &= ~SB_RDONLY;
3031         }
3032 #ifdef CONFIG_QUOTA
3033         /*
3034          * Turn on quotas which were not enabled for read-only mounts if
3035          * filesystem has quota feature, so that they are updated correctly.
3036          */
3037         if (ext4_has_feature_quota(sb) && (s_flags & SB_RDONLY)) {
3038                 int ret = ext4_enable_quotas(sb);
3039
3040                 if (!ret)
3041                         quota_update = 1;
3042                 else
3043                         ext4_msg(sb, KERN_ERR,
3044                                 "Cannot turn on quotas: error %d", ret);
3045         }
3046
3047         /* Turn on journaled quotas used for old sytle */
3048         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
3049                 if (EXT4_SB(sb)->s_qf_names[i]) {
3050                         int ret = ext4_quota_on_mount(sb, i);
3051
3052                         if (!ret)
3053                                 quota_update = 1;
3054                         else
3055                                 ext4_msg(sb, KERN_ERR,
3056                                         "Cannot turn on journaled "
3057                                         "quota: type %d: error %d", i, ret);
3058                 }
3059         }
3060 #endif
3061
3062         while (es->s_last_orphan) {
3063                 struct inode *inode;
3064
3065                 /*
3066                  * We may have encountered an error during cleanup; if
3067                  * so, skip the rest.
3068                  */
3069                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
3070                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
3071                         es->s_last_orphan = 0;
3072                         break;
3073                 }
3074
3075                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
3076                 if (IS_ERR(inode)) {
3077                         es->s_last_orphan = 0;
3078                         break;
3079                 }
3080
3081                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
3082                 dquot_initialize(inode);
3083                 if (inode->i_nlink) {
3084                         if (test_opt(sb, DEBUG))
3085                                 ext4_msg(sb, KERN_DEBUG,
3086                                         "%s: truncating inode %lu to %lld bytes",
3087                                         __func__, inode->i_ino, inode->i_size);
3088                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
3089                                   inode->i_ino, inode->i_size);
3090                         inode_lock(inode);
3091                         truncate_inode_pages(inode->i_mapping, inode->i_size);
3092                         ret = ext4_truncate(inode);
3093                         if (ret) {
3094                                 /*
3095                                  * We need to clean up the in-core orphan list
3096                                  * manually if ext4_truncate() failed to get a
3097                                  * transaction handle.
3098                                  */
3099                                 ext4_orphan_del(NULL, inode);
3100                                 ext4_std_error(inode->i_sb, ret);
3101                         }
3102                         inode_unlock(inode);
3103                         nr_truncates++;
3104                 } else {
3105                         if (test_opt(sb, DEBUG))
3106                                 ext4_msg(sb, KERN_DEBUG,
3107                                         "%s: deleting unreferenced inode %lu",
3108                                         __func__, inode->i_ino);
3109                         jbd_debug(2, "deleting unreferenced inode %lu\n",
3110                                   inode->i_ino);
3111                         nr_orphans++;
3112                 }
3113                 iput(inode);  /* The delete magic happens here! */
3114         }
3115
3116 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
3117
3118         if (nr_orphans)
3119                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
3120                        PLURAL(nr_orphans));
3121         if (nr_truncates)
3122                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
3123                        PLURAL(nr_truncates));
3124 #ifdef CONFIG_QUOTA
3125         /* Turn off quotas if they were enabled for orphan cleanup */
3126         if (quota_update) {
3127                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
3128                         if (sb_dqopt(sb)->files[i])
3129                                 dquot_quota_off(sb, i);
3130                 }
3131         }
3132 #endif
3133         sb->s_flags = s_flags; /* Restore SB_RDONLY status */
3134 }
3135
3136 /*
3137  * Maximal extent format file size.
3138  * Resulting logical blkno at s_maxbytes must fit in our on-disk
3139  * extent format containers, within a sector_t, and within i_blocks
3140  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
3141  * so that won't be a limiting factor.
3142  *
3143  * However there is other limiting factor. We do store extents in the form
3144  * of starting block and length, hence the resulting length of the extent
3145  * covering maximum file size must fit into on-disk format containers as
3146  * well. Given that length is always by 1 unit bigger than max unit (because
3147  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
3148  *
3149  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
3150  */
3151 static loff_t ext4_max_size(int blkbits, int has_huge_files)
3152 {
3153         loff_t res;
3154         loff_t upper_limit = MAX_LFS_FILESIZE;
3155
3156         BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64));
3157
3158         if (!has_huge_files) {
3159                 upper_limit = (1LL << 32) - 1;
3160
3161                 /* total blocks in file system block size */
3162                 upper_limit >>= (blkbits - 9);
3163                 upper_limit <<= blkbits;
3164         }
3165
3166         /*
3167          * 32-bit extent-start container, ee_block. We lower the maxbytes
3168          * by one fs block, so ee_len can cover the extent of maximum file
3169          * size
3170          */
3171         res = (1LL << 32) - 1;
3172         res <<= blkbits;
3173
3174         /* Sanity check against vm- & vfs- imposed limits */
3175         if (res > upper_limit)
3176                 res = upper_limit;
3177
3178         return res;
3179 }
3180
3181 /*
3182  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
3183  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
3184  * We need to be 1 filesystem block less than the 2^48 sector limit.
3185  */
3186 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
3187 {
3188         loff_t res = EXT4_NDIR_BLOCKS;
3189         int meta_blocks;
3190         loff_t upper_limit;
3191         /* This is calculated to be the largest file size for a dense, block
3192          * mapped file such that the file's total number of 512-byte sectors,
3193          * including data and all indirect blocks, does not exceed (2^48 - 1).
3194          *
3195          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
3196          * number of 512-byte sectors of the file.
3197          */
3198
3199         if (!has_huge_files) {
3200                 /*
3201                  * !has_huge_files or implies that the inode i_block field
3202                  * represents total file blocks in 2^32 512-byte sectors ==
3203                  * size of vfs inode i_blocks * 8
3204                  */
3205                 upper_limit = (1LL << 32) - 1;
3206
3207                 /* total blocks in file system block size */
3208                 upper_limit >>= (bits - 9);
3209
3210         } else {
3211                 /*
3212                  * We use 48 bit ext4_inode i_blocks
3213                  * With EXT4_HUGE_FILE_FL set the i_blocks
3214                  * represent total number of blocks in
3215                  * file system block size
3216                  */
3217                 upper_limit = (1LL << 48) - 1;
3218
3219         }
3220
3221         /* indirect blocks */
3222         meta_blocks = 1;
3223         /* double indirect blocks */
3224         meta_blocks += 1 + (1LL << (bits-2));
3225         /* tripple indirect blocks */
3226         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
3227
3228         upper_limit -= meta_blocks;
3229         upper_limit <<= bits;
3230
3231         res += 1LL << (bits-2);
3232         res += 1LL << (2*(bits-2));
3233         res += 1LL << (3*(bits-2));
3234         res <<= bits;
3235         if (res > upper_limit)
3236                 res = upper_limit;
3237
3238         if (res > MAX_LFS_FILESIZE)
3239                 res = MAX_LFS_FILESIZE;
3240
3241         return res;
3242 }
3243
3244 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
3245                                    ext4_fsblk_t logical_sb_block, int nr)
3246 {
3247         struct ext4_sb_info *sbi = EXT4_SB(sb);
3248         ext4_group_t bg, first_meta_bg;
3249         int has_super = 0;
3250
3251         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
3252
3253         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
3254                 return logical_sb_block + nr + 1;
3255         bg = sbi->s_desc_per_block * nr;
3256         if (ext4_bg_has_super(sb, bg))
3257                 has_super = 1;
3258
3259         /*
3260          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
3261          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
3262          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
3263          * compensate.
3264          */
3265         if (sb->s_blocksize == 1024 && nr == 0 &&
3266             le32_to_cpu(sbi->s_es->s_first_data_block) == 0)
3267                 has_super++;
3268
3269         return (has_super + ext4_group_first_block_no(sb, bg));
3270 }
3271
3272 /**
3273  * ext4_get_stripe_size: Get the stripe size.
3274  * @sbi: In memory super block info
3275  *
3276  * If we have specified it via mount option, then
3277  * use the mount option value. If the value specified at mount time is
3278  * greater than the blocks per group use the super block value.
3279  * If the super block value is greater than blocks per group return 0.
3280  * Allocator needs it be less than blocks per group.
3281  *
3282  */
3283 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
3284 {
3285         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
3286         unsigned long stripe_width =
3287                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
3288         int ret;
3289
3290         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
3291                 ret = sbi->s_stripe;
3292         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
3293                 ret = stripe_width;
3294         else if (stride && stride <= sbi->s_blocks_per_group)
3295                 ret = stride;
3296         else
3297                 ret = 0;
3298
3299         /*
3300          * If the stripe width is 1, this makes no sense and
3301          * we set it to 0 to turn off stripe handling code.
3302          */
3303         if (ret <= 1)
3304                 ret = 0;
3305
3306         return ret;
3307 }
3308
3309 /*
3310  * Check whether this filesystem can be mounted based on
3311  * the features present and the RDONLY/RDWR mount requested.
3312  * Returns 1 if this filesystem can be mounted as requested,
3313  * 0 if it cannot be.
3314  */
3315 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
3316 {
3317         if (ext4_has_unknown_ext4_incompat_features(sb)) {
3318                 ext4_msg(sb, KERN_ERR,
3319                         "Couldn't mount because of "
3320                         "unsupported optional features (%x)",
3321                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
3322                         ~EXT4_FEATURE_INCOMPAT_SUPP));
3323                 return 0;
3324         }
3325
3326 #ifndef CONFIG_UNICODE
3327         if (ext4_has_feature_casefold(sb)) {
3328                 ext4_msg(sb, KERN_ERR,
3329                          "Filesystem with casefold feature cannot be "
3330                          "mounted without CONFIG_UNICODE");
3331                 return 0;
3332         }
3333 #endif
3334
3335         if (readonly)
3336                 return 1;
3337
3338         if (ext4_has_feature_readonly(sb)) {
3339                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
3340                 sb->s_flags |= SB_RDONLY;
3341                 return 1;
3342         }
3343
3344         /* Check that feature set is OK for a read-write mount */
3345         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
3346                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
3347                          "unsupported optional features (%x)",
3348                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
3349                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
3350                 return 0;
3351         }
3352         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
3353                 ext4_msg(sb, KERN_ERR,
3354                          "Can't support bigalloc feature without "
3355                          "extents feature\n");
3356                 return 0;
3357         }
3358
3359 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2)
3360         if (!readonly && (ext4_has_feature_quota(sb) ||
3361                           ext4_has_feature_project(sb))) {
3362                 ext4_msg(sb, KERN_ERR,
3363                          "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2");
3364                 return 0;
3365         }
3366 #endif  /* CONFIG_QUOTA */
3367         return 1;
3368 }
3369
3370 /*
3371  * This function is called once a day if we have errors logged
3372  * on the file system
3373  */
3374 static void print_daily_error_info(struct timer_list *t)
3375 {
3376         struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report);
3377         struct super_block *sb = sbi->s_sb;
3378         struct ext4_super_block *es = sbi->s_es;
3379
3380         if (es->s_error_count)
3381                 /* fsck newer than v1.41.13 is needed to clean this condition. */
3382                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
3383                          le32_to_cpu(es->s_error_count));
3384         if (es->s_first_error_time) {
3385                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d",
3386                        sb->s_id,
3387                        ext4_get_tstamp(es, s_first_error_time),
3388                        (int) sizeof(es->s_first_error_func),
3389                        es->s_first_error_func,
3390                        le32_to_cpu(es->s_first_error_line));
3391                 if (es->s_first_error_ino)
3392                         printk(KERN_CONT ": inode %u",
3393                                le32_to_cpu(es->s_first_error_ino));
3394                 if (es->s_first_error_block)
3395                         printk(KERN_CONT ": block %llu", (unsigned long long)
3396                                le64_to_cpu(es->s_first_error_block));
3397                 printk(KERN_CONT "\n");
3398         }
3399         if (es->s_last_error_time) {
3400                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d",
3401                        sb->s_id,
3402                        ext4_get_tstamp(es, s_last_error_time),
3403                        (int) sizeof(es->s_last_error_func),
3404                        es->s_last_error_func,
3405                        le32_to_cpu(es->s_last_error_line));
3406                 if (es->s_last_error_ino)
3407                         printk(KERN_CONT ": inode %u",
3408                                le32_to_cpu(es->s_last_error_ino));
3409                 if (es->s_last_error_block)
3410                         printk(KERN_CONT ": block %llu", (unsigned long long)
3411                                le64_to_cpu(es->s_last_error_block));
3412                 printk(KERN_CONT "\n");
3413         }
3414         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
3415 }
3416
3417 /* Find next suitable group and run ext4_init_inode_table */
3418 static int ext4_run_li_request(struct ext4_li_request *elr)
3419 {
3420         struct ext4_group_desc *gdp = NULL;
3421         struct super_block *sb = elr->lr_super;
3422         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3423         ext4_group_t group = elr->lr_next_group;
3424         unsigned long timeout = 0;
3425         unsigned int prefetch_ios = 0;
3426         int ret = 0;
3427
3428         if (elr->lr_mode == EXT4_LI_MODE_PREFETCH_BBITMAP) {
3429                 elr->lr_next_group = ext4_mb_prefetch(sb, group,
3430                                 EXT4_SB(sb)->s_mb_prefetch, &prefetch_ios);
3431                 if (prefetch_ios)
3432                         ext4_mb_prefetch_fini(sb, elr->lr_next_group,
3433                                               prefetch_ios);
3434                 trace_ext4_prefetch_bitmaps(sb, group, elr->lr_next_group,
3435                                             prefetch_ios);
3436                 if (group >= elr->lr_next_group) {
3437                         ret = 1;
3438                         if (elr->lr_first_not_zeroed != ngroups &&
3439                             !sb_rdonly(sb) && test_opt(sb, INIT_INODE_TABLE)) {
3440                                 elr->lr_next_group = elr->lr_first_not_zeroed;
3441                                 elr->lr_mode = EXT4_LI_MODE_ITABLE;
3442                                 ret = 0;
3443                         }
3444                 }
3445                 return ret;
3446         }
3447
3448         for (; group < ngroups; group++) {
3449                 gdp = ext4_get_group_desc(sb, group, NULL);
3450                 if (!gdp) {
3451                         ret = 1;
3452                         break;
3453                 }
3454
3455                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3456                         break;
3457         }
3458
3459         if (group >= ngroups)
3460                 ret = 1;
3461
3462         if (!ret) {
3463                 timeout = jiffies;
3464                 ret = ext4_init_inode_table(sb, group,
3465                                             elr->lr_timeout ? 0 : 1);
3466                 trace_ext4_lazy_itable_init(sb, group);
3467                 if (elr->lr_timeout == 0) {
3468                         timeout = (jiffies - timeout) *
3469                                 EXT4_SB(elr->lr_super)->s_li_wait_mult;
3470                         elr->lr_timeout = timeout;
3471                 }
3472                 elr->lr_next_sched = jiffies + elr->lr_timeout;
3473                 elr->lr_next_group = group + 1;
3474         }
3475         return ret;
3476 }
3477
3478 /*
3479  * Remove lr_request from the list_request and free the
3480  * request structure. Should be called with li_list_mtx held
3481  */
3482 static void ext4_remove_li_request(struct ext4_li_request *elr)
3483 {
3484         if (!elr)
3485                 return;
3486
3487         list_del(&elr->lr_request);
3488         EXT4_SB(elr->lr_super)->s_li_request = NULL;
3489         kfree(elr);
3490 }
3491
3492 static void ext4_unregister_li_request(struct super_block *sb)
3493 {
3494         mutex_lock(&ext4_li_mtx);
3495         if (!ext4_li_info) {
3496                 mutex_unlock(&ext4_li_mtx);
3497                 return;
3498         }
3499
3500         mutex_lock(&ext4_li_info->li_list_mtx);
3501         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
3502         mutex_unlock(&ext4_li_info->li_list_mtx);
3503         mutex_unlock(&ext4_li_mtx);
3504 }
3505
3506 static struct task_struct *ext4_lazyinit_task;
3507
3508 /*
3509  * This is the function where ext4lazyinit thread lives. It walks
3510  * through the request list searching for next scheduled filesystem.
