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/uaccess.h>
  43 #include <linux/iversion.h>
  44 #include <linux/unicode.h>
  45 #include <linux/part_stat.h>
  46 #include <linux/kthread.h>
  47 #include <linux/freezer.h>
  48 #include <linux/fsnotify.h>
  49 #include <linux/fs_context.h>
  50 #include <linux/fs_parser.h>
  51
  52 #include "ext4.h"
  53 #include "ext4_extents.h"       /* Needed for trace points definition */
  54 #include "ext4_jbd2.h"
  55 #include "xattr.h"
  56 #include "acl.h"
  57 #include "mballoc.h"
  58 #include "fsmap.h"
  59
  60 #define CREATE_TRACE_POINTS
  61 #include <trace/events/ext4.h>
  62
  63 static struct ext4_lazy_init *ext4_li_info;
  64 static DEFINE_MUTEX(ext4_li_mtx);
  65 static struct ratelimit_state ext4_mount_msg_ratelimit;
  66
  67 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  68                              unsigned long journal_devnum);
  69 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  70 static void ext4_update_super(struct super_block *sb);
  71 static int ext4_commit_super(struct super_block *sb);
  72 static int ext4_mark_recovery_complete(struct super_block *sb,
  73                                         struct ext4_super_block *es);
  74 static int ext4_clear_journal_err(struct super_block *sb,
  75                                   struct ext4_super_block *es);
  76 static int ext4_sync_fs(struct super_block *sb, int wait);
  77 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  78 static int ext4_unfreeze(struct super_block *sb);
  79 static int ext4_freeze(struct super_block *sb);
  80 static inline int ext2_feature_set_ok(struct super_block *sb);
  81 static inline int ext3_feature_set_ok(struct super_block *sb);
  82 static void ext4_destroy_lazyinit_thread(void);
  83 static void ext4_unregister_li_request(struct super_block *sb);
  84 static void ext4_clear_request_list(void);
  85 static struct inode *ext4_get_journal_inode(struct super_block *sb,
  86                                             unsigned int journal_inum);
  87 static int ext4_validate_options(struct fs_context *fc);
  88 static int ext4_check_opt_consistency(struct fs_context *fc,
  89                                       struct super_block *sb);
  90 static int ext4_apply_options(struct fs_context *fc, struct super_block *sb);
  91 static int ext4_parse_param(struct fs_context *fc, struct fs_parameter *param);
  92 static int ext4_get_tree(struct fs_context *fc);
  93 static int ext4_reconfigure(struct fs_context *fc);
  94 static void ext4_fc_free(struct fs_context *fc);
  95 static int ext4_init_fs_context(struct fs_context *fc);
  96 static const struct fs_parameter_spec ext4_param_specs[];
  97
  98 /*
  99  * Lock ordering
 100  *
 101  * page fault path:
 102  * mmap_lock -> sb_start_pagefault -> invalidate_lock (r) -> transaction start
 103  *   -> page lock -> i_data_sem (rw)
 104  *
 105  * buffered write path:
 106  * sb_start_write -> i_mutex -> mmap_lock
 107  * sb_start_write -> i_mutex -> transaction start -> page lock ->
 108  *   i_data_sem (rw)
 109  *
 110  * truncate:
 111  * sb_start_write -> i_mutex -> invalidate_lock (w) -> i_mmap_rwsem (w) ->
 112  *   page lock
 113  * sb_start_write -> i_mutex -> invalidate_lock (w) -> transaction start ->
 114  *   i_data_sem (rw)
 115  *
 116  * direct IO:
 117  * sb_start_write -> i_mutex -> mmap_lock
 118  * sb_start_write -> i_mutex -> transaction start -> i_data_sem (rw)
 119  *
 120  * writepages:
 121  * transaction start -> page lock(s) -> i_data_sem (rw)
 122  */
 123
 124 static const struct fs_context_operations ext4_context_ops = {
 125         .parse_param    = ext4_parse_param,
 126         .get_tree       = ext4_get_tree,
 127         .reconfigure    = ext4_reconfigure,
 128         .free           = ext4_fc_free,
 129 };
 130
 131
 132 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
 133 static struct file_system_type ext2_fs_type = {
 134         .owner                  = THIS_MODULE,
 135         .name                   = "ext2",
 136         .init_fs_context        = ext4_init_fs_context,
 137         .parameters             = ext4_param_specs,
 138         .kill_sb                = kill_block_super,
 139         .fs_flags               = FS_REQUIRES_DEV,
 140 };
 141 MODULE_ALIAS_FS("ext2");
 142 MODULE_ALIAS("ext2");
 143 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
 144 #else
 145 #define IS_EXT2_SB(sb) (0)
 146 #endif
 147
 148
 149 static struct file_system_type ext3_fs_type = {
 150         .owner                  = THIS_MODULE,
 151         .name                   = "ext3",
 152         .init_fs_context        = ext4_init_fs_context,
 153         .parameters             = ext4_param_specs,
 154         .kill_sb                = kill_block_super,
 155         .fs_flags               = FS_REQUIRES_DEV,
 156 };
 157 MODULE_ALIAS_FS("ext3");
 158 MODULE_ALIAS("ext3");
 159 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 160
 161
 162 static inline void __ext4_read_bh(struct buffer_head *bh, int op_flags,
 163                                   bh_end_io_t *end_io)
 164 {
 165         /*
 166          * buffer's verified bit is no longer valid after reading from
 167          * disk again due to write out error, clear it to make sure we
 168          * recheck the buffer contents.
 169          */
 170         clear_buffer_verified(bh);
 171
 172         bh->b_end_io = end_io ? end_io : end_buffer_read_sync;
 173         get_bh(bh);
 174         submit_bh(REQ_OP_READ, op_flags, bh);
 175 }
 176
 177 void ext4_read_bh_nowait(struct buffer_head *bh, int op_flags,
 178                          bh_end_io_t *end_io)
 179 {
 180         BUG_ON(!buffer_locked(bh));
 181
 182         if (ext4_buffer_uptodate(bh)) {
 183                 unlock_buffer(bh);
 184                 return;
 185         }
 186         __ext4_read_bh(bh, op_flags, end_io);
 187 }
 188
 189 int ext4_read_bh(struct buffer_head *bh, int op_flags, bh_end_io_t *end_io)
 190 {
 191         BUG_ON(!buffer_locked(bh));
 192
 193         if (ext4_buffer_uptodate(bh)) {
 194                 unlock_buffer(bh);
 195                 return 0;
 196         }
 197
 198         __ext4_read_bh(bh, op_flags, end_io);
 199
 200         wait_on_buffer(bh);
 201         if (buffer_uptodate(bh))
 202                 return 0;
 203         return -EIO;
 204 }
 205
 206 int ext4_read_bh_lock(struct buffer_head *bh, int op_flags, bool wait)
 207 {
 208         if (trylock_buffer(bh)) {
 209                 if (wait)
 210                         return ext4_read_bh(bh, op_flags, NULL);
 211                 ext4_read_bh_nowait(bh, op_flags, NULL);
 212                 return 0;
 213         }
 214         if (wait) {
 215                 wait_on_buffer(bh);
 216                 if (buffer_uptodate(bh))
 217                         return 0;
 218                 return -EIO;
 219         }
 220         return 0;
 221 }
 222
 223 /*
 224  * This works like __bread_gfp() except it uses ERR_PTR for error
 225  * returns.  Currently with sb_bread it's impossible to distinguish
 226  * between ENOMEM and EIO situations (since both result in a NULL
 227  * return.
 228  */
 229 static struct buffer_head *__ext4_sb_bread_gfp(struct super_block *sb,
 230                                                sector_t block, int op_flags,
 231                                                gfp_t gfp)
 232 {
 233         struct buffer_head *bh;
 234         int ret;
 235
 236         bh = sb_getblk_gfp(sb, block, gfp);
 237         if (bh == NULL)
 238                 return ERR_PTR(-ENOMEM);
 239         if (ext4_buffer_uptodate(bh))
 240                 return bh;
 241
 242         ret = ext4_read_bh_lock(bh, REQ_META | op_flags, true);
 243         if (ret) {
 244                 put_bh(bh);
 245                 return ERR_PTR(ret);
 246         }
 247         return bh;
 248 }
 249
 250 struct buffer_head *ext4_sb_bread(struct super_block *sb, sector_t block,
 251                                    int op_flags)
 252 {
 253         return __ext4_sb_bread_gfp(sb, block, op_flags, __GFP_MOVABLE);
 254 }
 255
 256 struct buffer_head *ext4_sb_bread_unmovable(struct super_block *sb,
 257                                             sector_t block)
 258 {
 259         return __ext4_sb_bread_gfp(sb, block, 0, 0);
 260 }
 261
 262 void ext4_sb_breadahead_unmovable(struct super_block *sb, sector_t block)
 263 {
 264         struct buffer_head *bh = sb_getblk_gfp(sb, block, 0);
 265
 266         if (likely(bh)) {
 267                 ext4_read_bh_lock(bh, REQ_RAHEAD, false);
 268                 brelse(bh);
 269         }
 270 }
 271
 272 static int ext4_verify_csum_type(struct super_block *sb,
 273                                  struct ext4_super_block *es)
 274 {
 275         if (!ext4_has_feature_metadata_csum(sb))
 276                 return 1;
 277
 278         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 279 }
 280
 281 __le32 ext4_superblock_csum(struct super_block *sb,
 282                             struct ext4_super_block *es)
 283 {
 284         struct ext4_sb_info *sbi = EXT4_SB(sb);
 285         int offset = offsetof(struct ext4_super_block, s_checksum);
 286         __u32 csum;
 287
 288         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 289
 290         return cpu_to_le32(csum);
 291 }
 292
 293 static int ext4_superblock_csum_verify(struct super_block *sb,
 294                                        struct ext4_super_block *es)
 295 {
 296         if (!ext4_has_metadata_csum(sb))
 297                 return 1;
 298
 299         return es->s_checksum == ext4_superblock_csum(sb, es);
 300 }
 301
 302 void ext4_superblock_csum_set(struct super_block *sb)
 303 {
 304         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 305
 306         if (!ext4_has_metadata_csum(sb))
 307                 return;
 308
 309         es->s_checksum = ext4_superblock_csum(sb, es);
 310 }
 311
 312 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 313                                struct ext4_group_desc *bg)
 314 {
 315         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 316                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 317                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 318 }
 319
 320 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 321                                struct ext4_group_desc *bg)
 322 {
 323         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 324                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 325                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 326 }
 327
 328 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 329                               struct ext4_group_desc *bg)
 330 {
 331         return le32_to_cpu(bg->bg_inode_table_lo) |
 332                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 333                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 334 }
 335
 336 __u32 ext4_free_group_clusters(struct super_block *sb,
 337                                struct ext4_group_desc *bg)
 338 {
 339         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 340                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 341                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 342 }
 343
 344 __u32 ext4_free_inodes_count(struct super_block *sb,
 345                               struct ext4_group_desc *bg)
 346 {
 347         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 348                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 349                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 350 }
 351
 352 __u32 ext4_used_dirs_count(struct super_block *sb,
 353                               struct ext4_group_desc *bg)
 354 {
 355         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 356                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 357                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 358 }
 359
 360 __u32 ext4_itable_unused_count(struct super_block *sb,
 361                               struct ext4_group_desc *bg)
 362 {
 363         return le16_to_cpu(bg->bg_itable_unused_lo) |
 364                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 365                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 366 }
 367
 368 void ext4_block_bitmap_set(struct super_block *sb,
 369                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 370 {
 371         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 372         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 373                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 374 }
 375
 376 void ext4_inode_bitmap_set(struct super_block *sb,
 377                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 378 {
 379         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 380         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 381                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 382 }
 383
 384 void ext4_inode_table_set(struct super_block *sb,
 385                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 386 {
 387         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 388         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 389                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 390 }
 391
 392 void ext4_free_group_clusters_set(struct super_block *sb,
 393                                   struct ext4_group_desc *bg, __u32 count)
 394 {
 395         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 396         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 397                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 398 }
 399
 400 void ext4_free_inodes_set(struct super_block *sb,
 401                           struct ext4_group_desc *bg, __u32 count)
 402 {
 403         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 404         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 405                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 406 }
 407
 408 void ext4_used_dirs_set(struct super_block *sb,
 409                           struct ext4_group_desc *bg, __u32 count)
 410 {
 411         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 412         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 413                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 414 }
 415
 416 void ext4_itable_unused_set(struct super_block *sb,
 417                           struct ext4_group_desc *bg, __u32 count)
 418 {
 419         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 420         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 421                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 422 }
 423
 424 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi, time64_t now)
 425 {
 426         now = clamp_val(now, 0, (1ull << 40) - 1);
 427
 428         *lo = cpu_to_le32(lower_32_bits(now));
 429         *hi = upper_32_bits(now);
 430 }
 431
 432 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi)
 433 {
 434         return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo);
 435 }
 436 #define ext4_update_tstamp(es, tstamp) \
 437         __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi, \
 438                              ktime_get_real_seconds())
 439 #define ext4_get_tstamp(es, tstamp) \
 440         __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
 441
 442 /*
 443  * The del_gendisk() function uninitializes the disk-specific data
 444  * structures, including the bdi structure, without telling anyone
 445  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 446  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 447  * This is a kludge to prevent these oops until we can put in a proper
 448  * hook in del_gendisk() to inform the VFS and file system layers.
 449  */
 450 static int block_device_ejected(struct super_block *sb)
 451 {
 452         struct inode *bd_inode = sb->s_bdev->bd_inode;
 453         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 454
 455         return bdi->dev == NULL;
 456 }
 457
 458 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 459 {
 460         struct super_block              *sb = journal->j_private;
 461         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 462         int                             error = is_journal_aborted(journal);
 463         struct ext4_journal_cb_entry    *jce;
 464
 465         BUG_ON(txn->t_state == T_FINISHED);
 466
 467         ext4_process_freed_data(sb, txn->t_tid);
 468
 469         spin_lock(&sbi->s_md_lock);
 470         while (!list_empty(&txn->t_private_list)) {
 471                 jce = list_entry(txn->t_private_list.next,
 472                                  struct ext4_journal_cb_entry, jce_list);
 473                 list_del_init(&jce->jce_list);
 474                 spin_unlock(&sbi->s_md_lock);
 475                 jce->jce_func(sb, jce, error);
 476                 spin_lock(&sbi->s_md_lock);
 477         }
 478         spin_unlock(&sbi->s_md_lock);
 479 }
 480
 481 /*
 482  * This writepage callback for write_cache_pages()
 483  * takes care of a few cases after page cleaning.
 484  *
 485  * write_cache_pages() already checks for dirty pages
 486  * and calls clear_page_dirty_for_io(), which we want,
 487  * to write protect the pages.
 488  *
 489  * However, we may have to redirty a page (see below.)
 490  */
 491 static int ext4_journalled_writepage_callback(struct page *page,
 492                                               struct writeback_control *wbc,
 493                                               void *data)
 494 {
 495         transaction_t *transaction = (transaction_t *) data;
 496         struct buffer_head *bh, *head;
 497         struct journal_head *jh;
 498
 499         bh = head = page_buffers(page);
 500         do {
 501                 /*
 502                  * We have to redirty a page in these cases:
 503                  * 1) If buffer is dirty, it means the page was dirty because it
 504                  * contains a buffer that needs checkpointing. So the dirty bit
 505                  * needs to be preserved so that checkpointing writes the buffer
 506                  * properly.
 507                  * 2) If buffer is not part of the committing transaction
 508                  * (we may have just accidentally come across this buffer because
 509                  * inode range tracking is not exact) or if the currently running
 510                  * transaction already contains this buffer as well, dirty bit
 511                  * needs to be preserved so that the buffer gets writeprotected
 512                  * properly on running transaction's commit.
 513                  */
 514                 jh = bh2jh(bh);
 515                 if (buffer_dirty(bh) ||
 516                     (jh && (jh->b_transaction != transaction ||
 517                             jh->b_next_transaction))) {
 518                         redirty_page_for_writepage(wbc, page);
 519                         goto out;
 520                 }
 521         } while ((bh = bh->b_this_page) != head);
 522
 523 out:
 524         return AOP_WRITEPAGE_ACTIVATE;
 525 }
 526
 527 static int ext4_journalled_submit_inode_data_buffers(struct jbd2_inode *jinode)
 528 {
 529         struct address_space *mapping = jinode->i_vfs_inode->i_mapping;
 530         struct writeback_control wbc = {
 531                 .sync_mode =  WB_SYNC_ALL,
 532                 .nr_to_write = LONG_MAX,
 533                 .range_start = jinode->i_dirty_start,
 534                 .range_end = jinode->i_dirty_end,
 535         };
 536
 537         return write_cache_pages(mapping, &wbc,
 538                                  ext4_journalled_writepage_callback,
 539                                  jinode->i_transaction);
 540 }
 541
 542 static int ext4_journal_submit_inode_data_buffers(struct jbd2_inode *jinode)
 543 {
 544         int ret;
 545
 546         if (ext4_should_journal_data(jinode->i_vfs_inode))
 547                 ret = ext4_journalled_submit_inode_data_buffers(jinode);
 548         else
 549                 ret = jbd2_journal_submit_inode_data_buffers(jinode);
 550
 551         return ret;
 552 }
 553
 554 static int ext4_journal_finish_inode_data_buffers(struct jbd2_inode *jinode)
 555 {
 556         int ret = 0;
 557
 558         if (!ext4_should_journal_data(jinode->i_vfs_inode))
 559                 ret = jbd2_journal_finish_inode_data_buffers(jinode);
 560
 561         return ret;
 562 }
 563
 564 static bool system_going_down(void)
 565 {
 566         return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF
 567                 || system_state == SYSTEM_RESTART;
 568 }
 569
 570 struct ext4_err_translation {
 571         int code;
 572         int errno;
 573 };
 574
 575 #define EXT4_ERR_TRANSLATE(err) { .code = EXT4_ERR_##err, .errno = err }
 576
 577 static struct ext4_err_translation err_translation[] = {
 578         EXT4_ERR_TRANSLATE(EIO),
 579         EXT4_ERR_TRANSLATE(ENOMEM),
 580         EXT4_ERR_TRANSLATE(EFSBADCRC),
 581         EXT4_ERR_TRANSLATE(EFSCORRUPTED),
 582         EXT4_ERR_TRANSLATE(ENOSPC),
 583         EXT4_ERR_TRANSLATE(ENOKEY),
 584         EXT4_ERR_TRANSLATE(EROFS),
 585         EXT4_ERR_TRANSLATE(EFBIG),
 586         EXT4_ERR_TRANSLATE(EEXIST),
 587         EXT4_ERR_TRANSLATE(ERANGE),
 588         EXT4_ERR_TRANSLATE(EOVERFLOW),
 589         EXT4_ERR_TRANSLATE(EBUSY),
 590         EXT4_ERR_TRANSLATE(ENOTDIR),
 591         EXT4_ERR_TRANSLATE(ENOTEMPTY),
 592         EXT4_ERR_TRANSLATE(ESHUTDOWN),
 593         EXT4_ERR_TRANSLATE(EFAULT),
 594 };
 595
 596 static int ext4_errno_to_code(int errno)
 597 {
 598         int i;
 599
 600         for (i = 0; i < ARRAY_SIZE(err_translation); i++)
 601                 if (err_translation[i].errno == errno)
 602                         return err_translation[i].code;
 603         return EXT4_ERR_UNKNOWN;
 604 }
 605
 606 static void save_error_info(struct super_block *sb, int error,
 607                             __u32 ino, __u64 block,
 608                             const char *func, unsigned int line)
 609 {
 610         struct ext4_sb_info *sbi = EXT4_SB(sb);
 611
 612         /* We default to EFSCORRUPTED error... */
 613         if (error == 0)
 614                 error = EFSCORRUPTED;
 615
 616         spin_lock(&sbi->s_error_lock);
 617         sbi->s_add_error_count++;
 618         sbi->s_last_error_code = error;
 619         sbi->s_last_error_line = line;
 620         sbi->s_last_error_ino = ino;
 621         sbi->s_last_error_block = block;
 622         sbi->s_last_error_func = func;
 623         sbi->s_last_error_time = ktime_get_real_seconds();
 624         if (!sbi->s_first_error_time) {
 625                 sbi->s_first_error_code = error;
 626                 sbi->s_first_error_line = line;
 627                 sbi->s_first_error_ino = ino;
 628                 sbi->s_first_error_block = block;
 629                 sbi->s_first_error_func = func;
 630                 sbi->s_first_error_time = sbi->s_last_error_time;
 631         }
 632         spin_unlock(&sbi->s_error_lock);
 633 }
 634
 635 /* Deal with the reporting of failure conditions on a filesystem such as
 636  * inconsistencies detected or read IO failures.
 637  *
 638  * On ext2, we can store the error state of the filesystem in the
 639  * superblock.  That is not possible on ext4, because we may have other
 640  * write ordering constraints on the superblock which prevent us from
 641  * writing it out straight away; and given that the journal is about to
 642  * be aborted, we can't rely on the current, or future, transactions to
 643  * write out the superblock safely.
 644  *
 645  * We'll just use the jbd2_journal_abort() error code to record an error in
 646  * the journal instead.  On recovery, the journal will complain about
 647  * that error until we've noted it down and cleared it.
 648  *
 649  * If force_ro is set, we unconditionally force the filesystem into an
 650  * ABORT|READONLY state, unless the error response on the fs has been set to
 651  * panic in which case we take the easy way out and panic immediately. This is
 652  * used to deal with unrecoverable failures such as journal IO errors or ENOMEM
 653  * at a critical moment in log management.
 654  */
 655 static void ext4_handle_error(struct super_block *sb, bool force_ro, int error,
 656                               __u32 ino, __u64 block,
 657                               const char *func, unsigned int line)
 658 {
 659         journal_t *journal = EXT4_SB(sb)->s_journal;
 660         bool continue_fs = !force_ro && test_opt(sb, ERRORS_CONT);
 661
 662         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 663         if (test_opt(sb, WARN_ON_ERROR))
 664                 WARN_ON_ONCE(1);
 665
 666         if (!continue_fs && !sb_rdonly(sb)) {
 667                 ext4_set_mount_flag(sb, EXT4_MF_FS_ABORTED);
 668                 if (journal)
 669                         jbd2_journal_abort(journal, -EIO);
 670         }
 671
 672         if (!bdev_read_only(sb->s_bdev)) {
 673                 save_error_info(sb, error, ino, block, func, line);
 674                 /*
 675                  * In case the fs should keep running, we need to writeout
 676                  * superblock through the journal. Due to lock ordering
 677                  * constraints, it may not be safe to do it right here so we
 678                  * defer superblock flushing to a workqueue.
 679                  */
 680                 if (continue_fs && journal)
 681                         schedule_work(&EXT4_SB(sb)->s_error_work);
 682                 else
 683                         ext4_commit_super(sb);
 684         }
 685
 686         /*
 687          * We force ERRORS_RO behavior when system is rebooting. Otherwise we
 688          * could panic during 'reboot -f' as the underlying device got already
 689          * disabled.
 690          */
 691         if (test_opt(sb, ERRORS_PANIC) && !system_going_down()) {
 692                 panic("EXT4-fs (device %s): panic forced after error\n",
 693                         sb->s_id);
 694         }
 695
 696         if (sb_rdonly(sb) || continue_fs)
 697                 return;
 698
 699         ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 700         /*
 701          * Make sure updated value of ->s_mount_flags will be visible before
 702          * ->s_flags update
 703          */
 704         smp_wmb();
 705         sb->s_flags |= SB_RDONLY;
 706 }
 707
 708 static void flush_stashed_error_work(struct work_struct *work)
 709 {
 710         struct ext4_sb_info *sbi = container_of(work, struct ext4_sb_info,
 711                                                 s_error_work);
 712         journal_t *journal = sbi->s_journal;
 713         handle_t *handle;
 714
 715         /*
 716          * If the journal is still running, we have to write out superblock
 717          * through the journal to avoid collisions of other journalled sb
 718          * updates.
 719          *
 720          * We use directly jbd2 functions here to avoid recursing back into
 721          * ext4 error handling code during handling of previous errors.
 722          */
 723         if (!sb_rdonly(sbi->s_sb) && journal) {
 724                 struct buffer_head *sbh = sbi->s_sbh;
 725                 handle = jbd2_journal_start(journal, 1);
 726                 if (IS_ERR(handle))
 727                         goto write_directly;
 728                 if (jbd2_journal_get_write_access(handle, sbh)) {
 729                         jbd2_journal_stop(handle);
 730                         goto write_directly;
 731                 }
 732                 ext4_update_super(sbi->s_sb);
 733                 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
 734                         ext4_msg(sbi->s_sb, KERN_ERR, "previous I/O error to "
 735                                  "superblock detected");
 736                         clear_buffer_write_io_error(sbh);
 737                         set_buffer_uptodate(sbh);
 738                 }
 739
 740                 if (jbd2_journal_dirty_metadata(handle, sbh)) {
 741                         jbd2_journal_stop(handle);
 742                         goto write_directly;
 743                 }
 744                 jbd2_journal_stop(handle);
 745                 ext4_notify_error_sysfs(sbi);
 746                 return;
 747         }
 748 write_directly:
 749         /*
 750          * Write through journal failed. Write sb directly to get error info
 751          * out and hope for the best.
 752          */
 753         ext4_commit_super(sbi->s_sb);
 754         ext4_notify_error_sysfs(sbi);
 755 }
 756
 757 #define ext4_error_ratelimit(sb)                                        \
 758                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 759                              "EXT4-fs error")
 760
 761 void __ext4_error(struct super_block *sb, const char *function,
 762                   unsigned int line, bool force_ro, int error, __u64 block,
 763                   const char *fmt, ...)
 764 {
 765         struct va_format vaf;
 766         va_list args;
 767
 768         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 769                 return;
 770
 771         trace_ext4_error(sb, function, line);
 772         if (ext4_error_ratelimit(sb)) {
 773                 va_start(args, fmt);
 774                 vaf.fmt = fmt;
 775                 vaf.va = &args;
 776                 printk(KERN_CRIT
 777                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 778                        sb->s_id, function, line, current->comm, &vaf);
 779                 va_end(args);
 780         }
 781         fsnotify_sb_error(sb, NULL, error ? error : EFSCORRUPTED);
 782
 783         ext4_handle_error(sb, force_ro, error, 0, block, function, line);
 784 }
 785
 786 void __ext4_error_inode(struct inode *inode, const char *function,
 787                         unsigned int line, ext4_fsblk_t block, int error,
 788                         const char *fmt, ...)
 789 {
 790         va_list args;
 791         struct va_format vaf;
 792
 793         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 794                 return;
 795
 796         trace_ext4_error(inode->i_sb, function, line);
 797         if (ext4_error_ratelimit(inode->i_sb)) {
 798                 va_start(args, fmt);
 799                 vaf.fmt = fmt;
 800                 vaf.va = &args;
 801                 if (block)
 802                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 803                                "inode #%lu: block %llu: comm %s: %pV\n",
 804                                inode->i_sb->s_id, function, line, inode->i_ino,
 805                                block, current->comm, &vaf);
 806                 else
 807                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 808                                "inode #%lu: comm %s: %pV\n",
 809                                inode->i_sb->s_id, function, line, inode->i_ino,
 810                                current->comm, &vaf);
 811                 va_end(args);
 812         }
 813         fsnotify_sb_error(inode->i_sb, inode, error ? error : EFSCORRUPTED);
 814
 815         ext4_handle_error(inode->i_sb, false, error, inode->i_ino, block,
 816                           function, line);
 817 }
 818
 819 void __ext4_error_file(struct file *file, const char *function,
 820                        unsigned int line, ext4_fsblk_t block,
 821                        const char *fmt, ...)
 822 {
 823         va_list args;
 824         struct va_format vaf;
 825         struct inode *inode = file_inode(file);
 826         char pathname[80], *path;
 827
 828         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 829                 return;
 830
 831         trace_ext4_error(inode->i_sb, function, line);
 832         if (ext4_error_ratelimit(inode->i_sb)) {
 833                 path = file_path(file, pathname, sizeof(pathname));
 834                 if (IS_ERR(path))
 835                         path = "(unknown)";
 836                 va_start(args, fmt);
 837                 vaf.fmt = fmt;
 838                 vaf.va = &args;
 839                 if (block)
 840                         printk(KERN_CRIT
 841                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 842                                "block %llu: comm %s: path %s: %pV\n",
 843                                inode->i_sb->s_id, function, line, inode->i_ino,
 844                                block, current->comm, path, &vaf);
 845                 else
 846                         printk(KERN_CRIT
 847                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 848                                "comm %s: path %s: %pV\n",
 849                                inode->i_sb->s_id, function, line, inode->i_ino,
 850                                current->comm, path, &vaf);
 851                 va_end(args);
 852         }
 853         fsnotify_sb_error(inode->i_sb, inode, EFSCORRUPTED);
 854
 855         ext4_handle_error(inode->i_sb, false, EFSCORRUPTED, inode->i_ino, block,
 856                           function, line);
 857 }
 858
 859 const char *ext4_decode_error(struct super_block *sb, int errno,
 860                               char nbuf[16])
 861 {
 862         char *errstr = NULL;
 863
 864         switch (errno) {
 865         case -EFSCORRUPTED:
 866                 errstr = "Corrupt filesystem";
 867                 break;
 868         case -EFSBADCRC:
 869                 errstr = "Filesystem failed CRC";
 870                 break;
 871         case -EIO:
 872                 errstr = "IO failure";
 873                 break;
 874         case -ENOMEM:
 875                 errstr = "Out of memory";
 876                 break;
 877         case -EROFS:
 878                 if (!sb || (EXT4_SB(sb)->s_journal &&
 879                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 880                         errstr = "Journal has aborted";
 881                 else
 882                         errstr = "Readonly filesystem";
 883                 break;
 884         default:
 885                 /* If the caller passed in an extra buffer for unknown
 886                  * errors, textualise them now.  Else we just return
 887                  * NULL. */
 888                 if (nbuf) {
 889                         /* Check for truncated error codes... */
 890                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 891                                 errstr = nbuf;
 892                 }
 893                 break;
 894         }
 895
 896         return errstr;
 897 }
 898
 899 /* __ext4_std_error decodes expected errors from journaling functions
 900  * automatically and invokes the appropriate error response.  */
 901
 902 void __ext4_std_error(struct super_block *sb, const char *function,
 903                       unsigned int line, int errno)
 904 {
 905         char nbuf[16];
 906         const char *errstr;
 907
 908         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 909                 return;
 910
 911         /* Special case: if the error is EROFS, and we're not already
 912          * inside a transaction, then there's really no point in logging
 913          * an error. */
 914         if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb))
 915                 return;
 916
 917         if (ext4_error_ratelimit(sb)) {
 918                 errstr = ext4_decode_error(sb, errno, nbuf);
 919                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 920                        sb->s_id, function, line, errstr);
 921         }
 922         fsnotify_sb_error(sb, NULL, errno ? errno : EFSCORRUPTED);
 923
 924         ext4_handle_error(sb, false, -errno, 0, 0, function, line);
 925 }
 926
 927 void __ext4_msg(struct super_block *sb,
 928                 const char *prefix, const char *fmt, ...)
 929 {
 930         struct va_format vaf;
 931         va_list args;
 932
 933         if (sb) {
 934                 atomic_inc(&EXT4_SB(sb)->s_msg_count);
 935                 if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state),
 936                                   "EXT4-fs"))
 937                         return;
 938         }
 939
 940         va_start(args, fmt);
 941         vaf.fmt = fmt;
 942         vaf.va = &args;
 943         if (sb)
 944                 printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 945         else
 946                 printk("%sEXT4-fs: %pV\n", prefix, &vaf);
 947         va_end(args);
 948 }
 949
 950 static int ext4_warning_ratelimit(struct super_block *sb)
 951 {
 952         atomic_inc(&EXT4_SB(sb)->s_warning_count);
 953         return ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state),
 954                             "EXT4-fs warning");
 955 }
 956
 957 void __ext4_warning(struct super_block *sb, const char *function,
 958                     unsigned int line, const char *fmt, ...)
 959 {
 960         struct va_format vaf;
 961         va_list args;
 962
 963         if (!ext4_warning_ratelimit(sb))
 964                 return;
 965
 966         va_start(args, fmt);
 967         vaf.fmt = fmt;
 968         vaf.va = &args;
 969         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 970                sb->s_id, function, line, &vaf);
 971         va_end(args);
 972 }
 973
 974 void __ext4_warning_inode(const struct inode *inode, const char *function,
 975                           unsigned int line, const char *fmt, ...)
 976 {
 977         struct va_format vaf;
 978         va_list args;
 979
 980         if (!ext4_warning_ratelimit(inode->i_sb))
 981                 return;
 982
 983         va_start(args, fmt);
 984         vaf.fmt = fmt;
 985         vaf.va = &args;
 986         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 987                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 988                function, line, inode->i_ino, current->comm, &vaf);
 989         va_end(args);
 990 }
 991
 992 void __ext4_grp_locked_error(const char *function, unsigned int line,
 993                              struct super_block *sb, ext4_group_t grp,
 994                              unsigned long ino, ext4_fsblk_t block,
 995                              const char *fmt, ...)
 996 __releases(bitlock)
 997 __acquires(bitlock)
 998 {
 999         struct va_format vaf;
1000         va_list args;
1001
1002         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
1003                 return;
1004
1005         trace_ext4_error(sb, function, line);
1006         if (ext4_error_ratelimit(sb)) {
1007                 va_start(args, fmt);
1008                 vaf.fmt = fmt;
1009                 vaf.va = &args;
1010                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
1011                        sb->s_id, function, line, grp);
1012                 if (ino)
1013                         printk(KERN_CONT "inode %lu: ", ino);
1014                 if (block)
1015                         printk(KERN_CONT "block %llu:",
1016                                (unsigned long long) block);
1017                 printk(KERN_CONT "%pV\n", &vaf);
1018                 va_end(args);
1019         }
1020
1021         if (test_opt(sb, ERRORS_CONT)) {
1022                 if (test_opt(sb, WARN_ON_ERROR))
1023                         WARN_ON_ONCE(1);
1024                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
1025                 if (!bdev_read_only(sb->s_bdev)) {
1026                         save_error_info(sb, EFSCORRUPTED, ino, block, function,
1027                                         line);
1028                         schedule_work(&EXT4_SB(sb)->s_error_work);
1029                 }
1030                 return;
1031         }
1032         ext4_unlock_group(sb, grp);
1033         ext4_handle_error(sb, false, EFSCORRUPTED, ino, block, function, line);
1034         /*
1035          * We only get here in the ERRORS_RO case; relocking the group
1036          * may be dangerous, but nothing bad will happen since the
1037          * filesystem will have already been marked read/only and the
1038          * journal has been aborted.  We return 1 as a hint to callers
1039          * who might what to use the return value from
1040          * ext4_grp_locked_error() to distinguish between the
1041          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
1042          * aggressively from the ext4 function in question, with a
1043          * more appropriate error code.