3511  * When such a fs is found, run the lazy initialization request
3512  * (ext4_rn_li_request) and keep track of the time spend in this
3513  * function. Based on that time we compute next schedule time of
3514  * the request. When walking through the list is complete, compute
3515  * next waking time and put itself into sleep.
3516  */
3517 static int ext4_lazyinit_thread(void *arg)
3518 {
3519         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
3520         struct list_head *pos, *n;
3521         struct ext4_li_request *elr;
3522         unsigned long next_wakeup, cur;
3523
3524         BUG_ON(NULL == eli);
3525
3526 cont_thread:
3527         while (true) {
3528                 next_wakeup = MAX_JIFFY_OFFSET;
3529
3530                 mutex_lock(&eli->li_list_mtx);
3531                 if (list_empty(&eli->li_request_list)) {
3532                         mutex_unlock(&eli->li_list_mtx);
3533                         goto exit_thread;
3534                 }
3535                 list_for_each_safe(pos, n, &eli->li_request_list) {
3536                         int err = 0;
3537                         int progress = 0;
3538                         elr = list_entry(pos, struct ext4_li_request,
3539                                          lr_request);
3540
3541                         if (time_before(jiffies, elr->lr_next_sched)) {
3542                                 if (time_before(elr->lr_next_sched, next_wakeup))
3543                                         next_wakeup = elr->lr_next_sched;
3544                                 continue;
3545                         }
3546                         if (down_read_trylock(&elr->lr_super->s_umount)) {
3547                                 if (sb_start_write_trylock(elr->lr_super)) {
3548                                         progress = 1;
3549                                         /*
3550                                          * We hold sb->s_umount, sb can not
3551                                          * be removed from the list, it is
3552                                          * now safe to drop li_list_mtx
3553                                          */
3554                                         mutex_unlock(&eli->li_list_mtx);
3555                                         err = ext4_run_li_request(elr);
3556                                         sb_end_write(elr->lr_super);
3557                                         mutex_lock(&eli->li_list_mtx);
3558                                         n = pos->next;
3559                                 }
3560                                 up_read((&elr->lr_super->s_umount));
3561                         }
3562                         /* error, remove the lazy_init job */
3563                         if (err) {
3564                                 ext4_remove_li_request(elr);
3565                                 continue;
3566                         }
3567                         if (!progress) {
3568                                 elr->lr_next_sched = jiffies +
3569                                         (prandom_u32()
3570                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3571                         }
3572                         if (time_before(elr->lr_next_sched, next_wakeup))
3573                                 next_wakeup = elr->lr_next_sched;
3574                 }
3575                 mutex_unlock(&eli->li_list_mtx);
3576
3577                 try_to_freeze();
3578
3579                 cur = jiffies;
3580                 if ((time_after_eq(cur, next_wakeup)) ||
3581                     (MAX_JIFFY_OFFSET == next_wakeup)) {
3582                         cond_resched();
3583                         continue;
3584                 }
3585
3586                 schedule_timeout_interruptible(next_wakeup - cur);
3587
3588                 if (kthread_should_stop()) {
3589                         ext4_clear_request_list();
3590                         goto exit_thread;
3591                 }
3592         }
3593
3594 exit_thread:
3595         /*
3596          * It looks like the request list is empty, but we need
3597          * to check it under the li_list_mtx lock, to prevent any
3598          * additions into it, and of course we should lock ext4_li_mtx
3599          * to atomically free the list and ext4_li_info, because at
3600          * this point another ext4 filesystem could be registering
3601          * new one.
3602          */
3603         mutex_lock(&ext4_li_mtx);
3604         mutex_lock(&eli->li_list_mtx);
3605         if (!list_empty(&eli->li_request_list)) {
3606                 mutex_unlock(&eli->li_list_mtx);
3607                 mutex_unlock(&ext4_li_mtx);
3608                 goto cont_thread;
3609         }
3610         mutex_unlock(&eli->li_list_mtx);
3611         kfree(ext4_li_info);
3612         ext4_li_info = NULL;
3613         mutex_unlock(&ext4_li_mtx);
3614
3615         return 0;
3616 }
3617
3618 static void ext4_clear_request_list(void)
3619 {
3620         struct list_head *pos, *n;
3621         struct ext4_li_request *elr;
3622
3623         mutex_lock(&ext4_li_info->li_list_mtx);
3624         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3625                 elr = list_entry(pos, struct ext4_li_request,
3626                                  lr_request);
3627                 ext4_remove_li_request(elr);
3628         }
3629         mutex_unlock(&ext4_li_info->li_list_mtx);
3630 }
3631
3632 static int ext4_run_lazyinit_thread(void)
3633 {
3634         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3635                                          ext4_li_info, "ext4lazyinit");
3636         if (IS_ERR(ext4_lazyinit_task)) {
3637                 int err = PTR_ERR(ext4_lazyinit_task);
3638                 ext4_clear_request_list();
3639                 kfree(ext4_li_info);
3640                 ext4_li_info = NULL;
3641                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3642                                  "initialization thread\n",
3643                                  err);
3644                 return err;
3645         }
3646         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3647         return 0;
3648 }
3649
3650 /*
3651  * Check whether it make sense to run itable init. thread or not.
3652  * If there is at least one uninitialized inode table, return
3653  * corresponding group number, else the loop goes through all
3654  * groups and return total number of groups.
3655  */
3656 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3657 {
3658         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3659         struct ext4_group_desc *gdp = NULL;
3660
3661         if (!ext4_has_group_desc_csum(sb))
3662                 return ngroups;
3663
3664         for (group = 0; group < ngroups; group++) {
3665                 gdp = ext4_get_group_desc(sb, group, NULL);
3666                 if (!gdp)
3667                         continue;
3668
3669                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3670                         break;
3671         }
3672
3673         return group;
3674 }
3675
3676 static int ext4_li_info_new(void)
3677 {
3678         struct ext4_lazy_init *eli = NULL;
3679
3680         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3681         if (!eli)
3682                 return -ENOMEM;
3683
3684         INIT_LIST_HEAD(&eli->li_request_list);
3685         mutex_init(&eli->li_list_mtx);
3686
3687         eli->li_state |= EXT4_LAZYINIT_QUIT;
3688
3689         ext4_li_info = eli;
3690
3691         return 0;
3692 }
3693
3694 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3695                                             ext4_group_t start)
3696 {
3697         struct ext4_li_request *elr;
3698
3699         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3700         if (!elr)
3701                 return NULL;
3702
3703         elr->lr_super = sb;
3704         elr->lr_first_not_zeroed = start;
3705         if (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS)) {
3706                 elr->lr_mode = EXT4_LI_MODE_ITABLE;
3707                 elr->lr_next_group = start;
3708         } else {
3709                 elr->lr_mode = EXT4_LI_MODE_PREFETCH_BBITMAP;
3710         }
3711
3712         /*
3713          * Randomize first schedule time of the request to
3714          * spread the inode table initialization requests
3715          * better.
3716          */
3717         elr->lr_next_sched = jiffies + (prandom_u32() %
3718                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3719         return elr;
3720 }
3721
3722 int ext4_register_li_request(struct super_block *sb,
3723                              ext4_group_t first_not_zeroed)
3724 {
3725         struct ext4_sb_info *sbi = EXT4_SB(sb);
3726         struct ext4_li_request *elr = NULL;
3727         ext4_group_t ngroups = sbi->s_groups_count;
3728         int ret = 0;
3729
3730         mutex_lock(&ext4_li_mtx);
3731         if (sbi->s_li_request != NULL) {
3732                 /*
3733                  * Reset timeout so it can be computed again, because
3734                  * s_li_wait_mult might have changed.
3735                  */
3736                 sbi->s_li_request->lr_timeout = 0;
3737                 goto out;
3738         }
3739
3740         if (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS) &&
3741             (first_not_zeroed == ngroups || sb_rdonly(sb) ||
3742              !test_opt(sb, INIT_INODE_TABLE)))
3743                 goto out;
3744
3745         elr = ext4_li_request_new(sb, first_not_zeroed);
3746         if (!elr) {
3747                 ret = -ENOMEM;
3748                 goto out;
3749         }
3750
3751         if (NULL == ext4_li_info) {
3752                 ret = ext4_li_info_new();
3753                 if (ret)
3754                         goto out;
3755         }
3756
3757         mutex_lock(&ext4_li_info->li_list_mtx);
3758         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3759         mutex_unlock(&ext4_li_info->li_list_mtx);
3760
3761         sbi->s_li_request = elr;
3762         /*
3763          * set elr to NULL here since it has been inserted to
3764          * the request_list and the removal and free of it is
3765          * handled by ext4_clear_request_list from now on.
3766          */
3767         elr = NULL;
3768
3769         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3770                 ret = ext4_run_lazyinit_thread();
3771                 if (ret)
3772                         goto out;
3773         }
3774 out:
3775         mutex_unlock(&ext4_li_mtx);
3776         if (ret)
3777                 kfree(elr);
3778         return ret;
3779 }
3780
3781 /*
3782  * We do not need to lock anything since this is called on
3783  * module unload.
3784  */
3785 static void ext4_destroy_lazyinit_thread(void)
3786 {
3787         /*
3788          * If thread exited earlier
3789          * there's nothing to be done.
3790          */
3791         if (!ext4_li_info || !ext4_lazyinit_task)
3792                 return;
3793
3794         kthread_stop(ext4_lazyinit_task);
3795 }
3796
3797 static int set_journal_csum_feature_set(struct super_block *sb)
3798 {
3799         int ret = 1;
3800         int compat, incompat;
3801         struct ext4_sb_info *sbi = EXT4_SB(sb);
3802
3803         if (ext4_has_metadata_csum(sb)) {
3804                 /* journal checksum v3 */
3805                 compat = 0;
3806                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3807         } else {
3808                 /* journal checksum v1 */
3809                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3810                 incompat = 0;
3811         }
3812
3813         jbd2_journal_clear_features(sbi->s_journal,
3814                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3815                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3816                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3817         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3818                 ret = jbd2_journal_set_features(sbi->s_journal,
3819                                 compat, 0,
3820                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3821                                 incompat);
3822         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3823                 ret = jbd2_journal_set_features(sbi->s_journal,
3824                                 compat, 0,
3825                                 incompat);
3826                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3827                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3828         } else {
3829                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3830                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3831         }
3832
3833         return ret;
3834 }
3835
3836 /*
3837  * Note: calculating the overhead so we can be compatible with
3838  * historical BSD practice is quite difficult in the face of
3839  * clusters/bigalloc.  This is because multiple metadata blocks from
3840  * different block group can end up in the same allocation cluster.
3841  * Calculating the exact overhead in the face of clustered allocation
3842  * requires either O(all block bitmaps) in memory or O(number of block
3843  * groups**2) in time.  We will still calculate the superblock for
3844  * older file systems --- and if we come across with a bigalloc file
3845  * system with zero in s_overhead_clusters the estimate will be close to
3846  * correct especially for very large cluster sizes --- but for newer
3847  * file systems, it's better to calculate this figure once at mkfs
3848  * time, and store it in the superblock.  If the superblock value is
3849  * present (even for non-bigalloc file systems), we will use it.
3850  */
3851 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3852                           char *buf)
3853 {
3854         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3855         struct ext4_group_desc  *gdp;
3856         ext4_fsblk_t            first_block, last_block, b;
3857         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3858         int                     s, j, count = 0;
3859
3860         if (!ext4_has_feature_bigalloc(sb))
3861                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3862                         sbi->s_itb_per_group + 2);
3863
3864         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3865                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3866         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3867         for (i = 0; i < ngroups; i++) {
3868                 gdp = ext4_get_group_desc(sb, i, NULL);
3869                 b = ext4_block_bitmap(sb, gdp);
3870                 if (b >= first_block && b <= last_block) {
3871                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3872                         count++;
3873                 }
3874                 b = ext4_inode_bitmap(sb, gdp);
3875                 if (b >= first_block && b <= last_block) {
3876                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3877                         count++;
3878                 }
3879                 b = ext4_inode_table(sb, gdp);
3880                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3881                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3882                                 int c = EXT4_B2C(sbi, b - first_block);
3883                                 ext4_set_bit(c, buf);
3884                                 count++;
3885                         }
3886                 if (i != grp)
3887                         continue;
3888                 s = 0;
3889                 if (ext4_bg_has_super(sb, grp)) {
3890                         ext4_set_bit(s++, buf);
3891                         count++;
3892                 }
3893                 j = ext4_bg_num_gdb(sb, grp);
3894                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3895                         ext4_error(sb, "Invalid number of block group "
3896                                    "descriptor blocks: %d", j);
3897                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3898                 }
3899                 count += j;
3900                 for (; j > 0; j--)
3901                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3902         }
3903         if (!count)
3904                 return 0;
3905         return EXT4_CLUSTERS_PER_GROUP(sb) -
3906                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3907 }
3908
3909 /*
3910  * Compute the overhead and stash it in sbi->s_overhead
3911  */
3912 int ext4_calculate_overhead(struct super_block *sb)
3913 {
3914         struct ext4_sb_info *sbi = EXT4_SB(sb);
3915         struct ext4_super_block *es = sbi->s_es;
3916         struct inode *j_inode;
3917         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3918         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3919         ext4_fsblk_t overhead = 0;
3920         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3921
3922         if (!buf)
3923                 return -ENOMEM;
3924
3925         /*
3926          * Compute the overhead (FS structures).  This is constant
3927          * for a given filesystem unless the number of block groups
3928          * changes so we cache the previous value until it does.
3929          */
3930
3931         /*
3932          * All of the blocks before first_data_block are overhead
3933          */
3934         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3935
3936         /*
3937          * Add the overhead found in each block group
3938          */
3939         for (i = 0; i < ngroups; i++) {
3940                 int blks;
3941
3942                 blks = count_overhead(sb, i, buf);
3943                 overhead += blks;
3944                 if (blks)
3945                         memset(buf, 0, PAGE_SIZE);
3946                 cond_resched();
3947         }
3948
3949         /*
3950          * Add the internal journal blocks whether the journal has been
3951          * loaded or not
3952          */
3953         if (sbi->s_journal && !sbi->s_journal_bdev)
3954                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_total_len);
3955         else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) {
3956                 /* j_inum for internal journal is non-zero */
3957                 j_inode = ext4_get_journal_inode(sb, j_inum);
3958                 if (j_inode) {
3959                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3960                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
3961                         iput(j_inode);
3962                 } else {
3963                         ext4_msg(sb, KERN_ERR, "can't get journal size");
3964                 }
3965         }
3966         sbi->s_overhead = overhead;
3967         smp_wmb();
3968         free_page((unsigned long) buf);
3969         return 0;
3970 }
3971
3972 static void ext4_set_resv_clusters(struct super_block *sb)
3973 {
3974         ext4_fsblk_t resv_clusters;
3975         struct ext4_sb_info *sbi = EXT4_SB(sb);
3976
3977         /*
3978          * There's no need to reserve anything when we aren't using extents.
3979          * The space estimates are exact, there are no unwritten extents,
3980          * hole punching doesn't need new metadata... This is needed especially
3981          * to keep ext2/3 backward compatibility.
3982          */
3983         if (!ext4_has_feature_extents(sb))
3984                 return;
3985         /*
3986          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3987          * This should cover the situations where we can not afford to run
3988          * out of space like for example punch hole, or converting
3989          * unwritten extents in delalloc path. In most cases such
3990          * allocation would require 1, or 2 blocks, higher numbers are
3991          * very rare.
3992          */
3993         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3994                          sbi->s_cluster_bits);
3995
3996         do_div(resv_clusters, 50);
3997         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3998
3999         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
4000 }
4001
4002 static const char *ext4_quota_mode(struct super_block *sb)
4003 {
4004 #ifdef CONFIG_QUOTA
4005         if (!ext4_quota_capable(sb))
4006                 return "none";
4007
4008         if (EXT4_SB(sb)->s_journal && ext4_is_quota_journalled(sb))
4009                 return "journalled";
4010         else
4011                 return "writeback";
4012 #else
4013         return "disabled";
4014 #endif
4015 }
4016
4017 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
4018 {
4019         struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev);
4020         char *orig_data = kstrdup(data, GFP_KERNEL);
4021         struct buffer_head *bh, **group_desc;
4022         struct ext4_super_block *es = NULL;
4023         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
4024         struct flex_groups **flex_groups;
4025         ext4_fsblk_t block;
4026         ext4_fsblk_t sb_block = get_sb_block(&data);
4027         ext4_fsblk_t logical_sb_block;
4028         unsigned long offset = 0;
4029         unsigned long def_mount_opts;
4030         struct inode *root;
4031         const char *descr;
4032         int ret = -ENOMEM;
4033         int blocksize, clustersize;
4034         unsigned int db_count;
4035         unsigned int i;
4036         int needs_recovery, has_huge_files;
4037         __u64 blocks_count;
4038         int err = 0;
4039         ext4_group_t first_not_zeroed;
4040         struct ext4_parsed_options parsed_opts;
4041
4042         /* Set defaults for the variables that will be set during parsing */
4043         parsed_opts.journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4044         parsed_opts.journal_devnum = 0;
4045         parsed_opts.mb_optimize_scan = DEFAULT_MB_OPTIMIZE_SCAN;
4046
4047         if ((data && !orig_data) || !sbi)
4048                 goto out_free_base;
4049
4050         sbi->s_daxdev = dax_dev;
4051         sbi->s_blockgroup_lock =
4052                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
4053         if (!sbi->s_blockgroup_lock)
4054                 goto out_free_base;
4055
4056         sb->s_fs_info = sbi;
4057         sbi->s_sb = sb;
4058         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
4059         sbi->s_sb_block = sb_block;
4060         sbi->s_sectors_written_start =
4061                 part_stat_read(sb->s_bdev, sectors[STAT_WRITE]);
4062
4063         /* Cleanup superblock name */
4064         strreplace(sb->s_id, '/', '!');
4065
4066         /* -EINVAL is default */
4067         ret = -EINVAL;
4068         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
4069         if (!blocksize) {
4070                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
4071                 goto out_fail;
4072         }
4073
4074         /*
4075          * The ext4 superblock will not be buffer aligned for other than 1kB
4076          * block sizes.  We need to calculate the offset from buffer start.
4077          */
4078         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
4079                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
4080                 offset = do_div(logical_sb_block, blocksize);
4081         } else {
4082                 logical_sb_block = sb_block;
4083         }
4084
4085         bh = ext4_sb_bread_unmovable(sb, logical_sb_block);
4086         if (IS_ERR(bh)) {
4087                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
4088                 ret = PTR_ERR(bh);
4089                 goto out_fail;
4090         }
4091         /*
4092          * Note: s_es must be initialized as soon as possible because
4093          *       some ext4 macro-instructions depend on its value
4094          */
4095         es = (struct ext4_super_block *) (bh->b_data + offset);
4096         sbi->s_es = es;
4097         sb->s_magic = le16_to_cpu(es->s_magic);
4098         if (sb->s_magic != EXT4_SUPER_MAGIC)
4099                 goto cantfind_ext4;
4100         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
4101
4102         /* Warn if metadata_csum and gdt_csum are both set. */
4103         if (ext4_has_feature_metadata_csum(sb) &&
4104             ext4_has_feature_gdt_csum(sb))
4105                 ext4_warning(sb, "metadata_csum and uninit_bg are "
4106                              "redundant flags; please run fsck.");
4107
4108         /* Check for a known checksum algorithm */
4109         if (!ext4_verify_csum_type(sb, es)) {
4110                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
4111                          "unknown checksum algorithm.");
4112                 silent = 1;
4113                 goto cantfind_ext4;
4114         }
4115
4116         /* Load the checksum driver */
4117         sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
4118         if (IS_ERR(sbi->s_chksum_driver)) {
4119                 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
4120                 ret = PTR_ERR(sbi->s_chksum_driver);
4121                 sbi->s_chksum_driver = NULL;
4122                 goto failed_mount;
4123         }
4124
4125         /* Check superblock checksum */
4126         if (!ext4_superblock_csum_verify(sb, es)) {
4127                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
4128                          "invalid superblock checksum.  Run e2fsck?");
4129                 silent = 1;
4130                 ret = -EFSBADCRC;
4131                 goto cantfind_ext4;
4132         }
4133
4134         /* Precompute checksum seed for all metadata */
4135         if (ext4_has_feature_csum_seed(sb))
4136                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
4137         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
4138                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
4139                                                sizeof(es->s_uuid));
4140
4141         /* Set defaults before we parse the mount options */
4142         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
4143         set_opt(sb, INIT_INODE_TABLE);
4144         if (def_mount_opts & EXT4_DEFM_DEBUG)
4145                 set_opt(sb, DEBUG);
4146         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
4147                 set_opt(sb, GRPID);
4148         if (def_mount_opts & EXT4_DEFM_UID16)
4149                 set_opt(sb, NO_UID32);
4150         /* xattr user namespace & acls are now defaulted on */
4151         set_opt(sb, XATTR_USER);
4152 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4153         set_opt(sb, POSIX_ACL);
4154 #endif
4155         if (ext4_has_feature_fast_commit(sb))
4156                 set_opt2(sb, JOURNAL_FAST_COMMIT);
4157         /* don't forget to enable journal_csum when metadata_csum is enabled. */
4158         if (ext4_has_metadata_csum(sb))
4159                 set_opt(sb, JOURNAL_CHECKSUM);
4160
4161         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
4162                 set_opt(sb, JOURNAL_DATA);
4163         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
4164                 set_opt(sb, ORDERED_DATA);
4165         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
4166                 set_opt(sb, WRITEBACK_DATA);
4167
4168         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
4169                 set_opt(sb, ERRORS_PANIC);
4170         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
4171                 set_opt(sb, ERRORS_CONT);
4172         else
4173                 set_opt(sb, ERRORS_RO);
4174         /* block_validity enabled by default; disable with noblock_validity */
4175         set_opt(sb, BLOCK_VALIDITY);
4176         if (def_mount_opts & EXT4_DEFM_DISCARD)
4177                 set_opt(sb, DISCARD);
4178
4179         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
4180         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
4181         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
4182         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
4183         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
4184
4185         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
4186                 set_opt(sb, BARRIER);
4187
4188         /*
4189          * enable delayed allocation by default
4190          * Use -o nodelalloc to turn it off
4191          */
4192         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
4193             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
4194                 set_opt(sb, DELALLOC);
4195
4196         /*
4197          * set default s_li_wait_mult for lazyinit, for the case there is
4198          * no mount option specified.
4199          */
4200         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
4201
4202         if (le32_to_cpu(es->s_log_block_size) >
4203             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4204                 ext4_msg(sb, KERN_ERR,
4205                          "Invalid log block size: %u",
4206                          le32_to_cpu(es->s_log_block_size));
4207                 goto failed_mount;
4208         }
4209         if (le32_to_cpu(es->s_log_cluster_size) >
4210             (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4211                 ext4_msg(sb, KERN_ERR,
4212                          "Invalid log cluster size: %u",
4213                          le32_to_cpu(es->s_log_cluster_size));
4214                 goto failed_mount;
4215         }
4216
4217         blocksize = EXT4_MIN_BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
4218
4219         if (blocksize == PAGE_SIZE)
4220                 set_opt(sb, DIOREAD_NOLOCK);
4221
4222         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
4223                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
4224                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
4225         } else {
4226                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
4227                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
4228                 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) {
4229                         ext4_msg(sb, KERN_ERR, "invalid first ino: %u",
4230                                  sbi->s_first_ino);
4231                         goto failed_mount;
4232                 }
4233                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
4234                     (!is_power_of_2(sbi->s_inode_size)) ||
4235                     (sbi->s_inode_size > blocksize)) {
4236                         ext4_msg(sb, KERN_ERR,
4237                                "unsupported inode size: %d",
4238                                sbi->s_inode_size);
4239                         ext4_msg(sb, KERN_ERR, "blocksize: %d", blocksize);
4240                         goto failed_mount;
4241                 }
4242                 /*
4243                  * i_atime_extra is the last extra field available for
4244                  * [acm]times in struct ext4_inode. Checking for that
4245                  * field should suffice to ensure we have extra space
4246                  * for all three.
4247                  */
4248                 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) +
4249                         sizeof(((struct ext4_inode *)0)->i_atime_extra)) {
4250                         sb->s_time_gran = 1;
4251                         sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX;
4252                 } else {
4253                         sb->s_time_gran = NSEC_PER_SEC;
4254                         sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX;
4255                 }
4256                 sb->s_time_min = EXT4_TIMESTAMP_MIN;
4257         }
4258         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
4259                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4260                         EXT4_GOOD_OLD_INODE_SIZE;
4261                 if (ext4_has_feature_extra_isize(sb)) {
4262                         unsigned v, max = (sbi->s_inode_size -
4263                                            EXT4_GOOD_OLD_INODE_SIZE);
4264
4265                         v = le16_to_cpu(es->s_want_extra_isize);
4266                         if (v > max) {
4267                                 ext4_msg(sb, KERN_ERR,
4268                                          "bad s_want_extra_isize: %d", v);
4269                                 goto failed_mount;
4270                         }
4271                         if (sbi->s_want_extra_isize < v)
4272                                 sbi->s_want_extra_isize = v;
4273
4274                         v = le16_to_cpu(es->s_min_extra_isize);
4275                         if (v > max) {
4276                                 ext4_msg(sb, KERN_ERR,
4277                                          "bad s_min_extra_isize: %d", v);
4278                                 goto failed_mount;
4279                         }
4280                         if (sbi->s_want_extra_isize < v)
4281                                 sbi->s_want_extra_isize = v;
4282                 }
4283         }
4284
4285         if (sbi->s_es->s_mount_opts[0]) {
4286                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
4287                                               sizeof(sbi->s_es->s_mount_opts),
4288                                               GFP_KERNEL);
4289                 if (!s_mount_opts)
4290                         goto failed_mount;
4291                 if (!parse_options(s_mount_opts, sb, &parsed_opts, 0)) {
4292                         ext4_msg(sb, KERN_WARNING,
4293                                  "failed to parse options in superblock: %s",
4294                                  s_mount_opts);
4295                 }
4296                 kfree(s_mount_opts);
4297         }
4298         sbi->s_def_mount_opt = sbi->s_mount_opt;
4299         if (!parse_options((char *) data, sb, &parsed_opts, 0))
4300                 goto failed_mount;
4301
4302 #ifdef CONFIG_UNICODE
4303         if (ext4_has_feature_casefold(sb) && !sb->s_encoding) {
4304                 const struct ext4_sb_encodings *encoding_info;
4305                 struct unicode_map *encoding;
4306                 __u16 encoding_flags;
4307
4308                 if (ext4_sb_read_encoding(es, &encoding_info,
4309                                           &encoding_flags)) {
4310                         ext4_msg(sb, KERN_ERR,
4311                                  "Encoding requested by superblock is unknown");
4312                         goto failed_mount;
4313                 }
4314
4315                 encoding = utf8_load(encoding_info->version);
4316                 if (IS_ERR(encoding)) {
4317                         ext4_msg(sb, KERN_ERR,
4318                                  "can't mount with superblock charset: %s-%s "
4319                                  "not supported by the kernel. flags: 0x%x.",
4320                                  encoding_info->name, encoding_info->version,
4321                                  encoding_flags);
4322                         goto failed_mount;
4323                 }
4324                 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: "
4325                          "%s-%s with flags 0x%hx", encoding_info->name,
4326                          encoding_info->version?:"\b", encoding_flags);
4327
4328                 sb->s_encoding = encoding;
4329                 sb->s_encoding_flags = encoding_flags;
4330         }
4331 #endif
4332
4333         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4334                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with data=journal disables delayed allocation, dioread_nolock, O_DIRECT and fast_commit support!\n");
4335                 /* can't mount with both data=journal and dioread_nolock. */
4336                 clear_opt(sb, DIOREAD_NOLOCK);
4337                 clear_opt2(sb, JOURNAL_FAST_COMMIT);
4338                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4339                         ext4_msg(sb, KERN_ERR, "can't mount with "
4340                                  "both data=journal and delalloc");
4341                         goto failed_mount;
4342                 }
4343                 if (test_opt(sb, DAX_ALWAYS)) {
4344                         ext4_msg(sb, KERN_ERR, "can't mount with "
4345                                  "both data=journal and dax");
4346                         goto failed_mount;
4347                 }
4348                 if (ext4_has_feature_encrypt(sb)) {
4349                         ext4_msg(sb, KERN_WARNING,
4350                                  "encrypted files will use data=ordered "
4351                                  "instead of data journaling mode");
4352                 }
4353                 if (test_opt(sb, DELALLOC))
4354                         clear_opt(sb, DELALLOC);
4355         } else {
4356                 sb->s_iflags |= SB_I_CGROUPWB;
4357         }
4358
4359         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
4360                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
4361
4362         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
4363             (ext4_has_compat_features(sb) ||
4364              ext4_has_ro_compat_features(sb) ||
4365              ext4_has_incompat_features(sb)))
4366                 ext4_msg(sb, KERN_WARNING,
4367                        "feature flags set on rev 0 fs, "
4368                        "running e2fsck is recommended");
4369
4370         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
4371                 set_opt2(sb, HURD_COMPAT);
4372                 if (ext4_has_feature_64bit(sb)) {
4373                         ext4_msg(sb, KERN_ERR,
4374                                  "The Hurd can't support 64-bit file systems");
4375                         goto failed_mount;
4376                 }
4377
4378                 /*
4379                  * ea_inode feature uses l_i_version field which is not
4380                  * available in HURD_COMPAT mode.
4381                  */
4382                 if (ext4_has_feature_ea_inode(sb)) {
4383                         ext4_msg(sb, KERN_ERR,
4384                                  "ea_inode feature is not supported for Hurd");
4385                         goto failed_mount;
4386                 }
4387         }
4388
4389         if (IS_EXT2_SB(sb)) {
4390                 if (ext2_feature_set_ok(sb))
4391                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
4392                                  "using the ext4 subsystem");
4393                 else {
4394                         /*
4395                          * If we're probing be silent, if this looks like
4396                          * it's actually an ext[34] filesystem.
4397                          */
4398                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4399                                 goto failed_mount;
4400                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
4401                                  "to feature incompatibilities");
4402                         goto failed_mount;
4403                 }
4404         }
4405
4406         if (IS_EXT3_SB(sb)) {
4407                 if (ext3_feature_set_ok(sb))
4408                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
4409                                  "using the ext4 subsystem");
4410                 else {
4411                         /*
4412                          * If we're probing be silent, if this looks like
4413                          * it's actually an ext4 filesystem.
4414                          */
4415                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4416                                 goto failed_mount;
4417                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
4418                                  "to feature incompatibilities");
4419                         goto failed_mount;
4420                 }
4421         }
4422
4423         /*
4424          * Check feature flags regardless of the revision level, since we
4425          * previously didn't change the revision level when setting the flags,
4426          * so there is a chance incompat flags are set on a rev 0 filesystem.
4427          */
4428         if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
4429                 goto failed_mount;
4430
4431         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
4432                 ext4_msg(sb, KERN_ERR,
4433                          "Number of reserved GDT blocks insanely large: %d",
4434                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
4435                 goto failed_mount;
4436         }
4437
4438         if (bdev_dax_supported(sb->s_bdev, blocksize))
4439                 set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags);
4440
4441         if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) {
4442                 if (ext4_has_feature_inline_data(sb)) {
4443                         ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
4444                                         " that may contain inline data");
4445                         goto failed_mount;
4446                 }
4447                 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) {
4448                         ext4_msg(sb, KERN_ERR,
4449                                 "DAX unsupported by block device.");
4450                         goto failed_mount;
4451                 }
4452         }
4453
4454         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
4455                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
4456                          es->s_encryption_level);
4457                 goto failed_mount;
4458         }
4459
4460         if (sb->s_blocksize != blocksize) {
4461                 /*
4462                  * bh must be released before kill_bdev(), otherwise
4463                  * it won't be freed and its page also. kill_bdev()
4464                  * is called by sb_set_blocksize().