1044          */
1045         ext4_lock_group(sb, grp);
1046         return;
1047 }
1048
1049 void ext4_mark_group_bitmap_corrupted(struct super_block *sb,
1050                                      ext4_group_t group,
1051                                      unsigned int flags)
1052 {
1053         struct ext4_sb_info *sbi = EXT4_SB(sb);
1054         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1055         struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
1056         int ret;
1057
1058         if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) {
1059                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT,
1060                                             &grp->bb_state);
1061                 if (!ret)
1062                         percpu_counter_sub(&sbi->s_freeclusters_counter,
1063                                            grp->bb_free);
1064         }
1065
1066         if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) {
1067                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT,
1068                                             &grp->bb_state);
1069                 if (!ret && gdp) {
1070                         int count;
1071
1072                         count = ext4_free_inodes_count(sb, gdp);
1073                         percpu_counter_sub(&sbi->s_freeinodes_counter,
1074                                            count);
1075                 }
1076         }
1077 }
1078
1079 void ext4_update_dynamic_rev(struct super_block *sb)
1080 {
1081         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1082
1083         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
1084                 return;
1085
1086         ext4_warning(sb,
1087                      "updating to rev %d because of new feature flag, "
1088                      "running e2fsck is recommended",
1089                      EXT4_DYNAMIC_REV);
1090
1091         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
1092         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
1093         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
1094         /* leave es->s_feature_*compat flags alone */
1095         /* es->s_uuid will be set by e2fsck if empty */
1096
1097         /*
1098          * The rest of the superblock fields should be zero, and if not it
1099          * means they are likely already in use, so leave them alone.  We
1100          * can leave it up to e2fsck to clean up any inconsistencies there.
1101          */
1102 }
1103
1104 /*
1105  * Open the external journal device
1106  */
1107 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
1108 {
1109         struct block_device *bdev;
1110
1111         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
1112         if (IS_ERR(bdev))
1113                 goto fail;
1114         return bdev;
1115
1116 fail:
1117         ext4_msg(sb, KERN_ERR,
1118                  "failed to open journal device unknown-block(%u,%u) %ld",
1119                  MAJOR(dev), MINOR(dev), PTR_ERR(bdev));
1120         return NULL;
1121 }
1122
1123 /*
1124  * Release the journal device
1125  */
1126 static void ext4_blkdev_put(struct block_device *bdev)
1127 {
1128         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1129 }
1130
1131 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
1132 {
1133         struct block_device *bdev;
1134         bdev = sbi->s_journal_bdev;
1135         if (bdev) {
1136                 ext4_blkdev_put(bdev);
1137                 sbi->s_journal_bdev = NULL;
1138         }
1139 }
1140
1141 static inline struct inode *orphan_list_entry(struct list_head *l)
1142 {
1143         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
1144 }
1145
1146 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
1147 {
1148         struct list_head *l;
1149
1150         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
1151                  le32_to_cpu(sbi->s_es->s_last_orphan));
1152
1153         printk(KERN_ERR "sb_info orphan list:\n");
1154         list_for_each(l, &sbi->s_orphan) {
1155                 struct inode *inode = orphan_list_entry(l);
1156                 printk(KERN_ERR "  "
1157                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
1158                        inode->i_sb->s_id, inode->i_ino, inode,
1159                        inode->i_mode, inode->i_nlink,
1160                        NEXT_ORPHAN(inode));
1161         }
1162 }
1163
1164 #ifdef CONFIG_QUOTA
1165 static int ext4_quota_off(struct super_block *sb, int type);
1166
1167 static inline void ext4_quota_off_umount(struct super_block *sb)
1168 {
1169         int type;
1170
1171         /* Use our quota_off function to clear inode flags etc. */
1172         for (type = 0; type < EXT4_MAXQUOTAS; type++)
1173                 ext4_quota_off(sb, type);
1174 }
1175
1176 /*
1177  * This is a helper function which is used in the mount/remount
1178  * codepaths (which holds s_umount) to fetch the quota file name.
1179  */
1180 static inline char *get_qf_name(struct super_block *sb,
1181                                 struct ext4_sb_info *sbi,
1182                                 int type)
1183 {
1184         return rcu_dereference_protected(sbi->s_qf_names[type],
1185                                          lockdep_is_held(&sb->s_umount));
1186 }
1187 #else
1188 static inline void ext4_quota_off_umount(struct super_block *sb)
1189 {
1190 }
1191 #endif
1192
1193 static void ext4_put_super(struct super_block *sb)
1194 {
1195         struct ext4_sb_info *sbi = EXT4_SB(sb);
1196         struct ext4_super_block *es = sbi->s_es;
1197         struct buffer_head **group_desc;
1198         struct flex_groups **flex_groups;
1199         int aborted = 0;
1200         int i, err;
1201
1202         /*
1203          * Unregister sysfs before destroying jbd2 journal.
1204          * Since we could still access attr_journal_task attribute via sysfs
1205          * path which could have sbi->s_journal->j_task as NULL
1206          * Unregister sysfs before flush sbi->s_error_work.
1207          * Since user may read /proc/fs/ext4/xx/mb_groups during umount, If
1208          * read metadata verify failed then will queue error work.
1209          * flush_stashed_error_work will call start_this_handle may trigger
1210          * BUG_ON.
1211          */
1212         ext4_unregister_sysfs(sb);
1213
1214         ext4_unregister_li_request(sb);
1215         ext4_quota_off_umount(sb);
1216
1217         flush_work(&sbi->s_error_work);
1218         destroy_workqueue(sbi->rsv_conversion_wq);
1219         ext4_release_orphan_info(sb);
1220
1221         if (sbi->s_journal) {
1222                 aborted = is_journal_aborted(sbi->s_journal);
1223                 err = jbd2_journal_destroy(sbi->s_journal);
1224                 sbi->s_journal = NULL;
1225                 if ((err < 0) && !aborted) {
1226                         ext4_abort(sb, -err, "Couldn't clean up the journal");
1227                 }
1228         }
1229
1230         ext4_es_unregister_shrinker(sbi);
1231         del_timer_sync(&sbi->s_err_report);
1232         ext4_release_system_zone(sb);
1233         ext4_mb_release(sb);
1234         ext4_ext_release(sb);
1235
1236         if (!sb_rdonly(sb) && !aborted) {
1237                 ext4_clear_feature_journal_needs_recovery(sb);
1238                 ext4_clear_feature_orphan_present(sb);
1239                 es->s_state = cpu_to_le16(sbi->s_mount_state);
1240         }
1241         if (!sb_rdonly(sb))
1242                 ext4_commit_super(sb);
1243
1244         rcu_read_lock();
1245         group_desc = rcu_dereference(sbi->s_group_desc);
1246         for (i = 0; i < sbi->s_gdb_count; i++)
1247                 brelse(group_desc[i]);
1248         kvfree(group_desc);
1249         flex_groups = rcu_dereference(sbi->s_flex_groups);
1250         if (flex_groups) {
1251                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
1252                         kvfree(flex_groups[i]);
1253                 kvfree(flex_groups);
1254         }
1255         rcu_read_unlock();
1256         percpu_counter_destroy(&sbi->s_freeclusters_counter);
1257         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1258         percpu_counter_destroy(&sbi->s_dirs_counter);
1259         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
1260         percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit);
1261         percpu_free_rwsem(&sbi->s_writepages_rwsem);
1262 #ifdef CONFIG_QUOTA
1263         for (i = 0; i < EXT4_MAXQUOTAS; i++)
1264                 kfree(get_qf_name(sb, sbi, i));
1265 #endif
1266
1267         /* Debugging code just in case the in-memory inode orphan list
1268          * isn't empty.  The on-disk one can be non-empty if we've
1269          * detected an error and taken the fs readonly, but the
1270          * in-memory list had better be clean by this point. */
1271         if (!list_empty(&sbi->s_orphan))
1272                 dump_orphan_list(sb, sbi);
1273         ASSERT(list_empty(&sbi->s_orphan));
1274
1275         sync_blockdev(sb->s_bdev);
1276         invalidate_bdev(sb->s_bdev);
1277         if (sbi->s_journal_bdev && sbi->s_journal_bdev != sb->s_bdev) {
1278                 /*
1279                  * Invalidate the journal device's buffers.  We don't want them
1280                  * floating about in memory - the physical journal device may
1281                  * hotswapped, and it breaks the `ro-after' testing code.
1282                  */
1283                 sync_blockdev(sbi->s_journal_bdev);
1284                 invalidate_bdev(sbi->s_journal_bdev);
1285                 ext4_blkdev_remove(sbi);
1286         }
1287
1288         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
1289         sbi->s_ea_inode_cache = NULL;
1290
1291         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
1292         sbi->s_ea_block_cache = NULL;
1293
1294         ext4_stop_mmpd(sbi);
1295
1296         brelse(sbi->s_sbh);
1297         sb->s_fs_info = NULL;
1298         /*
1299          * Now that we are completely done shutting down the
1300          * superblock, we need to actually destroy the kobject.
1301          */
1302         kobject_put(&sbi->s_kobj);
1303         wait_for_completion(&sbi->s_kobj_unregister);
1304         if (sbi->s_chksum_driver)
1305                 crypto_free_shash(sbi->s_chksum_driver);
1306         kfree(sbi->s_blockgroup_lock);
1307         fs_put_dax(sbi->s_daxdev);
1308         fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy);
1309 #if IS_ENABLED(CONFIG_UNICODE)
1310         utf8_unload(sb->s_encoding);
1311 #endif
1312         kfree(sbi);
1313 }
1314
1315 static struct kmem_cache *ext4_inode_cachep;
1316
1317 /*
1318  * Called inside transaction, so use GFP_NOFS
1319  */
1320 static struct inode *ext4_alloc_inode(struct super_block *sb)
1321 {
1322         struct ext4_inode_info *ei;
1323
1324         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
1325         if (!ei)
1326                 return NULL;
1327
1328         inode_set_iversion(&ei->vfs_inode, 1);
1329         spin_lock_init(&ei->i_raw_lock);
1330         INIT_LIST_HEAD(&ei->i_prealloc_list);
1331         atomic_set(&ei->i_prealloc_active, 0);
1332         spin_lock_init(&ei->i_prealloc_lock);
1333         ext4_es_init_tree(&ei->i_es_tree);
1334         rwlock_init(&ei->i_es_lock);
1335         INIT_LIST_HEAD(&ei->i_es_list);
1336         ei->i_es_all_nr = 0;
1337         ei->i_es_shk_nr = 0;
1338         ei->i_es_shrink_lblk = 0;
1339         ei->i_reserved_data_blocks = 0;
1340         spin_lock_init(&(ei->i_block_reservation_lock));
1341         ext4_init_pending_tree(&ei->i_pending_tree);
1342 #ifdef CONFIG_QUOTA
1343         ei->i_reserved_quota = 0;
1344         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
1345 #endif
1346         ei->jinode = NULL;
1347         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
1348         spin_lock_init(&ei->i_completed_io_lock);
1349         ei->i_sync_tid = 0;
1350         ei->i_datasync_tid = 0;
1351         atomic_set(&ei->i_unwritten, 0);
1352         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
1353         ext4_fc_init_inode(&ei->vfs_inode);
1354         mutex_init(&ei->i_fc_lock);
1355         return &ei->vfs_inode;
1356 }
1357
1358 static int ext4_drop_inode(struct inode *inode)
1359 {
1360         int drop = generic_drop_inode(inode);
1361
1362         if (!drop)
1363                 drop = fscrypt_drop_inode(inode);
1364
1365         trace_ext4_drop_inode(inode, drop);
1366         return drop;
1367 }
1368
1369 static void ext4_free_in_core_inode(struct inode *inode)
1370 {
1371         fscrypt_free_inode(inode);
1372         if (!list_empty(&(EXT4_I(inode)->i_fc_list))) {
1373                 pr_warn("%s: inode %ld still in fc list",
1374                         __func__, inode->i_ino);
1375         }
1376         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1377 }
1378
1379 static void ext4_destroy_inode(struct inode *inode)
1380 {
1381         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1382                 ext4_msg(inode->i_sb, KERN_ERR,
1383                          "Inode %lu (%p): orphan list check failed!",
1384                          inode->i_ino, EXT4_I(inode));
1385                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1386                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1387                                 true);
1388                 dump_stack();
1389         }
1390
1391         if (EXT4_I(inode)->i_reserved_data_blocks)
1392                 ext4_msg(inode->i_sb, KERN_ERR,
1393                          "Inode %lu (%p): i_reserved_data_blocks (%u) not cleared!",
1394                          inode->i_ino, EXT4_I(inode),
1395                          EXT4_I(inode)->i_reserved_data_blocks);
1396 }
1397
1398 static void init_once(void *foo)
1399 {
1400         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1401
1402         INIT_LIST_HEAD(&ei->i_orphan);
1403         init_rwsem(&ei->xattr_sem);
1404         init_rwsem(&ei->i_data_sem);
1405         inode_init_once(&ei->vfs_inode);
1406         ext4_fc_init_inode(&ei->vfs_inode);
1407 }
1408
1409 static int __init init_inodecache(void)
1410 {
1411         ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache",
1412                                 sizeof(struct ext4_inode_info), 0,
1413                                 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
1414                                         SLAB_ACCOUNT),
1415                                 offsetof(struct ext4_inode_info, i_data),
1416                                 sizeof_field(struct ext4_inode_info, i_data),
1417                                 init_once);
1418         if (ext4_inode_cachep == NULL)
1419                 return -ENOMEM;
1420         return 0;
1421 }
1422
1423 static void destroy_inodecache(void)
1424 {
1425         /*
1426          * Make sure all delayed rcu free inodes are flushed before we
1427          * destroy cache.
1428          */
1429         rcu_barrier();
1430         kmem_cache_destroy(ext4_inode_cachep);
1431 }
1432
1433 void ext4_clear_inode(struct inode *inode)
1434 {
1435         ext4_fc_del(inode);
1436         invalidate_inode_buffers(inode);
1437         clear_inode(inode);
1438         ext4_discard_preallocations(inode, 0);
1439         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1440         dquot_drop(inode);
1441         if (EXT4_I(inode)->jinode) {
1442                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1443                                                EXT4_I(inode)->jinode);
1444                 jbd2_free_inode(EXT4_I(inode)->jinode);
1445                 EXT4_I(inode)->jinode = NULL;
1446         }
1447         fscrypt_put_encryption_info(inode);
1448         fsverity_cleanup_inode(inode);
1449 }
1450
1451 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1452                                         u64 ino, u32 generation)
1453 {
1454         struct inode *inode;
1455
1456         /*
1457          * Currently we don't know the generation for parent directory, so
1458          * a generation of 0 means "accept any"
1459          */
1460         inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE);
1461         if (IS_ERR(inode))
1462                 return ERR_CAST(inode);
1463         if (generation && inode->i_generation != generation) {
1464                 iput(inode);
1465                 return ERR_PTR(-ESTALE);
1466         }
1467
1468         return inode;
1469 }
1470
1471 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1472                                         int fh_len, int fh_type)
1473 {
1474         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1475                                     ext4_nfs_get_inode);
1476 }
1477
1478 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1479                                         int fh_len, int fh_type)
1480 {
1481         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1482                                     ext4_nfs_get_inode);
1483 }
1484
1485 static int ext4_nfs_commit_metadata(struct inode *inode)
1486 {
1487         struct writeback_control wbc = {
1488                 .sync_mode = WB_SYNC_ALL
1489         };
1490
1491         trace_ext4_nfs_commit_metadata(inode);
1492         return ext4_write_inode(inode, &wbc);
1493 }
1494
1495 #ifdef CONFIG_FS_ENCRYPTION
1496 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1497 {
1498         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1499                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1500 }
1501
1502 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1503                                                         void *fs_data)
1504 {
1505         handle_t *handle = fs_data;
1506         int res, res2, credits, retries = 0;
1507
1508         /*
1509          * Encrypting the root directory is not allowed because e2fsck expects
1510          * lost+found to exist and be unencrypted, and encrypting the root
1511          * directory would imply encrypting the lost+found directory as well as
1512          * the filename "lost+found" itself.
1513          */
1514         if (inode->i_ino == EXT4_ROOT_INO)
1515                 return -EPERM;
1516
1517         if (WARN_ON_ONCE(IS_DAX(inode) && i_size_read(inode)))
1518                 return -EINVAL;
1519
1520         if (ext4_test_inode_flag(inode, EXT4_INODE_DAX))
1521                 return -EOPNOTSUPP;
1522
1523         res = ext4_convert_inline_data(inode);
1524         if (res)
1525                 return res;
1526
1527         /*
1528          * If a journal handle was specified, then the encryption context is
1529          * being set on a new inode via inheritance and is part of a larger
1530          * transaction to create the inode.  Otherwise the encryption context is
1531          * being set on an existing inode in its own transaction.  Only in the
1532          * latter case should the "retry on ENOSPC" logic be used.