4465                  */
4466                 brelse(bh);
4467                 /* Validate the filesystem blocksize */
4468                 if (!sb_set_blocksize(sb, blocksize)) {
4469                         ext4_msg(sb, KERN_ERR, "bad block size %d",
4470                                         blocksize);
4471                         bh = NULL;
4472                         goto failed_mount;
4473                 }
4474
4475                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
4476                 offset = do_div(logical_sb_block, blocksize);
4477                 bh = ext4_sb_bread_unmovable(sb, logical_sb_block);
4478                 if (IS_ERR(bh)) {
4479                         ext4_msg(sb, KERN_ERR,
4480                                "Can't read superblock on 2nd try");
4481                         ret = PTR_ERR(bh);
4482                         bh = NULL;
4483                         goto failed_mount;
4484                 }
4485                 es = (struct ext4_super_block *)(bh->b_data + offset);
4486                 sbi->s_es = es;
4487                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
4488                         ext4_msg(sb, KERN_ERR,
4489                                "Magic mismatch, very weird!");
4490                         goto failed_mount;
4491                 }
4492         }
4493
4494         has_huge_files = ext4_has_feature_huge_file(sb);
4495         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
4496                                                       has_huge_files);
4497         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
4498
4499         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
4500         if (ext4_has_feature_64bit(sb)) {
4501                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
4502                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
4503                     !is_power_of_2(sbi->s_desc_size)) {
4504                         ext4_msg(sb, KERN_ERR,
4505                                "unsupported descriptor size %lu",
4506                                sbi->s_desc_size);
4507                         goto failed_mount;
4508                 }
4509         } else
4510                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
4511
4512         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
4513         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
4514
4515         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
4516         if (sbi->s_inodes_per_block == 0)
4517                 goto cantfind_ext4;
4518         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
4519             sbi->s_inodes_per_group > blocksize * 8) {
4520                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
4521                          sbi->s_inodes_per_group);
4522                 goto failed_mount;
4523         }
4524         sbi->s_itb_per_group = sbi->s_inodes_per_group /
4525                                         sbi->s_inodes_per_block;
4526         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
4527         sbi->s_sbh = bh;
4528         sbi->s_mount_state = le16_to_cpu(es->s_state);
4529         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
4530         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
4531
4532         for (i = 0; i < 4; i++)
4533                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
4534         sbi->s_def_hash_version = es->s_def_hash_version;
4535         if (ext4_has_feature_dir_index(sb)) {
4536                 i = le32_to_cpu(es->s_flags);
4537                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
4538                         sbi->s_hash_unsigned = 3;
4539                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
4540 #ifdef __CHAR_UNSIGNED__
4541                         if (!sb_rdonly(sb))
4542                                 es->s_flags |=
4543                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
4544                         sbi->s_hash_unsigned = 3;
4545 #else
4546                         if (!sb_rdonly(sb))
4547                                 es->s_flags |=
4548                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
4549 #endif
4550                 }
4551         }
4552
4553         /* Handle clustersize */
4554         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
4555         if (ext4_has_feature_bigalloc(sb)) {
4556                 if (clustersize < blocksize) {
4557                         ext4_msg(sb, KERN_ERR,
4558                                  "cluster size (%d) smaller than "
4559                                  "block size (%d)", clustersize, blocksize);
4560                         goto failed_mount;
4561                 }
4562                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
4563                         le32_to_cpu(es->s_log_block_size);
4564                 sbi->s_clusters_per_group =
4565                         le32_to_cpu(es->s_clusters_per_group);
4566                 if (sbi->s_clusters_per_group > blocksize * 8) {
4567                         ext4_msg(sb, KERN_ERR,
4568                                  "#clusters per group too big: %lu",
4569                                  sbi->s_clusters_per_group);
4570                         goto failed_mount;
4571                 }
4572                 if (sbi->s_blocks_per_group !=
4573                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
4574                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
4575                                  "clusters per group (%lu) inconsistent",
4576                                  sbi->s_blocks_per_group,
4577                                  sbi->s_clusters_per_group);
4578                         goto failed_mount;
4579                 }
4580         } else {
4581                 if (clustersize != blocksize) {
4582                         ext4_msg(sb, KERN_ERR,
4583                                  "fragment/cluster size (%d) != "
4584                                  "block size (%d)", clustersize, blocksize);
4585                         goto failed_mount;
4586                 }
4587                 if (sbi->s_blocks_per_group > blocksize * 8) {
4588                         ext4_msg(sb, KERN_ERR,
4589                                  "#blocks per group too big: %lu",
4590                                  sbi->s_blocks_per_group);
4591                         goto failed_mount;
4592                 }
4593                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
4594                 sbi->s_cluster_bits = 0;
4595         }
4596         sbi->s_cluster_ratio = clustersize / blocksize;
4597
4598         /* Do we have standard group size of clustersize * 8 blocks ? */
4599         if (sbi->s_blocks_per_group == clustersize << 3)
4600                 set_opt2(sb, STD_GROUP_SIZE);
4601
4602         /*
4603          * Test whether we have more sectors than will fit in sector_t,
4604          * and whether the max offset is addressable by the page cache.
4605          */
4606         err = generic_check_addressable(sb->s_blocksize_bits,
4607                                         ext4_blocks_count(es));
4608         if (err) {
4609                 ext4_msg(sb, KERN_ERR, "filesystem"
4610                          " too large to mount safely on this system");
4611                 goto failed_mount;
4612         }
4613
4614         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
4615                 goto cantfind_ext4;
4616
4617         /* check blocks count against device size */
4618         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
4619         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
4620                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
4621                        "exceeds size of device (%llu blocks)",
4622                        ext4_blocks_count(es), blocks_count);
4623                 goto failed_mount;
4624         }
4625
4626         /*
4627          * It makes no sense for the first data block to be beyond the end
4628          * of the filesystem.
4629          */
4630         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
4631                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4632                          "block %u is beyond end of filesystem (%llu)",
4633                          le32_to_cpu(es->s_first_data_block),
4634                          ext4_blocks_count(es));
4635                 goto failed_mount;
4636         }
4637         if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) &&
4638             (sbi->s_cluster_ratio == 1)) {
4639                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4640                          "block is 0 with a 1k block and cluster size");
4641                 goto failed_mount;
4642         }
4643
4644         blocks_count = (ext4_blocks_count(es) -
4645                         le32_to_cpu(es->s_first_data_block) +
4646                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
4647         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
4648         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
4649                 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu "
4650                        "(block count %llu, first data block %u, "
4651                        "blocks per group %lu)", blocks_count,
4652                        ext4_blocks_count(es),
4653                        le32_to_cpu(es->s_first_data_block),
4654                        EXT4_BLOCKS_PER_GROUP(sb));
4655                 goto failed_mount;
4656         }
4657         sbi->s_groups_count = blocks_count;
4658         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
4659                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
4660         if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) !=
4661             le32_to_cpu(es->s_inodes_count)) {
4662                 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu",
4663                          le32_to_cpu(es->s_inodes_count),
4664                          ((u64)sbi->s_groups_count * sbi->s_inodes_per_group));
4665                 ret = -EINVAL;
4666                 goto failed_mount;
4667         }
4668         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
4669                    EXT4_DESC_PER_BLOCK(sb);
4670         if (ext4_has_feature_meta_bg(sb)) {
4671                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
4672                         ext4_msg(sb, KERN_WARNING,
4673                                  "first meta block group too large: %u "
4674                                  "(group descriptor block count %u)",
4675                                  le32_to_cpu(es->s_first_meta_bg), db_count);
4676                         goto failed_mount;
4677                 }
4678         }
4679         rcu_assign_pointer(sbi->s_group_desc,
4680                            kvmalloc_array(db_count,
4681                                           sizeof(struct buffer_head *),
4682                                           GFP_KERNEL));
4683         if (sbi->s_group_desc == NULL) {
4684                 ext4_msg(sb, KERN_ERR, "not enough memory");
4685                 ret = -ENOMEM;
4686                 goto failed_mount;
4687         }
4688
4689         bgl_lock_init(sbi->s_blockgroup_lock);
4690
4691         /* Pre-read the descriptors into the buffer cache */
4692         for (i = 0; i < db_count; i++) {
4693                 block = descriptor_loc(sb, logical_sb_block, i);
4694                 ext4_sb_breadahead_unmovable(sb, block);
4695         }
4696
4697         for (i = 0; i < db_count; i++) {
4698                 struct buffer_head *bh;
4699
4700                 block = descriptor_loc(sb, logical_sb_block, i);
4701                 bh = ext4_sb_bread_unmovable(sb, block);
4702                 if (IS_ERR(bh)) {
4703                         ext4_msg(sb, KERN_ERR,
4704                                "can't read group descriptor %d", i);
4705                         db_count = i;
4706                         ret = PTR_ERR(bh);
4707                         goto failed_mount2;
4708                 }
4709                 rcu_read_lock();
4710                 rcu_dereference(sbi->s_group_desc)[i] = bh;
4711                 rcu_read_unlock();
4712         }
4713         sbi->s_gdb_count = db_count;
4714         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
4715                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
4716                 ret = -EFSCORRUPTED;
4717                 goto failed_mount2;
4718         }
4719
4720         timer_setup(&sbi->s_err_report, print_daily_error_info, 0);
4721         spin_lock_init(&sbi->s_error_lock);
4722         INIT_WORK(&sbi->s_error_work, flush_stashed_error_work);
4723
4724         /* Register extent status tree shrinker */
4725         if (ext4_es_register_shrinker(sbi))
4726                 goto failed_mount3;
4727
4728         sbi->s_stripe = ext4_get_stripe_size(sbi);
4729         sbi->s_extent_max_zeroout_kb = 32;
4730
4731         /*
4732          * set up enough so that it can read an inode
4733          */
4734         sb->s_op = &ext4_sops;
4735         sb->s_export_op = &ext4_export_ops;
4736         sb->s_xattr = ext4_xattr_handlers;
4737 #ifdef CONFIG_FS_ENCRYPTION
4738         sb->s_cop = &ext4_cryptops;
4739 #endif
4740 #ifdef CONFIG_FS_VERITY
4741         sb->s_vop = &ext4_verityops;
4742 #endif
4743 #ifdef CONFIG_QUOTA
4744         sb->dq_op = &ext4_quota_operations;
4745         if (ext4_has_feature_quota(sb))
4746                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4747         else
4748                 sb->s_qcop = &ext4_qctl_operations;
4749         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
4750 #endif
4751         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
4752
4753         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
4754         mutex_init(&sbi->s_orphan_lock);
4755
4756         /* Initialize fast commit stuff */
4757         atomic_set(&sbi->s_fc_subtid, 0);
4758         atomic_set(&sbi->s_fc_ineligible_updates, 0);
4759         INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_MAIN]);
4760         INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_STAGING]);
4761         INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_MAIN]);
4762         INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_STAGING]);
4763         sbi->s_fc_bytes = 0;
4764         ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE);
4765         ext4_clear_mount_flag(sb, EXT4_MF_FC_COMMITTING);
4766         spin_lock_init(&sbi->s_fc_lock);
4767         memset(&sbi->s_fc_stats, 0, sizeof(sbi->s_fc_stats));
4768         sbi->s_fc_replay_state.fc_regions = NULL;
4769         sbi->s_fc_replay_state.fc_regions_size = 0;
4770         sbi->s_fc_replay_state.fc_regions_used = 0;
4771         sbi->s_fc_replay_state.fc_regions_valid = 0;
4772         sbi->s_fc_replay_state.fc_modified_inodes = NULL;
4773         sbi->s_fc_replay_state.fc_modified_inodes_size = 0;
4774         sbi->s_fc_replay_state.fc_modified_inodes_used = 0;
4775
4776         sb->s_root = NULL;
4777
4778         needs_recovery = (es->s_last_orphan != 0 ||
4779                           ext4_has_feature_journal_needs_recovery(sb));
4780
4781         if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb))
4782                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
4783                         goto failed_mount3a;
4784
4785         /*
4786          * The first inode we look at is the journal inode.  Don't try
4787          * root first: it may be modified in the journal!
4788          */
4789         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
4790                 err = ext4_load_journal(sb, es, parsed_opts.journal_devnum);
4791                 if (err)
4792                         goto failed_mount3a;
4793         } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
4794                    ext4_has_feature_journal_needs_recovery(sb)) {
4795                 ext4_msg(sb, KERN_ERR, "required journal recovery "
4796                        "suppressed and not mounted read-only");
4797                 goto failed_mount_wq;
4798         } else {
4799                 /* Nojournal mode, all journal mount options are illegal */
4800                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
4801                         ext4_msg(sb, KERN_ERR, "can't mount with "
4802                                  "journal_checksum, fs mounted w/o journal");
4803                         goto failed_mount_wq;
4804                 }
4805                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4806                         ext4_msg(sb, KERN_ERR, "can't mount with "
4807                                  "journal_async_commit, fs mounted w/o journal");
4808                         goto failed_mount_wq;
4809                 }
4810                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
4811                         ext4_msg(sb, KERN_ERR, "can't mount with "
4812                                  "commit=%lu, fs mounted w/o journal",
4813                                  sbi->s_commit_interval / HZ);
4814                         goto failed_mount_wq;
4815                 }
4816                 if (EXT4_MOUNT_DATA_FLAGS &
4817                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
4818                         ext4_msg(sb, KERN_ERR, "can't mount with "
4819                                  "data=, fs mounted w/o journal");
4820                         goto failed_mount_wq;
4821                 }
4822                 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM;
4823                 clear_opt(sb, JOURNAL_CHECKSUM);
4824                 clear_opt(sb, DATA_FLAGS);
4825                 clear_opt2(sb, JOURNAL_FAST_COMMIT);
4826                 sbi->s_journal = NULL;
4827                 needs_recovery = 0;
4828                 goto no_journal;
4829         }
4830
4831         if (ext4_has_feature_64bit(sb) &&
4832             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4833                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4834                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4835                 goto failed_mount_wq;
4836         }
4837
4838         if (!set_journal_csum_feature_set(sb)) {
4839                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4840                          "feature set");
4841                 goto failed_mount_wq;
4842         }
4843
4844         if (test_opt2(sb, JOURNAL_FAST_COMMIT) &&
4845                 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4846                                           JBD2_FEATURE_INCOMPAT_FAST_COMMIT)) {
4847                 ext4_msg(sb, KERN_ERR,
4848                         "Failed to set fast commit journal feature");
4849                 goto failed_mount_wq;
4850         }
4851
4852         /* We have now updated the journal if required, so we can
4853          * validate the data journaling mode. */
4854         switch (test_opt(sb, DATA_FLAGS)) {
4855         case 0:
4856                 /* No mode set, assume a default based on the journal
4857                  * capabilities: ORDERED_DATA if the journal can
4858                  * cope, else JOURNAL_DATA
4859                  */
4860                 if (jbd2_journal_check_available_features
4861                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4862                         set_opt(sb, ORDERED_DATA);
4863                         sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
4864                 } else {
4865                         set_opt(sb, JOURNAL_DATA);
4866                         sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
4867                 }
4868                 break;
4869
4870         case EXT4_MOUNT_ORDERED_DATA:
4871         case EXT4_MOUNT_WRITEBACK_DATA:
4872                 if (!jbd2_journal_check_available_features
4873                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4874                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4875                                "requested data journaling mode");
4876                         goto failed_mount_wq;
4877                 }
4878                 break;
4879         default:
4880                 break;
4881         }
4882
4883         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4884             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4885                 ext4_msg(sb, KERN_ERR, "can't mount with "
4886                         "journal_async_commit in data=ordered mode");
4887                 goto failed_mount_wq;
4888         }
4889
4890         set_task_ioprio(sbi->s_journal->j_task, parsed_opts.journal_ioprio);
4891
4892         sbi->s_journal->j_submit_inode_data_buffers =
4893                 ext4_journal_submit_inode_data_buffers;
4894         sbi->s_journal->j_finish_inode_data_buffers =
4895                 ext4_journal_finish_inode_data_buffers;
4896
4897 no_journal:
4898         if (!test_opt(sb, NO_MBCACHE)) {
4899                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
4900                 if (!sbi->s_ea_block_cache) {
4901                         ext4_msg(sb, KERN_ERR,
4902                                  "Failed to create ea_block_cache");
4903                         goto failed_mount_wq;
4904                 }
4905
4906                 if (ext4_has_feature_ea_inode(sb)) {
4907                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
4908                         if (!sbi->s_ea_inode_cache) {
4909                                 ext4_msg(sb, KERN_ERR,
4910                                          "Failed to create ea_inode_cache");
4911                                 goto failed_mount_wq;
4912                         }
4913                 }
4914         }
4915
4916         if (ext4_has_feature_verity(sb) && blocksize != PAGE_SIZE) {
4917                 ext4_msg(sb, KERN_ERR, "Unsupported blocksize for fs-verity");
4918                 goto failed_mount_wq;
4919         }
4920
4921         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) &&
4922             !ext4_has_feature_encrypt(sb)) {
4923                 ext4_set_feature_encrypt(sb);
4924                 ext4_commit_super(sb);
4925         }
4926
4927         /*
4928          * Get the # of file system overhead blocks from the
4929          * superblock if present.