1533          */
1534
1535         if (handle) {
1536                 res = ext4_xattr_set_handle(handle, inode,
1537                                             EXT4_XATTR_INDEX_ENCRYPTION,
1538                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1539                                             ctx, len, 0);
1540                 if (!res) {
1541                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1542                         ext4_clear_inode_state(inode,
1543                                         EXT4_STATE_MAY_INLINE_DATA);
1544                         /*
1545                          * Update inode->i_flags - S_ENCRYPTED will be enabled,
1546                          * S_DAX may be disabled
1547                          */
1548                         ext4_set_inode_flags(inode, false);
1549                 }
1550                 return res;
1551         }
1552
1553         res = dquot_initialize(inode);
1554         if (res)
1555                 return res;
1556 retry:
1557         res = ext4_xattr_set_credits(inode, len, false /* is_create */,
1558                                      &credits);
1559         if (res)
1560                 return res;
1561
1562         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
1563         if (IS_ERR(handle))
1564                 return PTR_ERR(handle);
1565
1566         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1567                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1568                                     ctx, len, 0);
1569         if (!res) {
1570                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1571                 /*
1572                  * Update inode->i_flags - S_ENCRYPTED will be enabled,
1573                  * S_DAX may be disabled
1574                  */
1575                 ext4_set_inode_flags(inode, false);
1576                 res = ext4_mark_inode_dirty(handle, inode);
1577                 if (res)
1578                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1579         }
1580         res2 = ext4_journal_stop(handle);
1581
1582         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1583                 goto retry;
1584         if (!res)
1585                 res = res2;
1586         return res;
1587 }
1588
1589 static const union fscrypt_policy *ext4_get_dummy_policy(struct super_block *sb)
1590 {
1591         return EXT4_SB(sb)->s_dummy_enc_policy.policy;
1592 }
1593
1594 static bool ext4_has_stable_inodes(struct super_block *sb)
1595 {
1596         return ext4_has_feature_stable_inodes(sb);
1597 }
1598
1599 static void ext4_get_ino_and_lblk_bits(struct super_block *sb,
1600                                        int *ino_bits_ret, int *lblk_bits_ret)
1601 {
1602         *ino_bits_ret = 8 * sizeof(EXT4_SB(sb)->s_es->s_inodes_count);
1603         *lblk_bits_ret = 8 * sizeof(ext4_lblk_t);
1604 }
1605
1606 static const struct fscrypt_operations ext4_cryptops = {
1607         .key_prefix             = "ext4:",
1608         .get_context            = ext4_get_context,
1609         .set_context            = ext4_set_context,
1610         .get_dummy_policy       = ext4_get_dummy_policy,
1611         .empty_dir              = ext4_empty_dir,
1612         .has_stable_inodes      = ext4_has_stable_inodes,
1613         .get_ino_and_lblk_bits  = ext4_get_ino_and_lblk_bits,
1614 };
1615 #endif
1616
1617 #ifdef CONFIG_QUOTA
1618 static const char * const quotatypes[] = INITQFNAMES;
1619 #define QTYPE2NAME(t) (quotatypes[t])
1620
1621 static int ext4_write_dquot(struct dquot *dquot);
1622 static int ext4_acquire_dquot(struct dquot *dquot);
1623 static int ext4_release_dquot(struct dquot *dquot);
1624 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1625 static int ext4_write_info(struct super_block *sb, int type);
1626 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1627                          const struct path *path);
1628 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1629                                size_t len, loff_t off);
1630 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1631                                 const char *data, size_t len, loff_t off);
1632 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1633                              unsigned int flags);
1634
1635 static struct dquot **ext4_get_dquots(struct inode *inode)
1636 {
1637         return EXT4_I(inode)->i_dquot;
1638 }
1639
1640 static const struct dquot_operations ext4_quota_operations = {
1641         .get_reserved_space     = ext4_get_reserved_space,
1642         .write_dquot            = ext4_write_dquot,
1643         .acquire_dquot          = ext4_acquire_dquot,
1644         .release_dquot          = ext4_release_dquot,
1645         .mark_dirty             = ext4_mark_dquot_dirty,
1646         .write_info             = ext4_write_info,
1647         .alloc_dquot            = dquot_alloc,
1648         .destroy_dquot          = dquot_destroy,
1649         .get_projid             = ext4_get_projid,
1650         .get_inode_usage        = ext4_get_inode_usage,
1651         .get_next_id            = dquot_get_next_id,
1652 };
1653
1654 static const struct quotactl_ops ext4_qctl_operations = {
1655         .quota_on       = ext4_quota_on,
1656         .quota_off      = ext4_quota_off,
1657         .quota_sync     = dquot_quota_sync,
1658         .get_state      = dquot_get_state,
1659         .set_info       = dquot_set_dqinfo,
1660         .get_dqblk      = dquot_get_dqblk,
1661         .set_dqblk      = dquot_set_dqblk,
1662         .get_nextdqblk  = dquot_get_next_dqblk,
1663 };
1664 #endif
1665
1666 static const struct super_operations ext4_sops = {
1667         .alloc_inode    = ext4_alloc_inode,
1668         .free_inode     = ext4_free_in_core_inode,
1669         .destroy_inode  = ext4_destroy_inode,
1670         .write_inode    = ext4_write_inode,
1671         .dirty_inode    = ext4_dirty_inode,
1672         .drop_inode     = ext4_drop_inode,
1673         .evict_inode    = ext4_evict_inode,
1674         .put_super      = ext4_put_super,
1675         .sync_fs        = ext4_sync_fs,
1676         .freeze_fs      = ext4_freeze,
1677         .unfreeze_fs    = ext4_unfreeze,
1678         .statfs         = ext4_statfs,
1679         .show_options   = ext4_show_options,
1680 #ifdef CONFIG_QUOTA
1681         .quota_read     = ext4_quota_read,
1682         .quota_write    = ext4_quota_write,
1683         .get_dquots     = ext4_get_dquots,
1684 #endif
1685 };
1686
1687 static const struct export_operations ext4_export_ops = {
1688         .fh_to_dentry = ext4_fh_to_dentry,
1689         .fh_to_parent = ext4_fh_to_parent,
1690         .get_parent = ext4_get_parent,
1691         .commit_metadata = ext4_nfs_commit_metadata,
1692 };
1693
1694 enum {
1695         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1696         Opt_resgid, Opt_resuid, Opt_sb,
1697         Opt_nouid32, Opt_debug, Opt_removed,
1698         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1699         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1700         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1701         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1702         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1703         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1704         Opt_inlinecrypt,
1705         Opt_usrjquota, Opt_grpjquota, Opt_quota,
1706         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1707         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version,
1708         Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never,
1709         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
1710         Opt_nowarn_on_error, Opt_mblk_io_submit, Opt_debug_want_extra_isize,
1711         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1712         Opt_inode_readahead_blks, Opt_journal_ioprio,
1713         Opt_dioread_nolock, Opt_dioread_lock,
1714         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1715         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1716         Opt_no_prefetch_block_bitmaps, Opt_mb_optimize_scan,
1717         Opt_errors, Opt_data, Opt_data_err, Opt_jqfmt, Opt_dax_type,
1718 #ifdef CONFIG_EXT4_DEBUG
1719         Opt_fc_debug_max_replay, Opt_fc_debug_force
1720 #endif
1721 };
1722
1723 static const struct constant_table ext4_param_errors[] = {
1724         {"continue",    EXT4_MOUNT_ERRORS_CONT},
1725         {"panic",       EXT4_MOUNT_ERRORS_PANIC},
1726         {"remount-ro",  EXT4_MOUNT_ERRORS_RO},
1727         {}
1728 };
1729
1730 static const struct constant_table ext4_param_data[] = {
1731         {"journal",     EXT4_MOUNT_JOURNAL_DATA},
1732         {"ordered",     EXT4_MOUNT_ORDERED_DATA},
1733         {"writeback",   EXT4_MOUNT_WRITEBACK_DATA},
1734         {}
1735 };
1736
1737 static const struct constant_table ext4_param_data_err[] = {
1738         {"abort",       Opt_data_err_abort},
1739         {"ignore",      Opt_data_err_ignore},
1740         {}
1741 };
1742
1743 static const struct constant_table ext4_param_jqfmt[] = {
1744         {"vfsold",      QFMT_VFS_OLD},
1745         {"vfsv0",       QFMT_VFS_V0},
1746         {"vfsv1",       QFMT_VFS_V1},
1747         {}
1748 };
1749
1750 static const struct constant_table ext4_param_dax[] = {
1751         {"always",      Opt_dax_always},
1752         {"inode",       Opt_dax_inode},
1753         {"never",       Opt_dax_never},
1754         {}
1755 };
1756
1757 /* String parameter that allows empty argument */
1758 #define fsparam_string_empty(NAME, OPT) \
1759         __fsparam(fs_param_is_string, NAME, OPT, fs_param_can_be_empty, NULL)
1760
1761 /*
1762  * Mount option specification
1763  * We don't use fsparam_flag_no because of the way we set the
1764  * options and the way we show them in _ext4_show_options(). To
1765  * keep the changes to a minimum, let's keep the negative options
1766  * separate for now.
1767  */
1768 static const struct fs_parameter_spec ext4_param_specs[] = {
1769         fsparam_flag    ("bsddf",               Opt_bsd_df),
1770         fsparam_flag    ("minixdf",             Opt_minix_df),
1771         fsparam_flag    ("grpid",               Opt_grpid),
1772         fsparam_flag    ("bsdgroups",           Opt_grpid),
1773         fsparam_flag    ("nogrpid",             Opt_nogrpid),
1774         fsparam_flag    ("sysvgroups",          Opt_nogrpid),
1775         fsparam_u32     ("resgid",              Opt_resgid),
1776         fsparam_u32     ("resuid",              Opt_resuid),
1777         fsparam_u32     ("sb",                  Opt_sb),
1778         fsparam_enum    ("errors",              Opt_errors, ext4_param_errors),
1779         fsparam_flag    ("nouid32",             Opt_nouid32),
1780         fsparam_flag    ("debug",               Opt_debug),
1781         fsparam_flag    ("oldalloc",            Opt_removed),
1782         fsparam_flag    ("orlov",               Opt_removed),
1783         fsparam_flag    ("user_xattr",          Opt_user_xattr),
1784         fsparam_flag    ("nouser_xattr",        Opt_nouser_xattr),
1785         fsparam_flag    ("acl",                 Opt_acl),
1786         fsparam_flag    ("noacl",               Opt_noacl),
1787         fsparam_flag    ("norecovery",          Opt_noload),
1788         fsparam_flag    ("noload",              Opt_noload),
1789         fsparam_flag    ("bh",                  Opt_removed),
1790         fsparam_flag    ("nobh",                Opt_removed),
1791         fsparam_u32     ("commit",              Opt_commit),
1792         fsparam_u32     ("min_batch_time",      Opt_min_batch_time),
1793         fsparam_u32     ("max_batch_time",      Opt_max_batch_time),
1794         fsparam_u32     ("journal_dev",         Opt_journal_dev),
1795         fsparam_bdev    ("journal_path",        Opt_journal_path),
1796         fsparam_flag    ("journal_checksum",    Opt_journal_checksum),
1797         fsparam_flag    ("nojournal_checksum",  Opt_nojournal_checksum),
1798         fsparam_flag    ("journal_async_commit",Opt_journal_async_commit),
1799         fsparam_flag    ("abort",               Opt_abort),
1800         fsparam_enum    ("data",                Opt_data, ext4_param_data),
1801         fsparam_enum    ("data_err",            Opt_data_err,
1802                                                 ext4_param_data_err),
1803         fsparam_string_empty
1804                         ("usrjquota",           Opt_usrjquota),
1805         fsparam_string_empty
1806                         ("grpjquota",           Opt_grpjquota),
1807         fsparam_enum    ("jqfmt",               Opt_jqfmt, ext4_param_jqfmt),
1808         fsparam_flag    ("grpquota",            Opt_grpquota),
1809         fsparam_flag    ("quota",               Opt_quota),
1810         fsparam_flag    ("noquota",             Opt_noquota),
1811         fsparam_flag    ("usrquota",            Opt_usrquota),
1812         fsparam_flag    ("prjquota",            Opt_prjquota),
1813         fsparam_flag    ("barrier",             Opt_barrier),
1814         fsparam_u32     ("barrier",             Opt_barrier),
1815         fsparam_flag    ("nobarrier",           Opt_nobarrier),
1816         fsparam_flag    ("i_version",           Opt_i_version),
1817         fsparam_flag    ("dax",                 Opt_dax),
1818         fsparam_enum    ("dax",                 Opt_dax_type, ext4_param_dax),
1819         fsparam_u32     ("stripe",              Opt_stripe),
1820         fsparam_flag    ("delalloc",            Opt_delalloc),
1821         fsparam_flag    ("nodelalloc",          Opt_nodelalloc),
1822         fsparam_flag    ("warn_on_error",       Opt_warn_on_error),
1823         fsparam_flag    ("nowarn_on_error",     Opt_nowarn_on_error),
1824         fsparam_u32     ("debug_want_extra_isize",
1825                                                 Opt_debug_want_extra_isize),
1826         fsparam_flag    ("mblk_io_submit",      Opt_removed),
1827         fsparam_flag    ("nomblk_io_submit",    Opt_removed),
1828         fsparam_flag    ("block_validity",      Opt_block_validity),
1829         fsparam_flag    ("noblock_validity",    Opt_noblock_validity),
1830         fsparam_u32     ("inode_readahead_blks",
1831                                                 Opt_inode_readahead_blks),
1832         fsparam_u32     ("journal_ioprio",      Opt_journal_ioprio),
1833         fsparam_u32     ("auto_da_alloc",       Opt_auto_da_alloc),
1834         fsparam_flag    ("auto_da_alloc",       Opt_auto_da_alloc),
1835         fsparam_flag    ("noauto_da_alloc",     Opt_noauto_da_alloc),
1836         fsparam_flag    ("dioread_nolock",      Opt_dioread_nolock),
1837         fsparam_flag    ("nodioread_nolock",    Opt_dioread_lock),
1838         fsparam_flag    ("dioread_lock",        Opt_dioread_lock),
1839         fsparam_flag    ("discard",             Opt_discard),
1840         fsparam_flag    ("nodiscard",           Opt_nodiscard),
1841         fsparam_u32     ("init_itable",         Opt_init_itable),
1842         fsparam_flag    ("init_itable",         Opt_init_itable),
1843         fsparam_flag    ("noinit_itable",       Opt_noinit_itable),
1844 #ifdef CONFIG_EXT4_DEBUG
1845         fsparam_flag    ("fc_debug_force",      Opt_fc_debug_force),
1846         fsparam_u32     ("fc_debug_max_replay", Opt_fc_debug_max_replay),
1847 #endif
1848         fsparam_u32     ("max_dir_size_kb",     Opt_max_dir_size_kb),
1849         fsparam_flag    ("test_dummy_encryption",
1850                                                 Opt_test_dummy_encryption),
1851         fsparam_string  ("test_dummy_encryption",
1852                                                 Opt_test_dummy_encryption),
1853         fsparam_flag    ("inlinecrypt",         Opt_inlinecrypt),
1854         fsparam_flag    ("nombcache",           Opt_nombcache),
1855         fsparam_flag    ("no_mbcache",          Opt_nombcache), /* for backward compatibility */
1856         fsparam_flag    ("prefetch_block_bitmaps",
1857                                                 Opt_removed),
1858         fsparam_flag    ("no_prefetch_block_bitmaps",
1859                                                 Opt_no_prefetch_block_bitmaps),
1860         fsparam_s32     ("mb_optimize_scan",    Opt_mb_optimize_scan),
1861         fsparam_string  ("check",               Opt_removed),   /* mount option from ext2/3 */
1862         fsparam_flag    ("nocheck",             Opt_removed),   /* mount option from ext2/3 */
1863         fsparam_flag    ("reservation",         Opt_removed),   /* mount option from ext2/3 */
1864         fsparam_flag    ("noreservation",       Opt_removed),   /* mount option from ext2/3 */
1865         fsparam_u32     ("journal",             Opt_removed),   /* mount option from ext2/3 */
1866         {}
1867 };
1868
1869 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1870 #define DEFAULT_MB_OPTIMIZE_SCAN        (-1)
1871
1872 static const char deprecated_msg[] =
1873         "Mount option \"%s\" will be removed by %s\n"
1874         "Contact [email protected] if you think we should keep it.\n";
1875
1876 #define MOPT_SET        0x0001
1877 #define MOPT_CLEAR      0x0002
1878 #define MOPT_NOSUPPORT  0x0004
1879 #define MOPT_EXPLICIT   0x0008
1880 #ifdef CONFIG_QUOTA
1881 #define MOPT_Q          0
1882 #define MOPT_QFMT       0x0010
1883 #else
1884 #define MOPT_Q          MOPT_NOSUPPORT
1885 #define MOPT_QFMT       MOPT_NOSUPPORT
1886 #endif
1887 #define MOPT_NO_EXT2    0x0020
1888 #define MOPT_NO_EXT3    0x0040
1889 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1890 #define MOPT_SKIP       0x0080
1891 #define MOPT_2          0x0100
1892
1893 static const struct mount_opts {
1894         int     token;
1895         int     mount_opt;
1896         int     flags;
1897 } ext4_mount_opts[] = {
1898         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1899         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1900         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1901         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1902         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1903         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1904         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1905          MOPT_EXT4_ONLY | MOPT_SET},
1906         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1907          MOPT_EXT4_ONLY | MOPT_CLEAR},
1908         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1909         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1910         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1911          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1912         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1913          MOPT_EXT4_ONLY | MOPT_CLEAR},
1914         {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET},
1915         {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR},
1916         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1917          MOPT_EXT4_ONLY | MOPT_CLEAR},
1918         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1919          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1920         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1921                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1922          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1923         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1924         {Opt_data_err, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_NO_EXT2},
1925         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1926         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1927         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1928         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1929         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1930         {Opt_dax_type, 0, MOPT_EXT4_ONLY},
1931         {Opt_journal_dev, 0, MOPT_NO_EXT2},
1932         {Opt_journal_path, 0, MOPT_NO_EXT2},
1933         {Opt_journal_ioprio, 0, MOPT_NO_EXT2},
1934         {Opt_data, 0, MOPT_NO_EXT2},
1935         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1936         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1937 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1938         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1939         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1940 #else
1941         {Opt_acl, 0, MOPT_NOSUPPORT},
1942         {Opt_noacl, 0, MOPT_NOSUPPORT},
1943 #endif
1944         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1945         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1946         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1947         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1948                                                         MOPT_SET | MOPT_Q},
1949         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1950                                                         MOPT_SET | MOPT_Q},