4930          */
4931         if (es->s_overhead_clusters)
4932                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4933         else {
4934                 err = ext4_calculate_overhead(sb);
4935                 if (err)
4936                         goto failed_mount_wq;
4937         }
4938
4939         /*
4940          * The maximum number of concurrent works can be high and
4941          * concurrency isn't really necessary.  Limit it to 1.
4942          */
4943         EXT4_SB(sb)->rsv_conversion_wq =
4944                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4945         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4946                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4947                 ret = -ENOMEM;
4948                 goto failed_mount4;
4949         }
4950
4951         /*
4952          * The jbd2_journal_load will have done any necessary log recovery,
4953          * so we can safely mount the rest of the filesystem now.
4954          */
4955
4956         root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL);
4957         if (IS_ERR(root)) {
4958                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4959                 ret = PTR_ERR(root);
4960                 root = NULL;
4961                 goto failed_mount4;
4962         }
4963         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4964                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4965                 iput(root);
4966                 goto failed_mount4;
4967         }
4968
4969         sb->s_root = d_make_root(root);
4970         if (!sb->s_root) {
4971                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4972                 ret = -ENOMEM;
4973                 goto failed_mount4;
4974         }
4975
4976         ret = ext4_setup_super(sb, es, sb_rdonly(sb));
4977         if (ret == -EROFS) {
4978                 sb->s_flags |= SB_RDONLY;
4979                 ret = 0;
4980         } else if (ret)
4981                 goto failed_mount4a;
4982
4983         ext4_set_resv_clusters(sb);
4984
4985         if (test_opt(sb, BLOCK_VALIDITY)) {
4986                 err = ext4_setup_system_zone(sb);
4987                 if (err) {
4988                         ext4_msg(sb, KERN_ERR, "failed to initialize system "
4989                                  "zone (%d)", err);
4990                         goto failed_mount4a;
4991                 }
4992         }
4993         ext4_fc_replay_cleanup(sb);
4994
4995         ext4_ext_init(sb);
4996
4997         /*
4998          * Enable optimize_scan if number of groups is > threshold. This can be
4999          * turned off by passing "mb_optimize_scan=0". This can also be
5000          * turned on forcefully by passing "mb_optimize_scan=1".
5001          */
5002         if (parsed_opts.mb_optimize_scan == 1)
5003                 set_opt2(sb, MB_OPTIMIZE_SCAN);
5004         else if (parsed_opts.mb_optimize_scan == 0)
5005                 clear_opt2(sb, MB_OPTIMIZE_SCAN);
5006         else if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD)
5007                 set_opt2(sb, MB_OPTIMIZE_SCAN);
5008
5009         err = ext4_mb_init(sb);
5010         if (err) {
5011                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
5012                          err);
5013                 goto failed_mount5;
5014         }
5015
5016         /*
5017          * We can only set up the journal commit callback once
5018          * mballoc is initialized
5019          */
5020         if (sbi->s_journal)
5021                 sbi->s_journal->j_commit_callback =
5022                         ext4_journal_commit_callback;
5023
5024         block = ext4_count_free_clusters(sb);
5025         ext4_free_blocks_count_set(sbi->s_es,
5026                                    EXT4_C2B(sbi, block));
5027         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
5028                                   GFP_KERNEL);
5029         if (!err) {
5030                 unsigned long freei = ext4_count_free_inodes(sb);
5031                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
5032                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
5033                                           GFP_KERNEL);
5034         }
5035         if (!err)
5036                 err = percpu_counter_init(&sbi->s_dirs_counter,
5037                                           ext4_count_dirs(sb), GFP_KERNEL);
5038         if (!err)
5039                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
5040                                           GFP_KERNEL);
5041         if (!err)
5042                 err = percpu_counter_init(&sbi->s_sra_exceeded_retry_limit, 0,
5043                                           GFP_KERNEL);
5044         if (!err)
5045                 err = percpu_init_rwsem(&sbi->s_writepages_rwsem);
5046
5047         if (err) {
5048                 ext4_msg(sb, KERN_ERR, "insufficient memory");
5049                 goto failed_mount6;
5050         }
5051
5052         if (ext4_has_feature_flex_bg(sb))
5053                 if (!ext4_fill_flex_info(sb)) {
5054                         ext4_msg(sb, KERN_ERR,
5055                                "unable to initialize "
5056                                "flex_bg meta info!");
5057                         ret = -ENOMEM;
5058                         goto failed_mount6;
5059                 }
5060
5061         err = ext4_register_li_request(sb, first_not_zeroed);
5062         if (err)
5063                 goto failed_mount6;
5064
5065         err = ext4_register_sysfs(sb);
5066         if (err)
5067                 goto failed_mount7;
5068
5069 #ifdef CONFIG_QUOTA
5070         /* Enable quota usage during mount. */
5071         if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
5072                 err = ext4_enable_quotas(sb);
5073                 if (err)
5074                         goto failed_mount8;
5075         }
5076 #endif  /* CONFIG_QUOTA */
5077
5078         /*
5079          * Save the original bdev mapping's wb_err value which could be
5080          * used to detect the metadata async write error.
5081          */
5082         spin_lock_init(&sbi->s_bdev_wb_lock);
5083         errseq_check_and_advance(&sb->s_bdev->bd_inode->i_mapping->wb_err,
5084                                  &sbi->s_bdev_wb_err);
5085         sb->s_bdev->bd_super = sb;
5086         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
5087         ext4_orphan_cleanup(sb, es);
5088         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
5089         if (needs_recovery) {
5090                 ext4_msg(sb, KERN_INFO, "recovery complete");
5091                 err = ext4_mark_recovery_complete(sb, es);
5092                 if (err)
5093                         goto failed_mount8;
5094         }
5095         if (EXT4_SB(sb)->s_journal) {
5096                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
5097                         descr = " journalled data mode";
5098                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
5099                         descr = " ordered data mode";
5100                 else
5101                         descr = " writeback data mode";
5102         } else
5103                 descr = "out journal";
5104
5105         if (test_opt(sb, DISCARD)) {
5106                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
5107                 if (!blk_queue_discard(q))
5108                         ext4_msg(sb, KERN_WARNING,
5109                                  "mounting with \"discard\" option, but "
5110                                  "the device does not support discard");
5111         }
5112
5113         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
5114                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
5115                          "Opts: %.*s%s%s. Quota mode: %s.", descr,
5116                          (int) sizeof(sbi->s_es->s_mount_opts),
5117                          sbi->s_es->s_mount_opts,
5118                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data,
5119                          ext4_quota_mode(sb));
5120
5121         if (es->s_error_count)
5122                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
5123
5124         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
5125         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
5126         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
5127         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
5128         atomic_set(&sbi->s_warning_count, 0);
5129         atomic_set(&sbi->s_msg_count, 0);
5130
5131         kfree(orig_data);
5132         return 0;
5133
5134 cantfind_ext4:
5135         if (!silent)
5136                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
5137         goto failed_mount;
5138
5139 failed_mount8:
5140         ext4_unregister_sysfs(sb);
5141         kobject_put(&sbi->s_kobj);
5142 failed_mount7:
5143         ext4_unregister_li_request(sb);
5144 failed_mount6:
5145         ext4_mb_release(sb);
5146         rcu_read_lock();
5147         flex_groups = rcu_dereference(sbi->s_flex_groups);
5148         if (flex_groups) {
5149                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
5150                         kvfree(flex_groups[i]);
5151                 kvfree(flex_groups);
5152         }
5153         rcu_read_unlock();
5154         percpu_counter_destroy(&sbi->s_freeclusters_counter);
5155         percpu_counter_destroy(&sbi->s_freeinodes_counter);
5156         percpu_counter_destroy(&sbi->s_dirs_counter);
5157         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
5158         percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit);
5159         percpu_free_rwsem(&sbi->s_writepages_rwsem);
5160 failed_mount5:
5161         ext4_ext_release(sb);
5162         ext4_release_system_zone(sb);
5163 failed_mount4a:
5164         dput(sb->s_root);
5165         sb->s_root = NULL;
5166 failed_mount4:
5167         ext4_msg(sb, KERN_ERR, "mount failed");
5168         if (EXT4_SB(sb)->rsv_conversion_wq)
5169                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
5170 failed_mount_wq:
5171         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
5172         sbi->s_ea_inode_cache = NULL;
5173
5174         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
5175         sbi->s_ea_block_cache = NULL;
5176
5177         if (sbi->s_journal) {
5178                 jbd2_journal_destroy(sbi->s_journal);
5179                 sbi->s_journal = NULL;
5180         }
5181 failed_mount3a:
5182         ext4_es_unregister_shrinker(sbi);
5183 failed_mount3:
5184         flush_work(&sbi->s_error_work);
5185         del_timer_sync(&sbi->s_err_report);
5186         ext4_stop_mmpd(sbi);
5187 failed_mount2:
5188         rcu_read_lock();
5189         group_desc = rcu_dereference(sbi->s_group_desc);
5190         for (i = 0; i < db_count; i++)
5191                 brelse(group_desc[i]);
5192         kvfree(group_desc);
5193         rcu_read_unlock();
5194 failed_mount:
5195         if (sbi->s_chksum_driver)
5196                 crypto_free_shash(sbi->s_chksum_driver);
5197
5198 #ifdef CONFIG_UNICODE
5199         utf8_unload(sb->s_encoding);
5200 #endif
5201
5202 #ifdef CONFIG_QUOTA
5203         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5204                 kfree(get_qf_name(sb, sbi, i));
5205 #endif
5206         fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy);
5207         /* ext4_blkdev_remove() calls kill_bdev(), release bh before it. */
5208         brelse(bh);
5209         ext4_blkdev_remove(sbi);
5210 out_fail:
5211         sb->s_fs_info = NULL;
5212         kfree(sbi->s_blockgroup_lock);
5213 out_free_base:
5214         kfree(sbi);
5215         kfree(orig_data);
5216         fs_put_dax(dax_dev);
5217         return err ? err : ret;
5218 }
5219
5220 /*
5221  * Setup any per-fs journal parameters now.  We'll do this both on
5222  * initial mount, once the journal has been initialised but before we've
5223  * done any recovery; and again on any subsequent remount.
5224  */
5225 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
5226 {
5227         struct ext4_sb_info *sbi = EXT4_SB(sb);
5228
5229         journal->j_commit_interval = sbi->s_commit_interval;
5230         journal->j_min_batch_time = sbi->s_min_batch_time;
5231         journal->j_max_batch_time = sbi->s_max_batch_time;
5232         ext4_fc_init(sb, journal);
5233
5234         write_lock(&journal->j_state_lock);
5235         if (test_opt(sb, BARRIER))
5236                 journal->j_flags |= JBD2_BARRIER;
5237         else
5238                 journal->j_flags &= ~JBD2_BARRIER;
5239         if (test_opt(sb, DATA_ERR_ABORT))
5240                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
5241         else
5242                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
5243         write_unlock(&journal->j_state_lock);
5244 }
5245
5246 static struct inode *ext4_get_journal_inode(struct super_block *sb,
5247                                              unsigned int journal_inum)
5248 {
5249         struct inode *journal_inode;
5250
5251         /*
5252          * Test for the existence of a valid inode on disk.  Bad things
5253          * happen if we iget() an unused inode, as the subsequent iput()
5254          * will try to delete it.
5255          */
5256         journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL);
5257         if (IS_ERR(journal_inode)) {
5258                 ext4_msg(sb, KERN_ERR, "no journal found");
5259                 return NULL;
5260         }
5261         if (!journal_inode->i_nlink) {
5262                 make_bad_inode(journal_inode);
5263                 iput(journal_inode);
5264                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
5265                 return NULL;
5266         }
5267
5268         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
5269                   journal_inode, journal_inode->i_size);
5270         if (!S_ISREG(journal_inode->i_mode)) {
5271                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
5272                 iput(journal_inode);
5273                 return NULL;
5274         }
5275         return journal_inode;
5276 }
5277
5278 static journal_t *ext4_get_journal(struct super_block *sb,
5279                                    unsigned int journal_inum)
5280 {
5281         struct inode *journal_inode;
5282         journal_t *journal;
5283
5284         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5285                 return NULL;
5286
5287         journal_inode = ext4_get_journal_inode(sb, journal_inum);
5288         if (!journal_inode)
5289                 return NULL;
5290
5291         journal = jbd2_journal_init_inode(journal_inode);
5292         if (!journal) {
5293                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
5294                 iput(journal_inode);
5295                 return NULL;
5296         }
5297         journal->j_private = sb;
5298         ext4_init_journal_params(sb, journal);
5299         return journal;
5300 }
5301
5302 static journal_t *ext4_get_dev_journal(struct super_block *sb,
5303                                        dev_t j_dev)
5304 {
5305         struct buffer_head *bh;
5306         journal_t *journal;
5307         ext4_fsblk_t start;
5308         ext4_fsblk_t len;
5309         int hblock, blocksize;
5310         ext4_fsblk_t sb_block;
5311         unsigned long offset;
5312         struct ext4_super_block *es;
5313         struct block_device *bdev;
5314
5315         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5316                 return NULL;
5317
5318         bdev = ext4_blkdev_get(j_dev, sb);
5319         if (bdev == NULL)
5320                 return NULL;
5321
5322         blocksize = sb->s_blocksize;
5323         hblock = bdev_logical_block_size(bdev);
5324         if (blocksize < hblock) {
5325                 ext4_msg(sb, KERN_ERR,
5326                         "blocksize too small for journal device");
5327                 goto out_bdev;
5328         }
5329
5330         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
5331         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
5332         set_blocksize(bdev, blocksize);
5333         if (!(bh = __bread(bdev, sb_block, blocksize))) {
5334                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
5335                        "external journal");
5336                 goto out_bdev;
5337         }
5338
5339         es = (struct ext4_super_block *) (bh->b_data + offset);
5340         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
5341             !(le32_to_cpu(es->s_feature_incompat) &
5342               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
5343                 ext4_msg(sb, KERN_ERR, "external journal has "
5344                                         "bad superblock");
5345                 brelse(bh);
5346                 goto out_bdev;
5347         }
5348
5349         if ((le32_to_cpu(es->s_feature_ro_compat) &
5350              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
5351             es->s_checksum != ext4_superblock_csum(sb, es)) {
5352                 ext4_msg(sb, KERN_ERR, "external journal has "
5353                                        "corrupt superblock");
5354                 brelse(bh);
5355                 goto out_bdev;
5356         }
5357
5358         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
5359                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
5360                 brelse(bh);
5361                 goto out_bdev;
5362         }
5363
5364         len = ext4_blocks_count(es);
5365         start = sb_block + 1;
5366         brelse(bh);     /* we're done with the superblock */
5367
5368         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
5369                                         start, len, blocksize);
5370         if (!journal) {
5371                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
5372                 goto out_bdev;
5373         }
5374         journal->j_private = sb;
5375         if (ext4_read_bh_lock(journal->j_sb_buffer, REQ_META | REQ_PRIO, true)) {
5376                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
5377                 goto out_journal;
5378         }
5379         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
5380                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
5381                                         "user (unsupported) - %d",
5382                         be32_to_cpu(journal->j_superblock->s_nr_users));
5383                 goto out_journal;
5384         }
5385         EXT4_SB(sb)->s_journal_bdev = bdev;
5386         ext4_init_journal_params(sb, journal);
5387         return journal;
5388
5389 out_journal:
5390         jbd2_journal_destroy(journal);
5391 out_bdev:
5392         ext4_blkdev_put(bdev);
5393         return NULL;
5394 }
5395
5396 static int ext4_load_journal(struct super_block *sb,
5397                              struct ext4_super_block *es,
5398                              unsigned long journal_devnum)
5399 {
5400         journal_t *journal;
5401         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
5402         dev_t journal_dev;
5403         int err = 0;
5404         int really_read_only;
5405         int journal_dev_ro;
5406
5407         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5408                 return -EFSCORRUPTED;
5409
5410         if (journal_devnum &&
5411             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5412                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
5413                         "numbers have changed");
5414                 journal_dev = new_decode_dev(journal_devnum);
5415         } else
5416                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
5417
5418         if (journal_inum && journal_dev) {
5419                 ext4_msg(sb, KERN_ERR,
5420                          "filesystem has both journal inode and journal device!");
5421                 return -EINVAL;
5422         }
5423
5424         if (journal_inum) {
5425                 journal = ext4_get_journal(sb, journal_inum);
5426                 if (!journal)
5427                         return -EINVAL;
5428         } else {
5429                 journal = ext4_get_dev_journal(sb, journal_dev);
5430                 if (!journal)
5431                         return -EINVAL;
5432         }
5433
5434         journal_dev_ro = bdev_read_only(journal->j_dev);
5435         really_read_only = bdev_read_only(sb->s_bdev) | journal_dev_ro;
5436
5437         if (journal_dev_ro && !sb_rdonly(sb)) {
5438                 ext4_msg(sb, KERN_ERR,
5439                          "journal device read-only, try mounting with '-o ro'");
5440                 err = -EROFS;
5441                 goto err_out;
5442         }
5443
5444         /*
5445          * Are we loading a blank journal or performing recovery after a
5446          * crash?  For recovery, we need to check in advance whether we
5447          * can get read-write access to the device.