1951         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1952                                                         MOPT_SET | MOPT_Q},
1953         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1954                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1955                                                         MOPT_CLEAR | MOPT_Q},
1956         {Opt_usrjquota, 0, MOPT_Q},
1957         {Opt_grpjquota, 0, MOPT_Q},
1958         {Opt_jqfmt, 0, MOPT_QFMT},
1959         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1960         {Opt_no_prefetch_block_bitmaps, EXT4_MOUNT_NO_PREFETCH_BLOCK_BITMAPS,
1961          MOPT_SET},
1962 #ifdef CONFIG_EXT4_DEBUG
1963         {Opt_fc_debug_force, EXT4_MOUNT2_JOURNAL_FAST_COMMIT,
1964          MOPT_SET | MOPT_2 | MOPT_EXT4_ONLY},
1965 #endif
1966         {Opt_err, 0, 0}
1967 };
1968
1969 #if IS_ENABLED(CONFIG_UNICODE)
1970 static const struct ext4_sb_encodings {
1971         __u16 magic;
1972         char *name;
1973         unsigned int version;
1974 } ext4_sb_encoding_map[] = {
1975         {EXT4_ENC_UTF8_12_1, "utf8", UNICODE_AGE(12, 1, 0)},
1976 };
1977
1978 static const struct ext4_sb_encodings *
1979 ext4_sb_read_encoding(const struct ext4_super_block *es)
1980 {
1981         __u16 magic = le16_to_cpu(es->s_encoding);
1982         int i;
1983
1984         for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++)
1985                 if (magic == ext4_sb_encoding_map[i].magic)
1986                         return &ext4_sb_encoding_map[i];
1987
1988         return NULL;
1989 }
1990 #endif
1991
1992 static int ext4_set_test_dummy_encryption(struct super_block *sb, char *arg)
1993 {
1994 #ifdef CONFIG_FS_ENCRYPTION
1995         struct ext4_sb_info *sbi = EXT4_SB(sb);
1996         int err;
1997
1998         err = fscrypt_set_test_dummy_encryption(sb, arg,
1999                                                 &sbi->s_dummy_enc_policy);
2000         if (err) {
2001                 ext4_msg(sb, KERN_WARNING,
2002                          "Error while setting test dummy encryption [%d]", err);
2003                 return err;
2004         }
2005         ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled");
2006 #endif
2007         return 0;
2008 }
2009
2010 #define EXT4_SPEC_JQUOTA                        (1 <<  0)
2011 #define EXT4_SPEC_JQFMT                         (1 <<  1)
2012 #define EXT4_SPEC_DATAJ                         (1 <<  2)
2013 #define EXT4_SPEC_SB_BLOCK                      (1 <<  3)
2014 #define EXT4_SPEC_JOURNAL_DEV                   (1 <<  4)
2015 #define EXT4_SPEC_JOURNAL_IOPRIO                (1 <<  5)
2016 #define EXT4_SPEC_DUMMY_ENCRYPTION              (1 <<  6)
2017 #define EXT4_SPEC_s_want_extra_isize            (1 <<  7)
2018 #define EXT4_SPEC_s_max_batch_time              (1 <<  8)
2019 #define EXT4_SPEC_s_min_batch_time              (1 <<  9)
2020 #define EXT4_SPEC_s_inode_readahead_blks        (1 << 10)
2021 #define EXT4_SPEC_s_li_wait_mult                (1 << 11)
2022 #define EXT4_SPEC_s_max_dir_size_kb             (1 << 12)
2023 #define EXT4_SPEC_s_stripe                      (1 << 13)
2024 #define EXT4_SPEC_s_resuid                      (1 << 14)
2025 #define EXT4_SPEC_s_resgid                      (1 << 15)
2026 #define EXT4_SPEC_s_commit_interval             (1 << 16)
2027 #define EXT4_SPEC_s_fc_debug_max_replay         (1 << 17)
2028 #define EXT4_SPEC_s_sb_block                    (1 << 18)
2029 #define EXT4_SPEC_mb_optimize_scan              (1 << 19)
2030
2031 struct ext4_fs_context {
2032         char            *s_qf_names[EXT4_MAXQUOTAS];
2033         char            *test_dummy_enc_arg;
2034         int             s_jquota_fmt;   /* Format of quota to use */
2035 #ifdef CONFIG_EXT4_DEBUG
2036         int s_fc_debug_max_replay;
2037 #endif
2038         unsigned short  qname_spec;
2039         unsigned long   vals_s_flags;   /* Bits to set in s_flags */
2040         unsigned long   mask_s_flags;   /* Bits changed in s_flags */
2041         unsigned long   journal_devnum;
2042         unsigned long   s_commit_interval;
2043         unsigned long   s_stripe;
2044         unsigned int    s_inode_readahead_blks;
2045         unsigned int    s_want_extra_isize;
2046         unsigned int    s_li_wait_mult;
2047         unsigned int    s_max_dir_size_kb;
2048         unsigned int    journal_ioprio;
2049         unsigned int    vals_s_mount_opt;
2050         unsigned int    mask_s_mount_opt;
2051         unsigned int    vals_s_mount_opt2;
2052         unsigned int    mask_s_mount_opt2;
2053         unsigned long   vals_s_mount_flags;
2054         unsigned long   mask_s_mount_flags;
2055         unsigned int    opt_flags;      /* MOPT flags */
2056         unsigned int    spec;
2057         u32             s_max_batch_time;
2058         u32             s_min_batch_time;
2059         kuid_t          s_resuid;
2060         kgid_t          s_resgid;
2061         ext4_fsblk_t    s_sb_block;
2062 };
2063
2064 static void ext4_fc_free(struct fs_context *fc)
2065 {
2066         struct ext4_fs_context *ctx = fc->fs_private;
2067         int i;
2068
2069         if (!ctx)
2070                 return;
2071
2072         for (i = 0; i < EXT4_MAXQUOTAS; i++)
2073                 kfree(ctx->s_qf_names[i]);
2074
2075         kfree(ctx->test_dummy_enc_arg);
2076         kfree(ctx);
2077 }
2078
2079 int ext4_init_fs_context(struct fs_context *fc)
2080 {
2081         struct ext4_fs_context *ctx;
2082
2083         ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL);
2084         if (!ctx)
2085                 return -ENOMEM;
2086
2087         fc->fs_private = ctx;
2088         fc->ops = &ext4_context_ops;
2089
2090         return 0;
2091 }
2092
2093 #ifdef CONFIG_QUOTA
2094 /*
2095  * Note the name of the specified quota file.
2096  */
2097 static int note_qf_name(struct fs_context *fc, int qtype,
2098                        struct fs_parameter *param)
2099 {
2100         struct ext4_fs_context *ctx = fc->fs_private;
2101         char *qname;
2102
2103         if (param->size < 1) {
2104                 ext4_msg(NULL, KERN_ERR, "Missing quota name");
2105                 return -EINVAL;
2106         }
2107         if (strchr(param->string, '/')) {
2108                 ext4_msg(NULL, KERN_ERR,
2109                          "quotafile must be on filesystem root");
2110                 return -EINVAL;
2111         }
2112         if (ctx->s_qf_names[qtype]) {
2113                 if (strcmp(ctx->s_qf_names[qtype], param->string) != 0) {
2114                         ext4_msg(NULL, KERN_ERR,
2115                                  "%s quota file already specified",
2116                                  QTYPE2NAME(qtype));
2117                         return -EINVAL;
2118                 }
2119                 return 0;
2120         }
2121
2122         qname = kmemdup_nul(param->string, param->size, GFP_KERNEL);
2123         if (!qname) {
2124                 ext4_msg(NULL, KERN_ERR,
2125                          "Not enough memory for storing quotafile name");
2126                 return -ENOMEM;
2127         }
2128         ctx->s_qf_names[qtype] = qname;
2129         ctx->qname_spec |= 1 << qtype;
2130         ctx->spec |= EXT4_SPEC_JQUOTA;
2131         return 0;
2132 }
2133
2134 /*
2135  * Clear the name of the specified quota file.
2136  */
2137 static int unnote_qf_name(struct fs_context *fc, int qtype)
2138 {
2139         struct ext4_fs_context *ctx = fc->fs_private;
2140
2141         if (ctx->s_qf_names[qtype])
2142                 kfree(ctx->s_qf_names[qtype]);
2143
2144         ctx->s_qf_names[qtype] = NULL;
2145         ctx->qname_spec |= 1 << qtype;
2146         ctx->spec |= EXT4_SPEC_JQUOTA;
2147         return 0;
2148 }
2149 #endif
2150
2151 #define EXT4_SET_CTX(name)                                              \
2152 static inline void ctx_set_##name(struct ext4_fs_context *ctx,          \
2153                                   unsigned long flag)                   \
2154 {                                                                       \
2155         ctx->mask_s_##name |= flag;                                     \
2156         ctx->vals_s_##name |= flag;                                     \
2157 }
2158
2159 #define EXT4_CLEAR_CTX(name)                                            \
2160 static inline void ctx_clear_##name(struct ext4_fs_context *ctx,        \
2161                                     unsigned long flag)                 \
2162 {                                                                       \
2163         ctx->mask_s_##name |= flag;                                     \
2164         ctx->vals_s_##name &= ~flag;                                    \
2165 }
2166
2167 #define EXT4_TEST_CTX(name)                                             \
2168 static inline unsigned long                                             \
2169 ctx_test_##name(struct ext4_fs_context *ctx, unsigned long flag)        \
2170 {                                                                       \
2171         return (ctx->vals_s_##name & flag);                             \
2172 }
2173
2174 EXT4_SET_CTX(flags); /* set only */
2175 EXT4_SET_CTX(mount_opt);
2176 EXT4_CLEAR_CTX(mount_opt);
2177 EXT4_TEST_CTX(mount_opt);
2178 EXT4_SET_CTX(mount_opt2);
2179 EXT4_CLEAR_CTX(mount_opt2);
2180 EXT4_TEST_CTX(mount_opt2);
2181
2182 static inline void ctx_set_mount_flag(struct ext4_fs_context *ctx, int bit)
2183 {
2184         set_bit(bit, &ctx->mask_s_mount_flags);
2185         set_bit(bit, &ctx->vals_s_mount_flags);
2186 }
2187
2188 static int ext4_parse_param(struct fs_context *fc, struct fs_parameter *param)
2189 {
2190         struct ext4_fs_context *ctx = fc->fs_private;
2191         struct fs_parse_result result;
2192         const struct mount_opts *m;
2193         int is_remount;
2194         kuid_t uid;
2195         kgid_t gid;
2196         int token;
2197
2198         token = fs_parse(fc, ext4_param_specs, param, &result);
2199         if (token < 0)
2200                 return token;
2201         is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE;
2202
2203         for (m = ext4_mount_opts; m->token != Opt_err; m++)
2204                 if (token == m->token)
2205                         break;
2206
2207         ctx->opt_flags |= m->flags;
2208
2209         if (m->flags & MOPT_EXPLICIT) {
2210                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
2211                         ctx_set_mount_opt2(ctx, EXT4_MOUNT2_EXPLICIT_DELALLOC);
2212                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
2213                         ctx_set_mount_opt2(ctx,
2214                                        EXT4_MOUNT2_EXPLICIT_JOURNAL_CHECKSUM);
2215                 } else
2216                         return -EINVAL;
2217         }
2218
2219         if (m->flags & MOPT_NOSUPPORT) {
2220                 ext4_msg(NULL, KERN_ERR, "%s option not supported",
2221                          param->key);
2222                 return 0;
2223         }
2224
2225         switch (token) {
2226 #ifdef CONFIG_QUOTA
2227         case Opt_usrjquota:
2228                 if (!*param->string)
2229                         return unnote_qf_name(fc, USRQUOTA);
2230                 else
2231                         return note_qf_name(fc, USRQUOTA, param);
2232         case Opt_grpjquota:
2233                 if (!*param->string)
2234                         return unnote_qf_name(fc, GRPQUOTA);
2235                 else
2236                         return note_qf_name(fc, GRPQUOTA, param);
2237 #endif
2238         case Opt_noacl:
2239         case Opt_nouser_xattr:
2240                 ext4_msg(NULL, KERN_WARNING, deprecated_msg, param->key, "3.5");
2241                 break;
2242         case Opt_sb:
2243                 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
2244                         ext4_msg(NULL, KERN_WARNING,
2245                                  "Ignoring %s option on remount", param->key);
2246                 } else {
2247                         ctx->s_sb_block = result.uint_32;
2248                         ctx->spec |= EXT4_SPEC_s_sb_block;
2249                 }
2250                 return 0;
2251         case Opt_removed:
2252                 ext4_msg(NULL, KERN_WARNING, "Ignoring removed %s option",
2253                          param->key);
2254                 return 0;
2255         case Opt_abort:
2256                 ctx_set_mount_flag(ctx, EXT4_MF_FS_ABORTED);
2257                 return 0;
2258         case Opt_i_version:
2259                 ext4_msg(NULL, KERN_WARNING, deprecated_msg, param->key, "5.20");
2260                 ext4_msg(NULL, KERN_WARNING, "Use iversion instead\n");
2261                 ctx_set_flags(ctx, SB_I_VERSION);
2262                 return 0;
2263         case Opt_inlinecrypt:
2264 #ifdef CONFIG_FS_ENCRYPTION_INLINE_CRYPT
2265                 ctx_set_flags(ctx, SB_INLINECRYPT);
2266 #else
2267                 ext4_msg(NULL, KERN_ERR, "inline encryption not supported");
2268 #endif
2269                 return 0;
2270         case Opt_errors:
2271                 ctx_clear_mount_opt(ctx, EXT4_MOUNT_ERRORS_MASK);
2272                 ctx_set_mount_opt(ctx, result.uint_32);
2273                 return 0;
2274 #ifdef CONFIG_QUOTA
2275         case Opt_jqfmt:
2276                 ctx->s_jquota_fmt = result.uint_32;
2277                 ctx->spec |= EXT4_SPEC_JQFMT;
2278                 return 0;
2279 #endif
2280         case Opt_data:
2281                 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS);
2282                 ctx_set_mount_opt(ctx, result.uint_32);
2283                 ctx->spec |= EXT4_SPEC_DATAJ;
2284                 return 0;
2285         case Opt_commit:
2286                 if (result.uint_32 == 0)
2287                         ctx->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE;
2288                 else if (result.uint_32 > INT_MAX / HZ) {
2289                         ext4_msg(NULL, KERN_ERR,
2290                                  "Invalid commit interval %d, "
2291                                  "must be smaller than %d",
2292                                  result.uint_32, INT_MAX / HZ);
2293                         return -EINVAL;
2294                 }
2295                 ctx->s_commit_interval = HZ * result.uint_32;
2296                 ctx->spec |= EXT4_SPEC_s_commit_interval;
2297                 return 0;
2298         case Opt_debug_want_extra_isize:
2299                 if ((result.uint_32 & 1) || (result.uint_32 < 4)) {
2300                         ext4_msg(NULL, KERN_ERR,
2301                                  "Invalid want_extra_isize %d", result.uint_32);
2302                         return -EINVAL;
2303                 }
2304                 ctx->s_want_extra_isize = result.uint_32;
2305                 ctx->spec |= EXT4_SPEC_s_want_extra_isize;
2306                 return 0;
2307         case Opt_max_batch_time:
2308                 ctx->s_max_batch_time = result.uint_32;
2309                 ctx->spec |= EXT4_SPEC_s_max_batch_time;
2310                 return 0;
2311         case Opt_min_batch_time:
2312                 ctx->s_min_batch_time = result.uint_32;
2313                 ctx->spec |= EXT4_SPEC_s_min_batch_time;
2314                 return 0;
2315         case Opt_inode_readahead_blks:
2316                 if (result.uint_32 &&
2317                     (result.uint_32 > (1 << 30) ||
2318                      !is_power_of_2(result.uint_32))) {
2319                         ext4_msg(NULL, KERN_ERR,
2320                                  "EXT4-fs: inode_readahead_blks must be "
2321                                  "0 or a power of 2 smaller than 2^31");
2322                         return -EINVAL;
2323                 }
2324                 ctx->s_inode_readahead_blks = result.uint_32;
2325                 ctx->spec |= EXT4_SPEC_s_inode_readahead_blks;
2326                 return 0;
2327         case Opt_init_itable:
2328                 ctx_set_mount_opt(ctx, EXT4_MOUNT_INIT_INODE_TABLE);
2329                 ctx->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
2330                 if (param->type == fs_value_is_string)
2331                         ctx->s_li_wait_mult = result.uint_32;
2332                 ctx->spec |= EXT4_SPEC_s_li_wait_mult;
2333                 return 0;
2334         case Opt_max_dir_size_kb:
2335                 ctx->s_max_dir_size_kb = result.uint_32;
2336                 ctx->spec |= EXT4_SPEC_s_max_dir_size_kb;
2337                 return 0;
2338 #ifdef CONFIG_EXT4_DEBUG
2339         case Opt_fc_debug_max_replay:
2340                 ctx->s_fc_debug_max_replay = result.uint_32;
2341                 ctx->spec |= EXT4_SPEC_s_fc_debug_max_replay;
2342                 return 0;
2343 #endif
2344         case Opt_stripe:
2345                 ctx->s_stripe = result.uint_32;
2346                 ctx->spec |= EXT4_SPEC_s_stripe;
2347                 return 0;
2348         case Opt_resuid:
2349                 uid = make_kuid(current_user_ns(), result.uint_32);
2350                 if (!uid_valid(uid)) {
2351                         ext4_msg(NULL, KERN_ERR, "Invalid uid value %d",
2352                                  result.uint_32);
2353                         return -EINVAL;
2354                 }
2355                 ctx->s_resuid = uid;
2356                 ctx->spec |= EXT4_SPEC_s_resuid;
2357                 return 0;
2358         case Opt_resgid:
2359                 gid = make_kgid(current_user_ns(), result.uint_32);
2360                 if (!gid_valid(gid)) {
2361                         ext4_msg(NULL, KERN_ERR, "Invalid gid value %d",
2362                                  result.uint_32);
2363                         return -EINVAL;
2364                 }
2365                 ctx->s_resgid = gid;
2366                 ctx->spec |= EXT4_SPEC_s_resgid;
2367                 return 0;
2368         case Opt_journal_dev:
2369                 if (is_remount) {
2370                         ext4_msg(NULL, KERN_ERR,
2371                                  "Cannot specify journal on remount");
2372                         return -EINVAL;
2373                 }
2374                 ctx->journal_devnum = result.uint_32;
2375                 ctx->spec |= EXT4_SPEC_JOURNAL_DEV;
2376                 return 0;
2377         case Opt_journal_path:
2378         {
2379                 struct inode *journal_inode;
2380                 struct path path;
2381                 int error;
2382
2383                 if (is_remount) {
2384                         ext4_msg(NULL, KERN_ERR,
2385                                  "Cannot specify journal on remount");
2386                         return -EINVAL;
2387                 }
2388
2389                 error = fs_lookup_param(fc, param, 1, &path);
2390                 if (error) {
2391                         ext4_msg(NULL, KERN_ERR, "error: could not find "
2392                                  "journal device path");
2393                         return -EINVAL;
2394                 }
2395
2396                 journal_inode = d_inode(path.dentry);
2397                 ctx->journal_devnum = new_encode_dev(journal_inode->i_rdev);
2398                 ctx->spec |= EXT4_SPEC_JOURNAL_DEV;
2399                 path_put(&path);
2400                 return 0;
2401         }
2402         case Opt_journal_ioprio:
2403                 if (result.uint_32 > 7) {
2404                         ext4_msg(NULL, KERN_ERR, "Invalid journal IO priority"
2405                                  " (must be 0-7)");
2406                         return -EINVAL;
2407                 }
2408                 ctx->journal_ioprio =
2409                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, result.uint_32);
2410                 ctx->spec |= EXT4_SPEC_JOURNAL_IOPRIO;
2411                 return 0;
2412         case Opt_test_dummy_encryption:
2413 #ifdef CONFIG_FS_ENCRYPTION
2414                 if (param->type == fs_value_is_flag) {
2415                         ctx->spec |= EXT4_SPEC_DUMMY_ENCRYPTION;
2416                         ctx->test_dummy_enc_arg = NULL;
2417                         return 0;
2418                 }
2419                 if (*param->string &&
2420                     !(!strcmp(param->string, "v1") ||
2421                       !strcmp(param->string, "v2"))) {
2422                         ext4_msg(NULL, KERN_WARNING,
2423                                  "Value of option \"%s\" is unrecognized",
2424                                  param->key);
2425                         return -EINVAL;
2426                 }
2427                 ctx->spec |= EXT4_SPEC_DUMMY_ENCRYPTION;
2428                 ctx->test_dummy_enc_arg = kmemdup_nul(param->string, param->size,
2429                                                       GFP_KERNEL);
2430 #else
2431                 ext4_msg(NULL, KERN_WARNING,
2432                          "Test dummy encryption mount option ignored");
2433 #endif
2434                 return 0;
2435         case Opt_dax:
2436         case Opt_dax_type:
2437 #ifdef CONFIG_FS_DAX
2438         {
2439                 int type = (token == Opt_dax) ?
2440                            Opt_dax : result.uint_32;
2441
2442                 switch (type) {
2443                 case Opt_dax:
2444                 case Opt_dax_always:
2445                         ctx_set_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS);
2446                         ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER);
2447                         break;
2448                 case Opt_dax_never:
2449                         ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER);
2450                         ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS);
2451                         break;
2452                 case Opt_dax_inode:
2453                         ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS);
2454                         ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER);
2455                         /* Strictly for printing options */
2456                         ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE);
2457                         break;
2458                 }
2459                 return 0;
2460         }
2461 #else
2462                 ext4_msg(NULL, KERN_INFO, "dax option not supported");
2463                 return -EINVAL;
2464 #endif
2465         case Opt_data_err:
2466                 if (result.uint_32 == Opt_data_err_abort)
2467                         ctx_set_mount_opt(ctx, m->mount_opt);
2468                 else if (result.uint_32 == Opt_data_err_ignore)
2469                         ctx_clear_mount_opt(ctx, m->mount_opt);
2470                 return 0;
2471         case Opt_mb_optimize_scan:
2472                 if (result.int_32 == 1) {
2473                         ctx_set_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN);
2474                         ctx->spec |= EXT4_SPEC_mb_optimize_scan;
2475                 } else if (result.int_32 == 0) {
2476                         ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN);
2477                         ctx->spec |= EXT4_SPEC_mb_optimize_scan;
2478                 } else {
2479                         ext4_msg(NULL, KERN_WARNING,
2480                                  "mb_optimize_scan should be set to 0 or 1.");
2481                         return -EINVAL;
2482                 }
2483                 return 0;
2484         }
2485
2486         /*
2487          * At this point we should only be getting options requiring MOPT_SET,
2488          * or MOPT_CLEAR. Anything else is a bug
2489          */
2490         if (m->token == Opt_err) {
2491                 ext4_msg(NULL, KERN_WARNING, "buggy handling of option %s",
2492                          param->key);
2493                 WARN_ON(1);
2494                 return -EINVAL;
2495         }
2496
2497         else {
2498                 unsigned int set = 0;
2499
2500                 if ((param->type == fs_value_is_flag) ||
2501                     result.uint_32 > 0)
2502                         set = 1;
2503
2504                 if (m->flags & MOPT_CLEAR)
2505                         set = !set;
2506                 else if (unlikely(!(m->flags & MOPT_SET))) {
2507                         ext4_msg(NULL, KERN_WARNING,
2508                                  "buggy handling of option %s",
2509                                  param->key);
2510                         WARN_ON(1);
2511                         return -EINVAL;
2512                 }
2513                 if (m->flags & MOPT_2) {
2514                         if (set != 0)
2515                                 ctx_set_mount_opt2(ctx, m->mount_opt);
2516                         else
2517                                 ctx_clear_mount_opt2(ctx, m->mount_opt);
2518                 } else {
2519                         if (set != 0)
2520                                 ctx_set_mount_opt(ctx, m->mount_opt);
2521                         else
2522                                 ctx_clear_mount_opt(ctx, m->mount_opt);
2523                 }
2524         }
2525
2526         return 0;
2527 }
2528
2529 static int parse_options(struct fs_context *fc, char *options)
2530 {
2531         struct fs_parameter param;
2532         int ret;
2533         char *key;
2534
2535         if (!options)
2536                 return 0;
2537
2538         while ((key = strsep(&options, ",")) != NULL) {
2539                 if (*key) {
2540                         size_t v_len = 0;
2541                         char *value = strchr(key, '=');
2542
2543                         param.type = fs_value_is_flag;
2544                         param.string = NULL;
2545
2546                         if (value) {
2547                                 if (value == key)
2548                                         continue;
2549
2550                                 *value++ = 0;
2551                                 v_len = strlen(value);
2552                                 param.string = kmemdup_nul(value, v_len,
2553                                                            GFP_KERNEL);
2554                                 if (!param.string)
2555                                         return -ENOMEM;
2556                                 param.type = fs_value_is_string;
2557                         }
2558
2559                         param.key = key;
2560                         param.size = v_len;
2561
2562                         ret = ext4_parse_param(fc, &param);
2563                         if (param.string)
2564                                 kfree(param.string);
2565                         if (ret < 0)
2566                                 return ret;
2567                 }
2568         }
2569
2570         ret = ext4_validate_options(fc);
2571         if (ret < 0)
2572                 return ret;
2573
2574         return 0;
2575 }
2576
2577 static int parse_apply_sb_mount_options(struct super_block *sb,
2578                                         struct ext4_fs_context *m_ctx)
2579 {
2580         struct ext4_sb_info *sbi = EXT4_SB(sb);
2581         char *s_mount_opts = NULL;
2582         struct ext4_fs_context *s_ctx = NULL;
2583         struct fs_context *fc = NULL;
2584         int ret = -ENOMEM;
2585
2586         if (!sbi->s_es->s_mount_opts[0])
2587                 return 0;
2588
2589         s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
2590                                 sizeof(sbi->s_es->s_mount_opts),
2591                                 GFP_KERNEL);
2592         if (!s_mount_opts)
2593                 return ret;
2594
2595         fc = kzalloc(sizeof(struct fs_context), GFP_KERNEL);
2596         if (!fc)
2597                 goto out_free;
2598
2599         s_ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL);
2600         if (!s_ctx)
2601                 goto out_free;
2602
2603         fc->fs_private = s_ctx;
2604         fc->s_fs_info = sbi;
2605
2606         ret = parse_options(fc, s_mount_opts);
2607         if (ret < 0)
2608                 goto parse_failed;
2609
2610         ret = ext4_check_opt_consistency(fc, sb);
2611         if (ret < 0) {
2612 parse_failed:
2613                 ext4_msg(sb, KERN_WARNING,
2614                          "failed to parse options in superblock: %s",
2615                          s_mount_opts);
2616                 ret = 0;
2617                 goto out_free;
2618         }
2619
2620         if (s_ctx->spec & EXT4_SPEC_JOURNAL_DEV)
2621                 m_ctx->journal_devnum = s_ctx->journal_devnum;
2622         if (s_ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)
2623                 m_ctx->journal_ioprio = s_ctx->journal_ioprio;
2624
2625         ret = ext4_apply_options(fc, sb);
2626
2627 out_free:
2628         kfree(s_ctx);
2629         kfree(fc);
2630         kfree(s_mount_opts);
2631         return ret;
2632 }
2633
2634 static void ext4_apply_quota_options(struct fs_context *fc,
2635                                      struct super_block *sb)
2636 {
2637 #ifdef CONFIG_QUOTA
2638         bool quota_feature = ext4_has_feature_quota(sb);
2639         struct ext4_fs_context *ctx = fc->fs_private;
2640         struct ext4_sb_info *sbi = EXT4_SB(sb);
2641         char *qname;
2642         int i;
2643
2644         if (quota_feature)
2645                 return;
2646
2647         if (ctx->spec & EXT4_SPEC_JQUOTA) {
2648                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2649                         if (!(ctx->qname_spec & (1 << i)))
2650                                 continue;
2651
2652                         qname = ctx->s_qf_names[i]; /* May be NULL */
2653                         if (qname)
2654                                 set_opt(sb, QUOTA);
2655                         ctx->s_qf_names[i] = NULL;
2656                         qname = rcu_replace_pointer(sbi->s_qf_names[i], qname,
2657                                                 lockdep_is_held(&sb->s_umount));
2658                         if (qname)
2659                                 kfree_rcu(qname);
2660                 }
2661         }
2662
2663         if (ctx->spec & EXT4_SPEC_JQFMT)
2664                 sbi->s_jquota_fmt = ctx->s_jquota_fmt;
2665 #endif
2666 }
2667
2668 /*
2669  * Check quota settings consistency.
2670  */
2671 static int ext4_check_quota_consistency(struct fs_context *fc,
2672                                         struct super_block *sb)
2673 {
2674 #ifdef CONFIG_QUOTA
2675         struct ext4_fs_context *ctx = fc->fs_private;
2676         struct ext4_sb_info *sbi = EXT4_SB(sb);
2677         bool quota_feature = ext4_has_feature_quota(sb);
2678         bool quota_loaded = sb_any_quota_loaded(sb);
2679         bool usr_qf_name, grp_qf_name, usrquota, grpquota;
2680         int quota_flags, i;
2681
2682         /*
2683          * We do the test below only for project quotas. 'usrquota' and
2684          * 'grpquota' mount options are allowed even without quota feature
2685          * to support legacy quotas in quota files.
2686          */
2687         if (ctx_test_mount_opt(ctx, EXT4_MOUNT_PRJQUOTA) &&
2688             !ext4_has_feature_project(sb)) {
2689                 ext4_msg(NULL, KERN_ERR, "Project quota feature not enabled. "
2690                          "Cannot enable project quota enforcement.");
2691                 return -EINVAL;
2692         }
2693
2694         quota_flags = EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
2695                       EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA;
2696         if (quota_loaded &&
2697             ctx->mask_s_mount_opt & quota_flags &&
2698             !ctx_test_mount_opt(ctx, quota_flags))
2699                 goto err_quota_change;
2700
2701         if (ctx->spec & EXT4_SPEC_JQUOTA) {
2702
2703                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2704                         if (!(ctx->qname_spec & (1 << i)))
2705                                 continue;
2706
2707                         if (quota_loaded &&
2708                             !!sbi->s_qf_names[i] != !!ctx->s_qf_names[i])
2709                                 goto err_jquota_change;
2710
2711                         if (sbi->s_qf_names[i] && ctx->s_qf_names[i] &&
2712                             strcmp(get_qf_name(sb, sbi, i),
2713                                    ctx->s_qf_names[i]) != 0)
2714                                 goto err_jquota_specified;
2715                 }
2716
2717                 if (quota_feature) {
2718                         ext4_msg(NULL, KERN_INFO,
2719                                  "Journaled quota options ignored when "
2720                                  "QUOTA feature is enabled");
2721                         return 0;
2722                 }
2723         }
2724
2725         if (ctx->spec & EXT4_SPEC_JQFMT) {
2726                 if (sbi->s_jquota_fmt != ctx->s_jquota_fmt && quota_loaded)
2727                         goto err_jquota_change;
2728                 if (quota_feature) {
2729                         ext4_msg(NULL, KERN_INFO, "Quota format mount options "
2730                                  "ignored when QUOTA feature is enabled");
2731                         return 0;
2732                 }
2733         }
2734
2735         /* Make sure we don't mix old and new quota format */
2736         usr_qf_name = (get_qf_name(sb, sbi, USRQUOTA) ||
2737                        ctx->s_qf_names[USRQUOTA]);
2738         grp_qf_name = (get_qf_name(sb, sbi, GRPQUOTA) ||
2739                        ctx->s_qf_names[GRPQUOTA]);
2740
2741         usrquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) ||
2742                     test_opt(sb, USRQUOTA));
2743
2744         grpquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) ||
2745                     test_opt(sb, GRPQUOTA));
2746
2747         if (usr_qf_name) {
2748                 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA);
2749                 usrquota = false;
2750         }
2751         if (grp_qf_name) {
2752                 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA);
2753                 grpquota = false;
2754         }
2755
2756         if (usr_qf_name || grp_qf_name) {
2757                 if (usrquota || grpquota) {
2758                         ext4_msg(NULL, KERN_ERR, "old and new quota "
2759                                  "format mixing");
2760                         return -EINVAL;
2761                 }
2762
2763                 if (!(ctx->spec & EXT4_SPEC_JQFMT || sbi->s_jquota_fmt)) {
2764                         ext4_msg(NULL, KERN_ERR, "journaled quota format "
2765                                  "not specified");
2766                         return -EINVAL;
2767                 }
2768         }
2769
2770         return 0;
2771
2772 err_quota_change:
2773         ext4_msg(NULL, KERN_ERR,
2774                  "Cannot change quota options when quota turned on");
2775         return -EINVAL;
2776 err_jquota_change:
2777         ext4_msg(NULL, KERN_ERR, "Cannot change journaled quota "
2778                  "options when quota turned on");
2779         return -EINVAL;
2780 err_jquota_specified:
2781         ext4_msg(NULL, KERN_ERR, "%s quota file already specified",
2782                  QTYPE2NAME(i));
2783         return -EINVAL;
2784 #else
2785         return 0;
2786 #endif
2787 }
2788
2789 static int ext4_check_opt_consistency(struct fs_context *fc,
2790                                       struct super_block *sb)
2791 {
2792         struct ext4_fs_context *ctx = fc->fs_private;
2793         struct ext4_sb_info *sbi = fc->s_fs_info;
2794         int is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE;
2795
2796         if ((ctx->opt_flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
2797                 ext4_msg(NULL, KERN_ERR,
2798                          "Mount option(s) incompatible with ext2");
2799                 return -EINVAL;
2800         }
2801         if ((ctx->opt_flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
2802                 ext4_msg(NULL, KERN_ERR,
2803                          "Mount option(s) incompatible with ext3");
2804                 return -EINVAL;
2805         }
2806
2807         if (ctx->s_want_extra_isize >
2808             (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE)) {
2809                 ext4_msg(NULL, KERN_ERR,
2810                          "Invalid want_extra_isize %d",
2811                          ctx->s_want_extra_isize);
2812                 return -EINVAL;
2813         }
2814
2815         if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DIOREAD_NOLOCK)) {
2816                 int blocksize =
2817                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
2818                 if (blocksize < PAGE_SIZE)
2819                         ext4_msg(NULL, KERN_WARNING, "Warning: mounting with an "
2820                                  "experimental mount option 'dioread_nolock' "
2821                                  "for blocksize < PAGE_SIZE");
2822         }
2823
2824 #ifdef CONFIG_FS_ENCRYPTION
2825         /*
2826          * This mount option is just for testing, and it's not worthwhile to
2827          * implement the extra complexity (e.g. RCU protection) that would be
2828          * needed to allow it to be set or changed during remount.  We do allow
2829          * it to be specified during remount, but only if there is no change.