5448          */
5449         if (ext4_has_feature_journal_needs_recovery(sb)) {
5450                 if (sb_rdonly(sb)) {
5451                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
5452                                         "required on readonly filesystem");
5453                         if (really_read_only) {
5454                                 ext4_msg(sb, KERN_ERR, "write access "
5455                                         "unavailable, cannot proceed "
5456                                         "(try mounting with noload)");
5457                                 err = -EROFS;
5458                                 goto err_out;
5459                         }
5460                         ext4_msg(sb, KERN_INFO, "write access will "
5461                                "be enabled during recovery");
5462                 }
5463         }
5464
5465         if (!(journal->j_flags & JBD2_BARRIER))
5466                 ext4_msg(sb, KERN_INFO, "barriers disabled");
5467
5468         if (!ext4_has_feature_journal_needs_recovery(sb))
5469                 err = jbd2_journal_wipe(journal, !really_read_only);
5470         if (!err) {
5471                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
5472                 if (save)
5473                         memcpy(save, ((char *) es) +
5474                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
5475                 err = jbd2_journal_load(journal);
5476                 if (save)
5477                         memcpy(((char *) es) + EXT4_S_ERR_START,
5478                                save, EXT4_S_ERR_LEN);
5479                 kfree(save);
5480         }
5481
5482         if (err) {
5483                 ext4_msg(sb, KERN_ERR, "error loading journal");
5484                 goto err_out;
5485         }
5486
5487         EXT4_SB(sb)->s_journal = journal;
5488         err = ext4_clear_journal_err(sb, es);
5489         if (err) {
5490                 EXT4_SB(sb)->s_journal = NULL;
5491                 jbd2_journal_destroy(journal);
5492                 return err;
5493         }
5494
5495         if (!really_read_only && journal_devnum &&
5496             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5497                 es->s_journal_dev = cpu_to_le32(journal_devnum);
5498
5499                 /* Make sure we flush the recovery flag to disk. */
5500                 ext4_commit_super(sb);
5501         }
5502
5503         return 0;
5504
5505 err_out:
5506         jbd2_journal_destroy(journal);
5507         return err;
5508 }
5509
5510 /* Copy state of EXT4_SB(sb) into buffer for on-disk superblock */
5511 static void ext4_update_super(struct super_block *sb)
5512 {
5513         struct ext4_sb_info *sbi = EXT4_SB(sb);
5514         struct ext4_super_block *es = sbi->s_es;
5515         struct buffer_head *sbh = sbi->s_sbh;
5516
5517         lock_buffer(sbh);
5518         /*
5519          * If the file system is mounted read-only, don't update the
5520          * superblock write time.  This avoids updating the superblock
5521          * write time when we are mounting the root file system
5522          * read/only but we need to replay the journal; at that point,
5523          * for people who are east of GMT and who make their clock
5524          * tick in localtime for Windows bug-for-bug compatibility,
5525          * the clock is set in the future, and this will cause e2fsck
5526          * to complain and force a full file system check.
5527          */
5528         if (!(sb->s_flags & SB_RDONLY))
5529                 ext4_update_tstamp(es, s_wtime);
5530         es->s_kbytes_written =
5531                 cpu_to_le64(sbi->s_kbytes_written +
5532                     ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) -
5533                       sbi->s_sectors_written_start) >> 1));
5534         if (percpu_counter_initialized(&sbi->s_freeclusters_counter))
5535                 ext4_free_blocks_count_set(es,
5536                         EXT4_C2B(sbi, percpu_counter_sum_positive(
5537                                 &sbi->s_freeclusters_counter)));
5538         if (percpu_counter_initialized(&sbi->s_freeinodes_counter))
5539                 es->s_free_inodes_count =
5540                         cpu_to_le32(percpu_counter_sum_positive(
5541                                 &sbi->s_freeinodes_counter));
5542         /* Copy error information to the on-disk superblock */
5543         spin_lock(&sbi->s_error_lock);
5544         if (sbi->s_add_error_count > 0) {
5545                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
5546                 if (!es->s_first_error_time && !es->s_first_error_time_hi) {
5547                         __ext4_update_tstamp(&es->s_first_error_time,
5548                                              &es->s_first_error_time_hi,
5549                                              sbi->s_first_error_time);
5550                         strncpy(es->s_first_error_func, sbi->s_first_error_func,
5551                                 sizeof(es->s_first_error_func));
5552                         es->s_first_error_line =
5553                                 cpu_to_le32(sbi->s_first_error_line);
5554                         es->s_first_error_ino =
5555                                 cpu_to_le32(sbi->s_first_error_ino);
5556                         es->s_first_error_block =
5557                                 cpu_to_le64(sbi->s_first_error_block);
5558                         es->s_first_error_errcode =
5559                                 ext4_errno_to_code(sbi->s_first_error_code);
5560                 }
5561                 __ext4_update_tstamp(&es->s_last_error_time,
5562                                      &es->s_last_error_time_hi,
5563                                      sbi->s_last_error_time);
5564                 strncpy(es->s_last_error_func, sbi->s_last_error_func,
5565                         sizeof(es->s_last_error_func));
5566                 es->s_last_error_line = cpu_to_le32(sbi->s_last_error_line);
5567                 es->s_last_error_ino = cpu_to_le32(sbi->s_last_error_ino);
5568                 es->s_last_error_block = cpu_to_le64(sbi->s_last_error_block);
5569                 es->s_last_error_errcode =
5570                                 ext4_errno_to_code(sbi->s_last_error_code);
5571                 /*
5572                  * Start the daily error reporting function if it hasn't been
5573                  * started already
5574                  */
5575                 if (!es->s_error_count)
5576                         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);
5577                 le32_add_cpu(&es->s_error_count, sbi->s_add_error_count);
5578                 sbi->s_add_error_count = 0;
5579         }
5580         spin_unlock(&sbi->s_error_lock);
5581
5582         ext4_superblock_csum_set(sb);
5583         unlock_buffer(sbh);
5584 }
5585
5586 static int ext4_commit_super(struct super_block *sb)
5587 {
5588         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
5589         int error = 0;
5590
5591         if (!sbh)
5592                 return -EINVAL;
5593         if (block_device_ejected(sb))
5594                 return -ENODEV;
5595
5596         ext4_update_super(sb);
5597
5598         if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
5599                 /*
5600                  * Oh, dear.  A previous attempt to write the
5601                  * superblock failed.  This could happen because the
5602                  * USB device was yanked out.  Or it could happen to
5603                  * be a transient write error and maybe the block will
5604                  * be remapped.  Nothing we can do but to retry the
5605                  * write and hope for the best.
5606                  */
5607                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
5608                        "superblock detected");
5609                 clear_buffer_write_io_error(sbh);
5610                 set_buffer_uptodate(sbh);
5611         }
5612         BUFFER_TRACE(sbh, "marking dirty");
5613         mark_buffer_dirty(sbh);
5614         error = __sync_dirty_buffer(sbh,
5615                 REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
5616         if (buffer_write_io_error(sbh)) {
5617                 ext4_msg(sb, KERN_ERR, "I/O error while writing "
5618                        "superblock");
5619                 clear_buffer_write_io_error(sbh);
5620                 set_buffer_uptodate(sbh);
5621         }
5622         return error;
5623 }
5624
5625 /*
5626  * Have we just finished recovery?  If so, and if we are mounting (or
5627  * remounting) the filesystem readonly, then we will end up with a
5628  * consistent fs on disk.  Record that fact.
5629  */
5630 static int ext4_mark_recovery_complete(struct super_block *sb,
5631                                        struct ext4_super_block *es)
5632 {
5633         int err;
5634         journal_t *journal = EXT4_SB(sb)->s_journal;
5635
5636         if (!ext4_has_feature_journal(sb)) {
5637                 if (journal != NULL) {
5638                         ext4_error(sb, "Journal got removed while the fs was "
5639                                    "mounted!");
5640                         return -EFSCORRUPTED;
5641                 }
5642                 return 0;
5643         }
5644         jbd2_journal_lock_updates(journal);
5645         err = jbd2_journal_flush(journal, 0);
5646         if (err < 0)
5647                 goto out;
5648
5649         if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) {
5650                 ext4_clear_feature_journal_needs_recovery(sb);
5651                 ext4_commit_super(sb);
5652         }
5653 out:
5654         jbd2_journal_unlock_updates(journal);
5655         return err;
5656 }
5657
5658 /*
5659  * If we are mounting (or read-write remounting) a filesystem whose journal
5660  * has recorded an error from a previous lifetime, move that error to the
5661  * main filesystem now.
5662  */
5663 static int ext4_clear_journal_err(struct super_block *sb,
5664                                    struct ext4_super_block *es)
5665 {
5666         journal_t *journal;
5667         int j_errno;
5668         const char *errstr;
5669
5670         if (!ext4_has_feature_journal(sb)) {
5671                 ext4_error(sb, "Journal got removed while the fs was mounted!");
5672                 return -EFSCORRUPTED;
5673         }
5674
5675         journal = EXT4_SB(sb)->s_journal;
5676
5677         /*
5678          * Now check for any error status which may have been recorded in the
5679          * journal by a prior ext4_error() or ext4_abort()
5680          */
5681
5682         j_errno = jbd2_journal_errno(journal);
5683         if (j_errno) {
5684                 char nbuf[16];
5685
5686                 errstr = ext4_decode_error(sb, j_errno, nbuf);
5687                 ext4_warning(sb, "Filesystem error recorded "
5688                              "from previous mount: %s", errstr);
5689                 ext4_warning(sb, "Marking fs in need of filesystem check.");
5690
5691                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
5692                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
5693                 ext4_commit_super(sb);
5694
5695                 jbd2_journal_clear_err(journal);
5696                 jbd2_journal_update_sb_errno(journal);
5697         }
5698         return 0;
5699 }
5700
5701 /*
5702  * Force the running and committing transactions to commit,
5703  * and wait on the commit.
5704  */
5705 int ext4_force_commit(struct super_block *sb)
5706 {
5707         journal_t *journal;
5708
5709         if (sb_rdonly(sb))
5710                 return 0;
5711
5712         journal = EXT4_SB(sb)->s_journal;
5713         return ext4_journal_force_commit(journal);
5714 }
5715
5716 static int ext4_sync_fs(struct super_block *sb, int wait)
5717 {
5718         int ret = 0;
5719         tid_t target;
5720         bool needs_barrier = false;
5721         struct ext4_sb_info *sbi = EXT4_SB(sb);
5722
5723         if (unlikely(ext4_forced_shutdown(sbi)))
5724                 return 0;
5725
5726         trace_ext4_sync_fs(sb, wait);
5727         flush_workqueue(sbi->rsv_conversion_wq);
5728         /*
5729          * Writeback quota in non-journalled quota case - journalled quota has
5730          * no dirty dquots
5731          */
5732         dquot_writeback_dquots(sb, -1);
5733         /*
5734          * Data writeback is possible w/o journal transaction, so barrier must
5735          * being sent at the end of the function. But we can skip it if
5736          * transaction_commit will do it for us.
5737          */
5738         if (sbi->s_journal) {
5739                 target = jbd2_get_latest_transaction(sbi->s_journal);
5740                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
5741                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
5742                         needs_barrier = true;
5743
5744                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
5745                         if (wait)
5746                                 ret = jbd2_log_wait_commit(sbi->s_journal,
5747                                                            target);
5748                 }
5749         } else if (wait && test_opt(sb, BARRIER))
5750                 needs_barrier = true;
5751         if (needs_barrier) {
5752                 int err;
5753                 err = blkdev_issue_flush(sb->s_bdev);
5754                 if (!ret)
5755                         ret = err;
5756         }
5757
5758         return ret;
5759 }
5760
5761 /*
5762  * LVM calls this function before a (read-only) snapshot is created.  This
5763  * gives us a chance to flush the journal completely and mark the fs clean.
5764  *
5765  * Note that only this function cannot bring a filesystem to be in a clean
5766  * state independently. It relies on upper layer to stop all data & metadata
5767  * modifications.
5768  */
5769 static int ext4_freeze(struct super_block *sb)
5770 {
5771         int error = 0;
5772         journal_t *journal;
5773
5774         if (sb_rdonly(sb))
5775                 return 0;
5776
5777         journal = EXT4_SB(sb)->s_journal;
5778
5779         if (journal) {
5780                 /* Now we set up the journal barrier. */
5781                 jbd2_journal_lock_updates(journal);
5782
5783                 /*
5784                  * Don't clear the needs_recovery flag if we failed to
5785                  * flush the journal.
5786                  */
5787                 error = jbd2_journal_flush(journal, 0);
5788                 if (error < 0)
5789                         goto out;
5790
5791                 /* Journal blocked and flushed, clear needs_recovery flag. */
5792                 ext4_clear_feature_journal_needs_recovery(sb);
5793         }
5794
5795         error = ext4_commit_super(sb);
5796 out:
5797         if (journal)
5798                 /* we rely on upper layer to stop further updates */
5799                 jbd2_journal_unlock_updates(journal);
5800         return error;
5801 }
5802
5803 /*
5804  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
5805  * flag here, even though the filesystem is not technically dirty yet.
5806  */
5807 static int ext4_unfreeze(struct super_block *sb)
5808 {
5809         if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
5810                 return 0;
5811
5812         if (EXT4_SB(sb)->s_journal) {
5813                 /* Reset the needs_recovery flag before the fs is unlocked. */
5814                 ext4_set_feature_journal_needs_recovery(sb);
5815         }
5816
5817         ext4_commit_super(sb);
5818         return 0;
5819 }
5820
5821 /*
5822  * Structure to save mount options for ext4_remount's benefit
5823  */
5824 struct ext4_mount_options {
5825         unsigned long s_mount_opt;
5826         unsigned long s_mount_opt2;
5827         kuid_t s_resuid;
5828         kgid_t s_resgid;
5829         unsigned long s_commit_interval;
5830         u32 s_min_batch_time, s_max_batch_time;
5831 #ifdef CONFIG_QUOTA
5832         int s_jquota_fmt;
5833         char *s_qf_names[EXT4_MAXQUOTAS];
5834 #endif
5835 };
5836
5837 static int ext4_remount(struct super_block *sb, int *flags, char *data)
5838 {
5839         struct ext4_super_block *es;
5840         struct ext4_sb_info *sbi = EXT4_SB(sb);
5841         unsigned long old_sb_flags, vfs_flags;
5842         struct ext4_mount_options old_opts;
5843         int enable_quota = 0;
5844         ext4_group_t g;
5845         int err = 0;
5846 #ifdef CONFIG_QUOTA
5847         int i, j;
5848         char *to_free[EXT4_MAXQUOTAS];
5849 #endif
5850         char *orig_data = kstrdup(data, GFP_KERNEL);
5851         struct ext4_parsed_options parsed_opts;
5852
5853         parsed_opts.journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
5854         parsed_opts.journal_devnum = 0;
5855
5856         if (data && !orig_data)
5857                 return -ENOMEM;
5858
5859         /* Store the original options */
5860         old_sb_flags = sb->s_flags;
5861         old_opts.s_mount_opt = sbi->s_mount_opt;
5862         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
5863         old_opts.s_resuid = sbi->s_resuid;
5864         old_opts.s_resgid = sbi->s_resgid;
5865         old_opts.s_commit_interval = sbi->s_commit_interval;
5866         old_opts.s_min_batch_time = sbi->s_min_batch_time;
5867         old_opts.s_max_batch_time = sbi->s_max_batch_time;
5868 #ifdef CONFIG_QUOTA
5869         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
5870         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5871                 if (sbi->s_qf_names[i]) {
5872                         char *qf_name = get_qf_name(sb, sbi, i);
5873
5874                         old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL);
5875                         if (!old_opts.s_qf_names[i]) {
5876                                 for (j = 0; j < i; j++)
5877                                         kfree(old_opts.s_qf_names[j]);
5878                                 kfree(orig_data);
5879                                 return -ENOMEM;
5880                         }
5881                 } else
5882                         old_opts.s_qf_names[i] = NULL;
5883 #endif
5884         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
5885                 parsed_opts.journal_ioprio =
5886                         sbi->s_journal->j_task->io_context->ioprio;
5887
5888         /*
5889          * Some options can be enabled by ext4 and/or by VFS mount flag
5890          * either way we need to make sure it matches in both *flags and
5891          * s_flags. Copy those selected flags from *flags to s_flags
5892          */
5893         vfs_flags = SB_LAZYTIME | SB_I_VERSION;
5894         sb->s_flags = (sb->s_flags & ~vfs_flags) | (*flags & vfs_flags);
5895
5896         if (!parse_options(data, sb, &parsed_opts, 1)) {
5897                 err = -EINVAL;
5898                 goto restore_opts;
5899         }
5900
5901         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
5902             test_opt(sb, JOURNAL_CHECKSUM)) {
5903                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
5904                          "during remount not supported; ignoring");
5905                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
5906         }
5907
5908         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
5909                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
5910                         ext4_msg(sb, KERN_ERR, "can't mount with "
5911                                  "both data=journal and delalloc");
5912                         err = -EINVAL;
5913                         goto restore_opts;
5914                 }
5915                 if (test_opt(sb, DIOREAD_NOLOCK)) {
5916                         ext4_msg(sb, KERN_ERR, "can't mount with "
5917                                  "both data=journal and dioread_nolock");
5918                         err = -EINVAL;
5919                         goto restore_opts;
5920                 }
5921         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
5922                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5923                         ext4_msg(sb, KERN_ERR, "can't mount with "
5924                                 "journal_async_commit in data=ordered mode");
5925                         err = -EINVAL;
5926                         goto restore_opts;
5927                 }
5928         }
5929
5930         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
5931                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
5932                 err = -EINVAL;
5933                 goto restore_opts;
5934         }
5935
5936         if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED))
5937                 ext4_abort(sb, EXT4_ERR_ESHUTDOWN, "Abort forced by user");
5938
5939         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
5940                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
5941
5942         es = sbi->s_es;
5943
5944         if (sbi->s_journal) {
5945                 ext4_init_journal_params(sb, sbi->s_journal);
5946                 set_task_ioprio(sbi->s_journal->j_task, parsed_opts.journal_ioprio);
5947         }
5948
5949         /* Flush outstanding errors before changing fs state */
5950         flush_work(&sbi->s_error_work);
5951
5952         if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) {
5953                 if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED)) {
5954                         err = -EROFS;
5955                         goto restore_opts;
5956                 }
5957
5958                 if (*flags & SB_RDONLY) {
5959                         err = sync_filesystem(sb);
5960                         if (err < 0)
5961                                 goto restore_opts;
5962                         err = dquot_suspend(sb, -1);
5963                         if (err < 0)
5964                                 goto restore_opts;
5965
5966                         /*
5967                          * First of all, the unconditional stuff we have to do
5968                          * to disable replay of the journal when we next remount
5969                          */
5970                         sb->s_flags |= SB_RDONLY;
5971
5972                         /*
5973                          * OK, test if we are remounting a valid rw partition
5974                          * readonly, and if so set the rdonly flag and then
5975                          * mark the partition as valid again.
5976                          */
5977                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5978                             (sbi->s_mount_state & EXT4_VALID_FS))
5979                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5980
5981                         if (sbi->s_journal) {
5982                                 /*
5983                                  * We let remount-ro finish even if marking fs
5984                                  * as clean failed...
5985                                  */
5986                                 ext4_mark_recovery_complete(sb, es);
5987                         }
5988                 } else {
5989                         /* Make sure we can mount this feature set readwrite */
5990                         if (ext4_has_feature_readonly(sb) ||
5991                             !ext4_feature_set_ok(sb, 0)) {
5992                                 err = -EROFS;
5993                                 goto restore_opts;
5994                         }
5995                         /*
5996                          * Make sure the group descriptor checksums
5997                          * are sane.  If they aren't, refuse to remount r/w.
5998                          */
5999                         for (g = 0; g < sbi->s_groups_count; g++) {
6000                                 struct ext4_group_desc *gdp =
6001                                         ext4_get_group_desc(sb, g, NULL);
6002
6003                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
6004                                         ext4_msg(sb, KERN_ERR,
6005                "ext4_remount: Checksum for group %u failed (%u!=%u)",
6006                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
6007                                                le16_to_cpu(gdp->bg_checksum));
6008                                         err = -EFSBADCRC;
6009                                         goto restore_opts;
6010                                 }
6011                         }
6012
6013                         /*
6014                          * If we have an unprocessed orphan list hanging
6015                          * around from a previously readonly bdev mount,
6016                          * require a full umount/remount for now.
6017                          */
6018                         if (es->s_last_orphan) {
6019                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
6020                                        "remount RDWR because of unprocessed "
6021                                        "orphan inode list.  Please "
6022                                        "umount/remount instead");
6023                                 err = -EINVAL;
6024                                 goto restore_opts;
6025                         }
6026
6027                         /*
6028                          * Mounting a RDONLY partition read-write, so reread
6029                          * and store the current valid flag.  (It may have
6030                          * been changed by e2fsck since we originally mounted
6031                          * the partition.)
6032                          */
6033                         if (sbi->s_journal) {
6034                                 err = ext4_clear_journal_err(sb, es);
6035                                 if (err)
6036                                         goto restore_opts;
6037                         }
6038                         sbi->s_mount_state = le16_to_cpu(es->s_state);
6039
6040                         err = ext4_setup_super(sb, es, 0);
6041                         if (err)
6042                                 goto restore_opts;
6043
6044                         sb->s_flags &= ~SB_RDONLY;
6045                         if (ext4_has_feature_mmp(sb))
6046                                 if (ext4_multi_mount_protect(sb,
6047                                                 le64_to_cpu(es->s_mmp_block))) {
6048                                         err = -EROFS;
6049                                         goto restore_opts;
6050                                 }
6051                         enable_quota = 1;
6052                 }
6053         }
6054
6055         /*
6056          * Reinitialize lazy itable initialization thread based on
6057          * current settings
6058          */
6059         if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
6060                 ext4_unregister_li_request(sb);
6061         else {
6062                 ext4_group_t first_not_zeroed;
6063                 first_not_zeroed = ext4_has_uninit_itable(sb);
6064                 ext4_register_li_request(sb, first_not_zeroed);
6065         }
6066
6067         /*
6068          * Handle creation of system zone data early because it can fail.
6069          * Releasing of existing data is done when we are sure remount will
6070          * succeed.
6071          */
6072         if (test_opt(sb, BLOCK_VALIDITY) && !sbi->s_system_blks) {
6073                 err = ext4_setup_system_zone(sb);
6074                 if (err)
6075                         goto restore_opts;
6076         }
6077
6078         if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) {
6079                 err = ext4_commit_super(sb);
6080                 if (err)
6081                         goto restore_opts;
6082         }
6083
6084 #ifdef CONFIG_QUOTA
6085         /* Release old quota file names */
6086         for (i = 0; i < EXT4_MAXQUOTAS; i++)
6087                 kfree(old_opts.s_qf_names[i]);
6088         if (enable_quota) {
6089                 if (sb_any_quota_suspended(sb))
6090                         dquot_resume(sb, -1);
6091                 else if (ext4_has_feature_quota(sb)) {
6092                         err = ext4_enable_quotas(sb);
6093                         if (err)
6094                                 goto restore_opts;
6095                 }
6096         }
6097 #endif
6098         if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks)
6099                 ext4_release_system_zone(sb);
6100
6101         if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb))
6102                 ext4_stop_mmpd(sbi);
6103
6104         /*
6105          * Some options can be enabled by ext4 and/or by VFS mount flag
6106          * either way we need to make sure it matches in both *flags and
6107          * s_flags. Copy those selected flags from s_flags to *flags
6108          */
6109         *flags = (*flags & ~vfs_flags) | (sb->s_flags & vfs_flags);
6110
6111         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s. Quota mode: %s.",
6112                  orig_data, ext4_quota_mode(sb));
6113         kfree(orig_data);
6114         return 0;
6115
6116 restore_opts:
6117         sb->s_flags = old_sb_flags;
6118         sbi->s_mount_opt = old_opts.s_mount_opt;
6119         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
6120         sbi->s_resuid = old_opts.s_resuid;
6121         sbi->s_resgid = old_opts.s_resgid;
6122         sbi->s_commit_interval = old_opts.s_commit_interval;
6123         sbi->s_min_batch_time = old_opts.s_min_batch_time;
6124         sbi->s_max_batch_time = old_opts.s_max_batch_time;
6125         if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks)
6126                 ext4_release_system_zone(sb);
6127 #ifdef CONFIG_QUOTA
6128         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
6129         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
6130                 to_free[i] = get_qf_name(sb, sbi, i);
6131                 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]);
6132         }
6133         synchronize_rcu();
6134         for (i = 0; i < EXT4_MAXQUOTAS; i++)
6135                 kfree(to_free[i]);
6136 #endif
6137         if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb))
6138                 ext4_stop_mmpd(sbi);
6139         kfree(orig_data);
6140         return err;
6141 }
6142
6143 #ifdef CONFIG_QUOTA
6144 static int ext4_statfs_project(struct super_block *sb,
6145                                kprojid_t projid, struct kstatfs *buf)
6146 {
6147         struct kqid qid;
6148         struct dquot *dquot;
6149         u64 limit;
6150         u64 curblock;
6151
6152         qid = make_kqid_projid(projid);
6153         dquot = dqget(sb, qid);
6154         if (IS_ERR(dquot))
6155                 return PTR_ERR(dquot);
6156         spin_lock(&dquot->dq_dqb_lock);
6157
6158         limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit,
6159                              dquot->dq_dqb.dqb_bhardlimit);
6160         limit >>= sb->s_blocksize_bits;
6161
6162         if (limit && buf->f_blocks > limit) {
6163                 curblock = (dquot->dq_dqb.dqb_curspace +
6164                             dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits;
6165                 buf->f_blocks = limit;
6166                 buf->f_bfree = buf->f_bavail =
6167                         (buf->f_blocks > curblock) ?
6168                          (buf->f_blocks - curblock) : 0;
6169         }
6170
6171         limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit,
6172                              dquot->dq_dqb.dqb_ihardlimit);
6173         if (limit && buf->f_files > limit) {
6174                 buf->f_files = limit;
6175                 buf->f_ffree =
6176                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
6177                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
6178         }
6179
6180         spin_unlock(&dquot->dq_dqb_lock);
6181         dqput(dquot);
6182         return 0;
6183 }
6184 #endif
6185
6186 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
6187 {
6188         struct super_block *sb = dentry->d_sb;
6189         struct ext4_sb_info *sbi = EXT4_SB(sb);
6190         struct ext4_super_block *es = sbi->s_es;
6191         ext4_fsblk_t overhead = 0, resv_blocks;
6192         s64 bfree;
6193         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
6194
6195         if (!test_opt(sb, MINIX_DF))
6196                 overhead = sbi->s_overhead;
6197
6198         buf->f_type = EXT4_SUPER_MAGIC;
6199         buf->f_bsize = sb->s_blocksize;
6200         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
6201         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
6202                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
6203         /* prevent underflow in case that few free space is available */
6204         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
6205         buf->f_bavail = buf->f_bfree -
6206                         (ext4_r_blocks_count(es) + resv_blocks);
6207         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
6208                 buf->f_bavail = 0;
6209         buf->f_files = le32_to_cpu(es->s_inodes_count);
6210         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
6211         buf->f_namelen = EXT4_NAME_LEN;
6212         buf->f_fsid = uuid_to_fsid(es->s_uuid);
6213
6214 #ifdef CONFIG_QUOTA
6215         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
6216             sb_has_quota_limits_enabled(sb, PRJQUOTA))
6217                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
6218 #endif
6219         return 0;
6220 }
6221
6222
6223 #ifdef CONFIG_QUOTA
6224
6225 /*
6226  * Helper functions so that transaction is started before we acquire dqio_sem
6227  * to keep correct lock ordering of transaction > dqio_sem
6228  */
6229 static inline struct inode *dquot_to_inode(struct dquot *dquot)
6230 {
6231         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
6232 }
6233
6234 static int ext4_write_dquot(struct dquot *dquot)
6235 {
6236         int ret, err;
6237         handle_t *handle;
6238         struct inode *inode;
6239
6240         inode = dquot_to_inode(dquot);
6241         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
6242                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
6243         if (IS_ERR(handle))
6244                 return PTR_ERR(handle);
6245         ret = dquot_commit(dquot);
6246         err = ext4_journal_stop(handle);
6247         if (!ret)
6248                 ret = err;
6249         return ret;
6250 }
6251
6252 static int ext4_acquire_dquot(struct dquot *dquot)
6253 {
6254         int ret, err;
6255         handle_t *handle;
6256
6257         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
6258                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
6259         if (IS_ERR(handle))
6260                 return PTR_ERR(handle);
6261         ret = dquot_acquire(dquot);
6262         err = ext4_journal_stop(handle);
6263         if (!ret)
6264                 ret = err;
6265         return ret;
6266 }
6267
6268 static int ext4_release_dquot(struct dquot *dquot)
6269 {
6270         int ret, err;
6271         handle_t *handle;
6272
6273         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
6274                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
6275         if (IS_ERR(handle)) {
6276                 /* Release dquot anyway to avoid endless cycle in dqput() */
6277                 dquot_release(dquot);
6278                 return PTR_ERR(handle);
6279         }
6280         ret = dquot_release(dquot);
6281         err = ext4_journal_stop(handle);
6282         if (!ret)
6283                 ret = err;
6284         return ret;
6285 }
6286
6287 static int ext4_mark_dquot_dirty(struct dquot *dquot)
6288 {
6289         struct super_block *sb = dquot->dq_sb;
6290
6291         if (ext4_is_quota_journalled(sb)) {
6292                 dquot_mark_dquot_dirty(dquot);
6293                 return ext4_write_dquot(dquot);
6294         } else {
6295                 return dquot_mark_dquot_dirty(dquot);
6296         }
6297 }
6298
6299 static int ext4_write_info(struct super_block *sb, int type)
6300 {
6301         int ret, err;
6302         handle_t *handle;
6303
6304         /* Data block + inode block */
6305         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
6306         if (IS_ERR(handle))
6307                 return PTR_ERR(handle);
6308         ret = dquot_commit_info(sb, type);
6309         err = ext4_journal_stop(handle);
6310         if (!ret)
6311                 ret = err;
6312         return ret;
6313 }
6314
6315 /*
6316  * Turn on quotas during mount time - we need to find
6317  * the quota file and such...