2830          */
2831         if ((ctx->spec & EXT4_SPEC_DUMMY_ENCRYPTION) &&
2832             is_remount && !sbi->s_dummy_enc_policy.policy) {
2833                 ext4_msg(NULL, KERN_WARNING,
2834                          "Can't set test_dummy_encryption on remount");
2835                 return -1;
2836         }
2837 #endif
2838
2839         if ((ctx->spec & EXT4_SPEC_DATAJ) && is_remount) {
2840                 if (!sbi->s_journal) {
2841                         ext4_msg(NULL, KERN_WARNING,
2842                                  "Remounting file system with no journal "
2843                                  "so ignoring journalled data option");
2844                         ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS);
2845                 } else if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS) !=
2846                            test_opt(sb, DATA_FLAGS)) {
2847                         ext4_msg(NULL, KERN_ERR, "Cannot change data mode "
2848                                  "on remount");
2849                         return -EINVAL;
2850                 }
2851         }
2852
2853         if (is_remount) {
2854                 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) &&
2855                     (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) {
2856                         ext4_msg(NULL, KERN_ERR, "can't mount with "
2857                                  "both data=journal and dax");
2858                         return -EINVAL;
2859                 }
2860
2861                 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) &&
2862                     (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2863                      (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) {
2864 fail_dax_change_remount:
2865                         ext4_msg(NULL, KERN_ERR, "can't change "
2866                                  "dax mount option while remounting");
2867                         return -EINVAL;
2868                 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER) &&
2869                          (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2870                           (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS))) {
2871                         goto fail_dax_change_remount;
2872                 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE) &&
2873                            ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2874                             (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2875                             !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE))) {
2876                         goto fail_dax_change_remount;
2877                 }
2878         }
2879
2880         return ext4_check_quota_consistency(fc, sb);
2881 }
2882
2883 static int ext4_apply_options(struct fs_context *fc, struct super_block *sb)
2884 {
2885         struct ext4_fs_context *ctx = fc->fs_private;
2886         struct ext4_sb_info *sbi = fc->s_fs_info;
2887         int ret = 0;
2888
2889         sbi->s_mount_opt &= ~ctx->mask_s_mount_opt;
2890         sbi->s_mount_opt |= ctx->vals_s_mount_opt;
2891         sbi->s_mount_opt2 &= ~ctx->mask_s_mount_opt2;
2892         sbi->s_mount_opt2 |= ctx->vals_s_mount_opt2;
2893         sbi->s_mount_flags &= ~ctx->mask_s_mount_flags;
2894         sbi->s_mount_flags |= ctx->vals_s_mount_flags;
2895         sb->s_flags &= ~ctx->mask_s_flags;
2896         sb->s_flags |= ctx->vals_s_flags;
2897
2898         /*
2899          * i_version differs from common mount option iversion so we have
2900          * to let vfs know that it was set, otherwise it would get cleared
2901          * on remount
2902          */
2903         if (ctx->mask_s_flags & SB_I_VERSION)
2904                 fc->sb_flags |= SB_I_VERSION;
2905
2906 #define APPLY(X) ({ if (ctx->spec & EXT4_SPEC_##X) sbi->X = ctx->X; })
2907         APPLY(s_commit_interval);
2908         APPLY(s_stripe);
2909         APPLY(s_max_batch_time);
2910         APPLY(s_min_batch_time);
2911         APPLY(s_want_extra_isize);
2912         APPLY(s_inode_readahead_blks);
2913         APPLY(s_max_dir_size_kb);
2914         APPLY(s_li_wait_mult);
2915         APPLY(s_resgid);
2916         APPLY(s_resuid);
2917
2918 #ifdef CONFIG_EXT4_DEBUG
2919         APPLY(s_fc_debug_max_replay);
2920 #endif
2921
2922         ext4_apply_quota_options(fc, sb);
2923
2924         if (ctx->spec & EXT4_SPEC_DUMMY_ENCRYPTION)
2925                 ret = ext4_set_test_dummy_encryption(sb, ctx->test_dummy_enc_arg);
2926
2927         return ret;
2928 }
2929
2930
2931 static int ext4_validate_options(struct fs_context *fc)
2932 {
2933 #ifdef CONFIG_QUOTA
2934         struct ext4_fs_context *ctx = fc->fs_private;
2935         char *usr_qf_name, *grp_qf_name;
2936
2937         usr_qf_name = ctx->s_qf_names[USRQUOTA];
2938         grp_qf_name = ctx->s_qf_names[GRPQUOTA];
2939
2940         if (usr_qf_name || grp_qf_name) {
2941                 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) && usr_qf_name)
2942                         ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA);
2943
2944                 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) && grp_qf_name)
2945                         ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA);
2946
2947                 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) ||
2948                     ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA)) {
2949                         ext4_msg(NULL, KERN_ERR, "old and new quota "
2950                                  "format mixing");
2951                         return -EINVAL;
2952                 }
2953         }
2954 #endif
2955         return 1;
2956 }
2957
2958 static inline void ext4_show_quota_options(struct seq_file *seq,
2959                                            struct super_block *sb)
2960 {
2961 #if defined(CONFIG_QUOTA)
2962         struct ext4_sb_info *sbi = EXT4_SB(sb);
2963         char *usr_qf_name, *grp_qf_name;
2964
2965         if (sbi->s_jquota_fmt) {
2966                 char *fmtname = "";
2967
2968                 switch (sbi->s_jquota_fmt) {
2969                 case QFMT_VFS_OLD:
2970                         fmtname = "vfsold";
2971                         break;
2972                 case QFMT_VFS_V0:
2973                         fmtname = "vfsv0";
2974                         break;
2975                 case QFMT_VFS_V1:
2976                         fmtname = "vfsv1";
2977                         break;
2978                 }
2979                 seq_printf(seq, ",jqfmt=%s", fmtname);
2980         }
2981
2982         rcu_read_lock();
2983         usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]);
2984         grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]);
2985         if (usr_qf_name)
2986                 seq_show_option(seq, "usrjquota", usr_qf_name);
2987         if (grp_qf_name)
2988                 seq_show_option(seq, "grpjquota", grp_qf_name);
2989         rcu_read_unlock();
2990 #endif
2991 }
2992
2993 static const char *token2str(int token)
2994 {
2995         const struct fs_parameter_spec *spec;
2996
2997         for (spec = ext4_param_specs; spec->name != NULL; spec++)
2998                 if (spec->opt == token && !spec->type)
2999                         break;
3000         return spec->name;
3001 }
3002
3003 /*
3004  * Show an option if
3005  *  - it's set to a non-default value OR
3006  *  - if the per-sb default is different from the global default
3007  */
3008 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
3009                               int nodefs)
3010 {
3011         struct ext4_sb_info *sbi = EXT4_SB(sb);
3012         struct ext4_super_block *es = sbi->s_es;
3013         int def_errors, def_mount_opt = sbi->s_def_mount_opt;
3014         const struct mount_opts *m;
3015         char sep = nodefs ? '\n' : ',';
3016
3017 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
3018 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
3019
3020         if (sbi->s_sb_block != 1)
3021                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
3022
3023         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
3024                 int want_set = m->flags & MOPT_SET;
3025                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
3026                     m->flags & MOPT_SKIP)
3027                         continue;
3028                 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
3029                         continue; /* skip if same as the default */
3030                 if ((want_set &&
3031                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
3032                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
3033                         continue; /* select Opt_noFoo vs Opt_Foo */
3034                 SEQ_OPTS_PRINT("%s", token2str(m->token));
3035         }
3036
3037         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
3038             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
3039                 SEQ_OPTS_PRINT("resuid=%u",
3040                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
3041         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
3042             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
3043                 SEQ_OPTS_PRINT("resgid=%u",
3044                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
3045         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
3046         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
3047                 SEQ_OPTS_PUTS("errors=remount-ro");
3048         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
3049                 SEQ_OPTS_PUTS("errors=continue");
3050         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
3051                 SEQ_OPTS_PUTS("errors=panic");
3052         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
3053                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
3054         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
3055                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
3056         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
3057                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
3058         if (sb->s_flags & SB_I_VERSION)
3059                 SEQ_OPTS_PUTS("i_version");
3060         if (nodefs || sbi->s_stripe)
3061                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
3062         if (nodefs || EXT4_MOUNT_DATA_FLAGS &
3063                         (sbi->s_mount_opt ^ def_mount_opt)) {
3064                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3065                         SEQ_OPTS_PUTS("data=journal");
3066                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3067                         SEQ_OPTS_PUTS("data=ordered");
3068                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
3069                         SEQ_OPTS_PUTS("data=writeback");
3070         }
3071         if (nodefs ||
3072             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
3073                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
3074                                sbi->s_inode_readahead_blks);
3075
3076         if (test_opt(sb, INIT_INODE_TABLE) && (nodefs ||
3077                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
3078                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
3079         if (nodefs || sbi->s_max_dir_size_kb)
3080                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
3081         if (test_opt(sb, DATA_ERR_ABORT))
3082                 SEQ_OPTS_PUTS("data_err=abort");
3083
3084         fscrypt_show_test_dummy_encryption(seq, sep, sb);
3085
3086         if (sb->s_flags & SB_INLINECRYPT)
3087                 SEQ_OPTS_PUTS("inlinecrypt");
3088
3089         if (test_opt(sb, DAX_ALWAYS)) {
3090                 if (IS_EXT2_SB(sb))
3091                         SEQ_OPTS_PUTS("dax");
3092                 else
3093                         SEQ_OPTS_PUTS("dax=always");
3094         } else if (test_opt2(sb, DAX_NEVER)) {
3095                 SEQ_OPTS_PUTS("dax=never");
3096         } else if (test_opt2(sb, DAX_INODE)) {
3097                 SEQ_OPTS_PUTS("dax=inode");
3098         }
3099         ext4_show_quota_options(seq, sb);
3100         return 0;
3101 }
3102
3103 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
3104 {
3105         return _ext4_show_options(seq, root->d_sb, 0);
3106 }
3107
3108 int ext4_seq_options_show(struct seq_file *seq, void *offset)
3109 {
3110         struct super_block *sb = seq->private;
3111         int rc;
3112
3113         seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
3114         rc = _ext4_show_options(seq, sb, 1);
3115         seq_puts(seq, "\n");
3116         return rc;
3117 }
3118
3119 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
3120                             int read_only)
3121 {
3122         struct ext4_sb_info *sbi = EXT4_SB(sb);
3123         int err = 0;
3124
3125         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
3126                 ext4_msg(sb, KERN_ERR, "revision level too high, "
3127                          "forcing read-only mode");
3128                 err = -EROFS;
3129                 goto done;
3130         }
3131         if (read_only)
3132                 goto done;
3133         if (!(sbi->s_mount_state & EXT4_VALID_FS))
3134                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
3135                          "running e2fsck is recommended");
3136         else if (sbi->s_mount_state & EXT4_ERROR_FS)
3137                 ext4_msg(sb, KERN_WARNING,
3138                          "warning: mounting fs with errors, "
3139                          "running e2fsck is recommended");
3140         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
3141                  le16_to_cpu(es->s_mnt_count) >=
3142                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
3143                 ext4_msg(sb, KERN_WARNING,
3144                          "warning: maximal mount count reached, "
3145                          "running e2fsck is recommended");
3146         else if (le32_to_cpu(es->s_checkinterval) &&
3147                  (ext4_get_tstamp(es, s_lastcheck) +
3148                   le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds()))
3149                 ext4_msg(sb, KERN_WARNING,
3150                          "warning: checktime reached, "
3151                          "running e2fsck is recommended");
3152         if (!sbi->s_journal)
3153                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
3154         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
3155                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
3156         le16_add_cpu(&es->s_mnt_count, 1);
3157         ext4_update_tstamp(es, s_mtime);
3158         if (sbi->s_journal) {
3159                 ext4_set_feature_journal_needs_recovery(sb);
3160                 if (ext4_has_feature_orphan_file(sb))
3161                         ext4_set_feature_orphan_present(sb);
3162         }
3163
3164         err = ext4_commit_super(sb);
3165 done:
3166         if (test_opt(sb, DEBUG))
3167                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
3168                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
3169                         sb->s_blocksize,
3170                         sbi->s_groups_count,
3171                         EXT4_BLOCKS_PER_GROUP(sb),
3172                         EXT4_INODES_PER_GROUP(sb),
3173                         sbi->s_mount_opt, sbi->s_mount_opt2);
3174         return err;
3175 }
3176
3177 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
3178 {
3179         struct ext4_sb_info *sbi = EXT4_SB(sb);
3180         struct flex_groups **old_groups, **new_groups;
3181         int size, i, j;
3182
3183         if (!sbi->s_log_groups_per_flex)
3184                 return 0;
3185
3186         size = ext4_flex_group(sbi, ngroup - 1) + 1;
3187         if (size <= sbi->s_flex_groups_allocated)
3188                 return 0;
3189
3190         new_groups = kvzalloc(roundup_pow_of_two(size *
3191                               sizeof(*sbi->s_flex_groups)), GFP_KERNEL);
3192         if (!new_groups) {
3193                 ext4_msg(sb, KERN_ERR,
3194                          "not enough memory for %d flex group pointers", size);
3195                 return -ENOMEM;
3196         }
3197         for (i = sbi->s_flex_groups_allocated; i < size; i++) {
3198                 new_groups[i] = kvzalloc(roundup_pow_of_two(
3199                                          sizeof(struct flex_groups)),
3200                                          GFP_KERNEL);
3201                 if (!new_groups[i]) {
3202                         for (j = sbi->s_flex_groups_allocated; j < i; j++)
3203                                 kvfree(new_groups[j]);
3204                         kvfree(new_groups);
3205                         ext4_msg(sb, KERN_ERR,
3206                                  "not enough memory for %d flex groups", size);
3207                         return -ENOMEM;
3208                 }
3209         }
3210         rcu_read_lock();
3211         old_groups = rcu_dereference(sbi->s_flex_groups);
3212         if (old_groups)
3213                 memcpy(new_groups, old_groups,
3214                        (sbi->s_flex_groups_allocated *
3215                         sizeof(struct flex_groups *)));
3216         rcu_read_unlock();
3217         rcu_assign_pointer(sbi->s_flex_groups, new_groups);
3218         sbi->s_flex_groups_allocated = size;
3219         if (old_groups)
3220                 ext4_kvfree_array_rcu(old_groups);
3221         return 0;
3222 }
3223
3224 static int ext4_fill_flex_info(struct super_block *sb)
3225 {
3226         struct ext4_sb_info *sbi = EXT4_SB(sb);
3227         struct ext4_group_desc *gdp = NULL;
3228         struct flex_groups *fg;
3229         ext4_group_t flex_group;
3230         int i, err;
3231
3232         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
3233         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
3234                 sbi->s_log_groups_per_flex = 0;
3235                 return 1;
3236         }
3237
3238         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
3239         if (err)
3240                 goto failed;
3241
3242         for (i = 0; i < sbi->s_groups_count; i++) {
3243                 gdp = ext4_get_group_desc(sb, i, NULL);
3244
3245                 flex_group = ext4_flex_group(sbi, i);
3246                 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
3247                 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes);
3248                 atomic64_add(ext4_free_group_clusters(sb, gdp),
3249                              &fg->free_clusters);
3250                 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs);
3251         }
3252
3253         return 1;
3254 failed:
3255         return 0;
3256 }
3257
3258 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
3259                                    struct ext4_group_desc *gdp)
3260 {
3261         int offset = offsetof(struct ext4_group_desc, bg_checksum);
3262         __u16 crc = 0;
3263         __le32 le_group = cpu_to_le32(block_group);
3264         struct ext4_sb_info *sbi = EXT4_SB(sb);
3265
3266         if (ext4_has_metadata_csum(sbi->s_sb)) {
3267                 /* Use new metadata_csum algorithm */
3268                 __u32 csum32;
3269                 __u16 dummy_csum = 0;
3270
3271                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
3272                                      sizeof(le_group));
3273                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
3274                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
3275                                      sizeof(dummy_csum));
3276                 offset += sizeof(dummy_csum);
3277                 if (offset < sbi->s_desc_size)
3278                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
3279                                              sbi->s_desc_size - offset);
3280
3281                 crc = csum32 & 0xFFFF;
3282                 goto out;
3283         }
3284
3285         /* old crc16 code */
3286         if (!ext4_has_feature_gdt_csum(sb))
3287                 return 0;
3288
3289         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
3290         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
3291         crc = crc16(crc, (__u8 *)gdp, offset);
3292         offset += sizeof(gdp->bg_checksum); /* skip checksum */
3293         /* for checksum of struct ext4_group_desc do the rest...*/
3294         if (ext4_has_feature_64bit(sb) &&
3295             offset < le16_to_cpu(sbi->s_es->s_desc_size))
3296                 crc = crc16(crc, (__u8 *)gdp + offset,
3297                             le16_to_cpu(sbi->s_es->s_desc_size) -
3298                                 offset);
3299
3300 out:
3301         return cpu_to_le16(crc);
3302 }
3303
3304 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
3305                                 struct ext4_group_desc *gdp)
3306 {
3307         if (ext4_has_group_desc_csum(sb) &&
3308             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
3309                 return 0;
3310
3311         return 1;
3312 }
3313
3314 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
3315                               struct ext4_group_desc *gdp)
3316 {
3317         if (!ext4_has_group_desc_csum(sb))
3318                 return;
3319         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
3320 }
3321
3322 /* Called at mount-time, super-block is locked */
3323 static int ext4_check_descriptors(struct super_block *sb,
3324                                   ext4_fsblk_t sb_block,
3325                                   ext4_group_t *first_not_zeroed)
3326 {
3327         struct ext4_sb_info *sbi = EXT4_SB(sb);
3328         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
3329         ext4_fsblk_t last_block;
3330         ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0);
3331         ext4_fsblk_t block_bitmap;
3332         ext4_fsblk_t inode_bitmap;
3333         ext4_fsblk_t inode_table;
3334         int flexbg_flag = 0;
3335         ext4_group_t i, grp = sbi->s_groups_count;
3336
3337         if (ext4_has_feature_flex_bg(sb))
3338                 flexbg_flag = 1;
3339
3340         ext4_debug("Checking group descriptors");
3341
3342         for (i = 0; i < sbi->s_groups_count; i++) {
3343                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
3344
3345                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
3346                         last_block = ext4_blocks_count(sbi->s_es) - 1;
3347                 else
3348                         last_block = first_block +
3349                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
3350
3351                 if ((grp == sbi->s_groups_count) &&
3352                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3353                         grp = i;
3354
3355                 block_bitmap = ext4_block_bitmap(sb, gdp);
3356                 if (block_bitmap == sb_block) {
3357                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3358                                  "Block bitmap for group %u overlaps "
3359                                  "superblock", i);
3360                         if (!sb_rdonly(sb))
3361                                 return 0;
3362                 }
3363                 if (block_bitmap >= sb_block + 1 &&
3364                     block_bitmap <= last_bg_block) {
3365                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3366                                  "Block bitmap for group %u overlaps "
3367                                  "block group descriptors", i);
3368                         if (!sb_rdonly(sb))
3369                                 return 0;
3370                 }
3371                 if (block_bitmap < first_block || block_bitmap > last_block) {
3372                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3373                                "Block bitmap for group %u not in group "
3374                                "(block %llu)!", i, block_bitmap);
3375                         return 0;
3376                 }
3377                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
3378                 if (inode_bitmap == sb_block) {
3379                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3380                                  "Inode bitmap for group %u overlaps "
3381                                  "superblock", i);
3382                         if (!sb_rdonly(sb))
3383                                 return 0;
3384                 }
3385                 if (inode_bitmap >= sb_block + 1 &&
3386                     inode_bitmap <= last_bg_block) {
3387                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3388                                  "Inode bitmap for group %u overlaps "
3389                                  "block group descriptors", i);
3390                         if (!sb_rdonly(sb))
3391                                 return 0;
3392                 }
3393                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
3394                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3395                                "Inode bitmap for group %u not in group "
3396                                "(block %llu)!", i, inode_bitmap);
3397                         return 0;
3398                 }
3399                 inode_table = ext4_inode_table(sb, gdp);
3400                 if (inode_table == sb_block) {
3401                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3402                                  "Inode table for group %u overlaps "
3403                                  "superblock", i);
3404                         if (!sb_rdonly(sb))
3405                                 return 0;
3406                 }
3407                 if (inode_table >= sb_block + 1 &&
3408                     inode_table <= last_bg_block) {
3409                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3410                                  "Inode table for group %u overlaps "
3411                                  "block group descriptors", i);
3412                         if (!sb_rdonly(sb))
3413                                 return 0;
3414                 }
3415                 if (inode_table < first_block ||
3416                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
3417                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3418                                "Inode table for group %u not in group "
3419                                "(block %llu)!", i, inode_table);
3420                         return 0;
3421                 }
3422                 ext4_lock_group(sb, i);
3423                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
3424                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3425                                  "Checksum for group %u failed (%u!=%u)",
3426                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
3427                                      gdp)), le16_to_cpu(gdp->bg_checksum));
3428                         if (!sb_rdonly(sb)) {
3429                                 ext4_unlock_group(sb, i);
3430                                 return 0;
3431                         }
3432                 }
3433                 ext4_unlock_group(sb, i);
3434                 if (!flexbg_flag)
3435                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
3436         }
3437         if (NULL != first_not_zeroed)
3438                 *first_not_zeroed = grp;
3439         return 1;
3440 }
3441
3442 /*
3443  * Maximal extent format file size.