6318  */
6319 static int ext4_quota_on_mount(struct super_block *sb, int type)
6320 {
6321         return dquot_quota_on_mount(sb, get_qf_name(sb, EXT4_SB(sb), type),
6322                                         EXT4_SB(sb)->s_jquota_fmt, type);
6323 }
6324
6325 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
6326 {
6327         struct ext4_inode_info *ei = EXT4_I(inode);
6328
6329         /* The first argument of lockdep_set_subclass has to be
6330          * *exactly* the same as the argument to init_rwsem() --- in
6331          * this case, in init_once() --- or lockdep gets unhappy
6332          * because the name of the lock is set using the
6333          * stringification of the argument to init_rwsem().
6334          */
6335         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
6336         lockdep_set_subclass(&ei->i_data_sem, subclass);
6337 }
6338
6339 /*
6340  * Standard function to be called on quota_on
6341  */
6342 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
6343                          const struct path *path)
6344 {
6345         int err;
6346
6347         if (!test_opt(sb, QUOTA))
6348                 return -EINVAL;
6349
6350         /* Quotafile not on the same filesystem? */
6351         if (path->dentry->d_sb != sb)
6352                 return -EXDEV;
6353
6354         /* Quota already enabled for this file? */
6355         if (IS_NOQUOTA(d_inode(path->dentry)))
6356                 return -EBUSY;
6357
6358         /* Journaling quota? */
6359         if (EXT4_SB(sb)->s_qf_names[type]) {
6360                 /* Quotafile not in fs root? */
6361                 if (path->dentry->d_parent != sb->s_root)
6362                         ext4_msg(sb, KERN_WARNING,
6363                                 "Quota file not on filesystem root. "
6364                                 "Journaled quota will not work");
6365                 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
6366         } else {
6367                 /*
6368                  * Clear the flag just in case mount options changed since
6369                  * last time.
6370                  */
6371                 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
6372         }
6373
6374         /*
6375          * When we journal data on quota file, we have to flush journal to see
6376          * all updates to the file when we bypass pagecache...
6377          */
6378         if (EXT4_SB(sb)->s_journal &&
6379             ext4_should_journal_data(d_inode(path->dentry))) {
6380                 /*
6381                  * We don't need to lock updates but journal_flush() could
6382                  * otherwise be livelocked...
6383                  */
6384                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
6385                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal, 0);
6386                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
6387                 if (err)
6388                         return err;
6389         }
6390
6391         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
6392         err = dquot_quota_on(sb, type, format_id, path);
6393         if (err) {
6394                 lockdep_set_quota_inode(path->dentry->d_inode,
6395                                              I_DATA_SEM_NORMAL);
6396         } else {
6397                 struct inode *inode = d_inode(path->dentry);
6398                 handle_t *handle;
6399
6400                 /*
6401                  * Set inode flags to prevent userspace from messing with quota
6402                  * files. If this fails, we return success anyway since quotas
6403                  * are already enabled and this is not a hard failure.
6404                  */
6405                 inode_lock(inode);
6406                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
6407                 if (IS_ERR(handle))
6408                         goto unlock_inode;
6409                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
6410                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
6411                                 S_NOATIME | S_IMMUTABLE);
6412                 err = ext4_mark_inode_dirty(handle, inode);
6413                 ext4_journal_stop(handle);
6414         unlock_inode:
6415                 inode_unlock(inode);
6416         }
6417         return err;
6418 }
6419
6420 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
6421                              unsigned int flags)
6422 {
6423         int err;
6424         struct inode *qf_inode;
6425         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
6426                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
6427                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
6428                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
6429         };
6430
6431         BUG_ON(!ext4_has_feature_quota(sb));
6432
6433         if (!qf_inums[type])
6434                 return -EPERM;
6435
6436         qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL);
6437         if (IS_ERR(qf_inode)) {
6438                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
6439                 return PTR_ERR(qf_inode);
6440         }
6441
6442         /* Don't account quota for quota files to avoid recursion */
6443         qf_inode->i_flags |= S_NOQUOTA;
6444         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
6445         err = dquot_load_quota_inode(qf_inode, type, format_id, flags);
6446         if (err)
6447                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
6448         iput(qf_inode);
6449
6450         return err;
6451 }
6452
6453 /* Enable usage tracking for all quota types. */
6454 static int ext4_enable_quotas(struct super_block *sb)
6455 {
6456         int type, err = 0;
6457         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
6458                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
6459                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
6460                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
6461         };
6462         bool quota_mopt[EXT4_MAXQUOTAS] = {
6463                 test_opt(sb, USRQUOTA),
6464                 test_opt(sb, GRPQUOTA),
6465                 test_opt(sb, PRJQUOTA),
6466         };
6467
6468         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
6469         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
6470                 if (qf_inums[type]) {
6471                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
6472                                 DQUOT_USAGE_ENABLED |
6473                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
6474                         if (err) {
6475                                 ext4_warning(sb,
6476                                         "Failed to enable quota tracking "
6477                                         "(type=%d, err=%d). Please run "
6478                                         "e2fsck to fix.", type, err);
6479                                 for (type--; type >= 0; type--)
6480                                         dquot_quota_off(sb, type);
6481
6482                                 return err;
6483                         }
6484                 }
6485         }
6486         return 0;
6487 }
6488
6489 static int ext4_quota_off(struct super_block *sb, int type)
6490 {
6491         struct inode *inode = sb_dqopt(sb)->files[type];
6492         handle_t *handle;
6493         int err;
6494
6495         /* Force all delayed allocation blocks to be allocated.
6496          * Caller already holds s_umount sem */
6497         if (test_opt(sb, DELALLOC))
6498                 sync_filesystem(sb);
6499
6500         if (!inode || !igrab(inode))
6501                 goto out;
6502
6503         err = dquot_quota_off(sb, type);
6504         if (err || ext4_has_feature_quota(sb))
6505                 goto out_put;
6506
6507         inode_lock(inode);
6508         /*
6509          * Update modification times of quota files when userspace can
6510          * start looking at them. If we fail, we return success anyway since
6511          * this is not a hard failure and quotas are already disabled.
6512          */
6513         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
6514         if (IS_ERR(handle)) {
6515                 err = PTR_ERR(handle);
6516                 goto out_unlock;
6517         }
6518         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
6519         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
6520         inode->i_mtime = inode->i_ctime = current_time(inode);
6521         err = ext4_mark_inode_dirty(handle, inode);
6522         ext4_journal_stop(handle);
6523 out_unlock:
6524         inode_unlock(inode);
6525 out_put:
6526         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
6527         iput(inode);
6528         return err;
6529 out:
6530         return dquot_quota_off(sb, type);
6531 }
6532
6533 /* Read data from quotafile - avoid pagecache and such because we cannot afford
6534  * acquiring the locks... As quota files are never truncated and quota code
6535  * itself serializes the operations (and no one else should touch the files)
6536  * we don't have to be afraid of races */
6537 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
6538                                size_t len, loff_t off)
6539 {
6540         struct inode *inode = sb_dqopt(sb)->files[type];
6541         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
6542         int offset = off & (sb->s_blocksize - 1);
6543         int tocopy;
6544         size_t toread;
6545         struct buffer_head *bh;
6546         loff_t i_size = i_size_read(inode);
6547
6548         if (off > i_size)
6549                 return 0;
6550         if (off+len > i_size)
6551                 len = i_size-off;
6552         toread = len;
6553         while (toread > 0) {
6554                 tocopy = sb->s_blocksize - offset < toread ?
6555                                 sb->s_blocksize - offset : toread;
6556                 bh = ext4_bread(NULL, inode, blk, 0);
6557                 if (IS_ERR(bh))
6558                         return PTR_ERR(bh);
6559                 if (!bh)        /* A hole? */
6560                         memset(data, 0, tocopy);
6561                 else
6562                         memcpy(data, bh->b_data+offset, tocopy);
6563                 brelse(bh);
6564                 offset = 0;
6565                 toread -= tocopy;
6566                 data += tocopy;
6567                 blk++;
6568         }
6569         return len;
6570 }
6571
6572 /* Write to quotafile (we know the transaction is already started and has
6573  * enough credits) */
6574 static ssize_t ext4_quota_write(struct super_block *sb, int type,
6575                                 const char *data, size_t len, loff_t off)
6576 {
6577         struct inode *inode = sb_dqopt(sb)->files[type];
6578         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
6579         int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1);
6580         int retries = 0;
6581         struct buffer_head *bh;
6582         handle_t *handle = journal_current_handle();
6583
6584         if (EXT4_SB(sb)->s_journal && !handle) {
6585                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
6586                         " cancelled because transaction is not started",
6587                         (unsigned long long)off, (unsigned long long)len);
6588                 return -EIO;
6589         }
6590         /*
6591          * Since we account only one data block in transaction credits,
6592          * then it is impossible to cross a block boundary.
6593          */
6594         if (sb->s_blocksize - offset < len) {
6595                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
6596                         " cancelled because not block aligned",
6597                         (unsigned long long)off, (unsigned long long)len);
6598                 return -EIO;
6599         }
6600
6601         do {
6602                 bh = ext4_bread(handle, inode, blk,
6603                                 EXT4_GET_BLOCKS_CREATE |
6604                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
6605         } while (PTR_ERR(bh) == -ENOSPC &&
6606                  ext4_should_retry_alloc(inode->i_sb, &retries));
6607         if (IS_ERR(bh))
6608                 return PTR_ERR(bh);
6609         if (!bh)
6610                 goto out;
6611         BUFFER_TRACE(bh, "get write access");
6612         err = ext4_journal_get_write_access(handle, bh);
6613         if (err) {
6614                 brelse(bh);
6615                 return err;
6616         }
6617         lock_buffer(bh);
6618         memcpy(bh->b_data+offset, data, len);
6619         flush_dcache_page(bh->b_page);
6620         unlock_buffer(bh);
6621         err = ext4_handle_dirty_metadata(handle, NULL, bh);
6622         brelse(bh);
6623 out:
6624         if (inode->i_size < off + len) {
6625                 i_size_write(inode, off + len);
6626                 EXT4_I(inode)->i_disksize = inode->i_size;
6627                 err2 = ext4_mark_inode_dirty(handle, inode);
6628                 if (unlikely(err2 && !err))
6629                         err = err2;
6630         }
6631         return err ? err : len;
6632 }
6633 #endif
6634
6635 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
6636                        const char *dev_name, void *data)
6637 {
6638         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
6639 }
6640
6641 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
6642 static inline void register_as_ext2(void)
6643 {
6644         int err = register_filesystem(&ext2_fs_type);
6645         if (err)
6646                 printk(KERN_WARNING
6647                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
6648 }
6649
6650 static inline void unregister_as_ext2(void)
6651 {
6652         unregister_filesystem(&ext2_fs_type);
6653 }
6654
6655 static inline int ext2_feature_set_ok(struct super_block *sb)
6656 {
6657         if (ext4_has_unknown_ext2_incompat_features(sb))
6658                 return 0;
6659         if (sb_rdonly(sb))
6660                 return 1;
6661         if (ext4_has_unknown_ext2_ro_compat_features(sb))
6662                 return 0;
6663         return 1;
6664 }
6665 #else
6666 static inline void register_as_ext2(void) { }
6667 static inline void unregister_as_ext2(void) { }
6668 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
6669 #endif
6670
6671 static inline void register_as_ext3(void)
6672 {
6673         int err = register_filesystem(&ext3_fs_type);
6674         if (err)
6675                 printk(KERN_WARNING
6676                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
6677 }
6678
6679 static inline void unregister_as_ext3(void)
6680 {
6681         unregister_filesystem(&ext3_fs_type);
6682 }
6683
6684 static inline int ext3_feature_set_ok(struct super_block *sb)
6685 {
6686         if (ext4_has_unknown_ext3_incompat_features(sb))
6687                 return 0;
6688         if (!ext4_has_feature_journal(sb))
6689                 return 0;
6690         if (sb_rdonly(sb))
6691                 return 1;
6692         if (ext4_has_unknown_ext3_ro_compat_features(sb))
6693                 return 0;
6694         return 1;
6695 }
6696
6697 static struct file_system_type ext4_fs_type = {
6698         .owner          = THIS_MODULE,
6699         .name           = "ext4",
6700         .mount          = ext4_mount,
6701         .kill_sb        = kill_block_super,
6702         .fs_flags       = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
6703 };
6704 MODULE_ALIAS_FS("ext4");
6705
6706 /* Shared across all ext4 file systems */
6707 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
6708
6709 static int __init ext4_init_fs(void)
6710 {
6711         int i, err;
6712
6713         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
6714         ext4_li_info = NULL;
6715
6716         /* Build-time check for flags consistency */
6717         ext4_check_flag_values();
6718
6719         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
6720                 init_waitqueue_head(&ext4__ioend_wq[i]);
6721
6722         err = ext4_init_es();
6723         if (err)
6724                 return err;
6725
6726         err = ext4_init_pending();
6727         if (err)
6728                 goto out7;
6729
6730         err = ext4_init_post_read_processing();
6731         if (err)
6732                 goto out6;
6733
6734         err = ext4_init_pageio();
6735         if (err)
6736                 goto out5;
6737
6738         err = ext4_init_system_zone();
6739         if (err)
6740                 goto out4;
6741
6742         err = ext4_init_sysfs();
6743         if (err)
6744                 goto out3;
6745
6746         err = ext4_init_mballoc();
6747         if (err)
6748                 goto out2;
6749         err = init_inodecache();
6750         if (err)
6751                 goto out1;
6752
6753         err = ext4_fc_init_dentry_cache();
6754         if (err)
6755                 goto out05;
6756
6757         register_as_ext3();
6758         register_as_ext2();
6759         err = register_filesystem(&ext4_fs_type);
6760         if (err)
6761                 goto out;
6762
6763         return 0;
6764 out:
6765         unregister_as_ext2();
6766         unregister_as_ext3();
6767 out05:
6768         destroy_inodecache();
6769 out1:
6770         ext4_exit_mballoc();
6771 out2:
6772         ext4_exit_sysfs();
6773 out3:
6774         ext4_exit_system_zone();
6775 out4:
6776         ext4_exit_pageio();
6777 out5:
6778         ext4_exit_post_read_processing();
6779 out6:
6780         ext4_exit_pending();
6781 out7:
6782         ext4_exit_es();
6783
6784         return err;
6785 }
6786
6787 static void __exit ext4_exit_fs(void)
6788 {
6789         ext4_destroy_lazyinit_thread();
6790         unregister_as_ext2();
6791         unregister_as_ext3();
6792         unregister_filesystem(&ext4_fs_type);
6793         destroy_inodecache();
6794         ext4_exit_mballoc();
6795         ext4_exit_sysfs();
6796         ext4_exit_system_zone();
6797         ext4_exit_pageio();
6798         ext4_exit_post_read_processing();
6799         ext4_exit_es();
6800         ext4_exit_pending();
6801 }
6802
6803 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
6804 MODULE_DESCRIPTION("Fourth Extended Filesystem");
6805 MODULE_LICENSE("GPL");
6806 MODULE_SOFTDEP("pre: crc32c");
6807 module_init(ext4_init_fs)
6808 module_exit(ext4_exit_fs)