3444  * Resulting logical blkno at s_maxbytes must fit in our on-disk
3445  * extent format containers, within a sector_t, and within i_blocks
3446  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
3447  * so that won't be a limiting factor.
3448  *
3449  * However there is other limiting factor. We do store extents in the form
3450  * of starting block and length, hence the resulting length of the extent
3451  * covering maximum file size must fit into on-disk format containers as
3452  * well. Given that length is always by 1 unit bigger than max unit (because
3453  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
3454  *
3455  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
3456  */
3457 static loff_t ext4_max_size(int blkbits, int has_huge_files)
3458 {
3459         loff_t res;
3460         loff_t upper_limit = MAX_LFS_FILESIZE;
3461
3462         BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64));
3463
3464         if (!has_huge_files) {
3465                 upper_limit = (1LL << 32) - 1;
3466
3467                 /* total blocks in file system block size */
3468                 upper_limit >>= (blkbits - 9);
3469                 upper_limit <<= blkbits;
3470         }
3471
3472         /*
3473          * 32-bit extent-start container, ee_block. We lower the maxbytes
3474          * by one fs block, so ee_len can cover the extent of maximum file
3475          * size
3476          */
3477         res = (1LL << 32) - 1;
3478         res <<= blkbits;
3479
3480         /* Sanity check against vm- & vfs- imposed limits */
3481         if (res > upper_limit)
3482                 res = upper_limit;
3483
3484         return res;
3485 }
3486
3487 /*
3488  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
3489  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
3490  * We need to be 1 filesystem block less than the 2^48 sector limit.
3491  */
3492 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
3493 {
3494         loff_t upper_limit, res = EXT4_NDIR_BLOCKS;
3495         int meta_blocks;
3496         unsigned int ppb = 1 << (bits - 2);
3497
3498         /*
3499          * This is calculated to be the largest file size for a dense, block
3500          * mapped file such that the file's total number of 512-byte sectors,
3501          * including data and all indirect blocks, does not exceed (2^48 - 1).
3502          *
3503          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
3504          * number of 512-byte sectors of the file.
3505          */
3506         if (!has_huge_files) {
3507                 /*
3508                  * !has_huge_files or implies that the inode i_block field
3509                  * represents total file blocks in 2^32 512-byte sectors ==
3510                  * size of vfs inode i_blocks * 8
3511                  */
3512                 upper_limit = (1LL << 32) - 1;
3513
3514                 /* total blocks in file system block size */
3515                 upper_limit >>= (bits - 9);
3516
3517         } else {
3518                 /*
3519                  * We use 48 bit ext4_inode i_blocks
3520                  * With EXT4_HUGE_FILE_FL set the i_blocks
3521                  * represent total number of blocks in
3522                  * file system block size
3523                  */
3524                 upper_limit = (1LL << 48) - 1;
3525
3526         }
3527
3528         /* Compute how many blocks we can address by block tree */
3529         res += ppb;
3530         res += ppb * ppb;
3531         res += ((loff_t)ppb) * ppb * ppb;
3532         /* Compute how many metadata blocks are needed */
3533         meta_blocks = 1;
3534         meta_blocks += 1 + ppb;
3535         meta_blocks += 1 + ppb + ppb * ppb;
3536         /* Does block tree limit file size? */
3537         if (res + meta_blocks <= upper_limit)
3538                 goto check_lfs;
3539
3540         res = upper_limit;
3541         /* How many metadata blocks are needed for addressing upper_limit? */
3542         upper_limit -= EXT4_NDIR_BLOCKS;
3543         /* indirect blocks */
3544         meta_blocks = 1;
3545         upper_limit -= ppb;
3546         /* double indirect blocks */
3547         if (upper_limit < ppb * ppb) {
3548                 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb);
3549                 res -= meta_blocks;
3550                 goto check_lfs;
3551         }
3552         meta_blocks += 1 + ppb;
3553         upper_limit -= ppb * ppb;
3554         /* tripple indirect blocks for the rest */
3555         meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb) +
3556                 DIV_ROUND_UP_ULL(upper_limit, ppb*ppb);
3557         res -= meta_blocks;
3558 check_lfs:
3559         res <<= bits;
3560         if (res > MAX_LFS_FILESIZE)
3561                 res = MAX_LFS_FILESIZE;
3562
3563         return res;
3564 }
3565
3566 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
3567                                    ext4_fsblk_t logical_sb_block, int nr)
3568 {
3569         struct ext4_sb_info *sbi = EXT4_SB(sb);
3570         ext4_group_t bg, first_meta_bg;
3571         int has_super = 0;
3572
3573         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
3574
3575         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
3576                 return logical_sb_block + nr + 1;
3577         bg = sbi->s_desc_per_block * nr;
3578         if (ext4_bg_has_super(sb, bg))
3579                 has_super = 1;
3580
3581         /*
3582          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
3583          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
3584          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
3585          * compensate.
3586          */
3587         if (sb->s_blocksize == 1024 && nr == 0 &&
3588             le32_to_cpu(sbi->s_es->s_first_data_block) == 0)
3589                 has_super++;
3590
3591         return (has_super + ext4_group_first_block_no(sb, bg));
3592 }
3593
3594 /**
3595  * ext4_get_stripe_size: Get the stripe size.
3596  * @sbi: In memory super block info
3597  *
3598  * If we have specified it via mount option, then
3599  * use the mount option value. If the value specified at mount time is
3600  * greater than the blocks per group use the super block value.
3601  * If the super block value is greater than blocks per group return 0.
3602  * Allocator needs it be less than blocks per group.
3603  *
3604  */
3605 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
3606 {
3607         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
3608         unsigned long stripe_width =
3609                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
3610         int ret;
3611
3612         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
3613                 ret = sbi->s_stripe;
3614         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
3615                 ret = stripe_width;
3616         else if (stride && stride <= sbi->s_blocks_per_group)
3617                 ret = stride;
3618         else
3619                 ret = 0;
3620
3621         /*
3622          * If the stripe width is 1, this makes no sense and
3623          * we set it to 0 to turn off stripe handling code.
3624          */
3625         if (ret <= 1)
3626                 ret = 0;
3627
3628         return ret;
3629 }
3630
3631 /*
3632  * Check whether this filesystem can be mounted based on
3633  * the features present and the RDONLY/RDWR mount requested.
3634  * Returns 1 if this filesystem can be mounted as requested,
3635  * 0 if it cannot be.
3636  */
3637 int ext4_feature_set_ok(struct super_block *sb, int readonly)
3638 {
3639         if (ext4_has_unknown_ext4_incompat_features(sb)) {
3640                 ext4_msg(sb, KERN_ERR,
3641                         "Couldn't mount because of "
3642                         "unsupported optional features (%x)",
3643                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
3644                         ~EXT4_FEATURE_INCOMPAT_SUPP));
3645                 return 0;
3646         }
3647
3648 #if !IS_ENABLED(CONFIG_UNICODE)
3649         if (ext4_has_feature_casefold(sb)) {
3650                 ext4_msg(sb, KERN_ERR,
3651                          "Filesystem with casefold feature cannot be "
3652                          "mounted without CONFIG_UNICODE");
3653                 return 0;
3654         }
3655 #endif
3656
3657         if (readonly)
3658                 return 1;
3659
3660         if (ext4_has_feature_readonly(sb)) {
3661                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
3662                 sb->s_flags |= SB_RDONLY;
3663                 return 1;
3664         }
3665
3666         /* Check that feature set is OK for a read-write mount */
3667         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
3668                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
3669                          "unsupported optional features (%x)",
3670                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
3671                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
3672                 return 0;
3673         }
3674         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
3675                 ext4_msg(sb, KERN_ERR,
3676                          "Can't support bigalloc feature without "
3677                          "extents feature\n");
3678                 return 0;
3679         }
3680
3681 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2)
3682         if (!readonly && (ext4_has_feature_quota(sb) ||
3683                           ext4_has_feature_project(sb))) {
3684                 ext4_msg(sb, KERN_ERR,
3685                          "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2");
3686                 return 0;
3687         }
3688 #endif  /* CONFIG_QUOTA */
3689         return 1;
3690 }
3691
3692 /*
3693  * This function is called once a day if we have errors logged
3694  * on the file system
3695  */
3696 static void print_daily_error_info(struct timer_list *t)
3697 {
3698         struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report);
3699         struct super_block *sb = sbi->s_sb;
3700         struct ext4_super_block *es = sbi->s_es;
3701
3702         if (es->s_error_count)
3703                 /* fsck newer than v1.41.13 is needed to clean this condition. */
3704                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
3705                          le32_to_cpu(es->s_error_count));
3706         if (es->s_first_error_time) {
3707                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d",
3708                        sb->s_id,
3709                        ext4_get_tstamp(es, s_first_error_time),
3710                        (int) sizeof(es->s_first_error_func),
3711                        es->s_first_error_func,
3712                        le32_to_cpu(es->s_first_error_line));
3713                 if (es->s_first_error_ino)
3714                         printk(KERN_CONT ": inode %u",
3715                                le32_to_cpu(es->s_first_error_ino));
3716                 if (es->s_first_error_block)
3717                         printk(KERN_CONT ": block %llu", (unsigned long long)
3718                                le64_to_cpu(es->s_first_error_block));
3719                 printk(KERN_CONT "\n");
3720         }
3721         if (es->s_last_error_time) {
3722                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d",
3723                        sb->s_id,
3724                        ext4_get_tstamp(es, s_last_error_time),
3725                        (int) sizeof(es->s_last_error_func),
3726                        es->s_last_error_func,
3727                        le32_to_cpu(es->s_last_error_line));
3728                 if (es->s_last_error_ino)
3729                         printk(KERN_CONT ": inode %u",
3730                                le32_to_cpu(es->s_last_error_ino));
3731                 if (es->s_last_error_block)
3732                         printk(KERN_CONT ": block %llu", (unsigned long long)
3733                                le64_to_cpu(es->s_last_error_block));
3734                 printk(KERN_CONT "\n");
3735         }
3736         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
3737 }
3738
3739 /* Find next suitable group and run ext4_init_inode_table */
3740 static int ext4_run_li_request(struct ext4_li_request *elr)
3741 {
3742         struct ext4_group_desc *gdp = NULL;
3743         struct super_block *sb = elr->lr_super;
3744         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3745         ext4_group_t group = elr->lr_next_group;
3746         unsigned int prefetch_ios = 0;
3747         int ret = 0;
3748         u64 start_time;
3749
3750         if (elr->lr_mode == EXT4_LI_MODE_PREFETCH_BBITMAP) {
3751                 elr->lr_next_group = ext4_mb_prefetch(sb, group,
3752                                 EXT4_SB(sb)->s_mb_prefetch, &prefetch_ios);
3753                 if (prefetch_ios)
3754                         ext4_mb_prefetch_fini(sb, elr->lr_next_group,
3755                                               prefetch_ios);
3756                 trace_ext4_prefetch_bitmaps(sb, group, elr->lr_next_group,
3757                                             prefetch_ios);
3758                 if (group >= elr->lr_next_group) {
3759                         ret = 1;
3760                         if (elr->lr_first_not_zeroed != ngroups &&
3761                             !sb_rdonly(sb) && test_opt(sb, INIT_INODE_TABLE)) {
3762                                 elr->lr_next_group = elr->lr_first_not_zeroed;
3763                                 elr->lr_mode = EXT4_LI_MODE_ITABLE;
3764                                 ret = 0;
3765                         }
3766                 }
3767                 return ret;
3768         }
3769
3770         for (; group < ngroups; group++) {
3771                 gdp = ext4_get_group_desc(sb, group, NULL);
3772                 if (!gdp) {
3773                         ret = 1;
3774                         break;
3775                 }
3776
3777                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3778                         break;
3779         }
3780
3781         if (group >= ngroups)
3782                 ret = 1;
3783
3784         if (!ret) {
3785                 start_time = ktime_get_real_ns();
3786                 ret = ext4_init_inode_table(sb, group,
3787                                             elr->lr_timeout ? 0 : 1);
3788                 trace_ext4_lazy_itable_init(sb, group);
3789                 if (elr->lr_timeout == 0) {
3790                         elr->lr_timeout = nsecs_to_jiffies((ktime_get_real_ns() - start_time) *
3791                                 EXT4_SB(elr->lr_super)->s_li_wait_mult);
3792                 }
3793                 elr->lr_next_sched = jiffies + elr->lr_timeout;
3794                 elr->lr_next_group = group + 1;
3795         }
3796         return ret;
3797 }
3798
3799 /*
3800  * Remove lr_request from the list_request and free the
3801  * request structure. Should be called with li_list_mtx held
3802  */
3803 static void ext4_remove_li_request(struct ext4_li_request *elr)
3804 {
3805         if (!elr)
3806                 return;
3807
3808         list_del(&elr->lr_request);
3809         EXT4_SB(elr->lr_super)->s_li_request = NULL;
3810         kfree(elr);
3811 }
3812
3813 static void ext4_unregister_li_request(struct super_block *sb)
3814 {
3815         mutex_lock(&ext4_li_mtx);
3816         if (!ext4_li_info) {
3817                 mutex_unlock(&ext4_li_mtx);
3818                 return;
3819         }
3820
3821         mutex_lock(&ext4_li_info->li_list_mtx);
3822         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
3823         mutex_unlock(&ext4_li_info->li_list_mtx);
3824         mutex_unlock(&ext4_li_mtx);
3825 }
3826
3827 static struct task_struct *ext4_lazyinit_task;
3828
3829 /*
3830  * This is the function where ext4lazyinit thread lives. It walks
3831  * through the request list searching for next scheduled filesystem.
3832  * When such a fs is found, run the lazy initialization request
3833  * (ext4_rn_li_request) and keep track of the time spend in this
3834  * function. Based on that time we compute next schedule time of
3835  * the request. When walking through the list is complete, compute
3836  * next waking time and put itself into sleep.
3837  */
3838 static int ext4_lazyinit_thread(void *arg)
3839 {
3840         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
3841         struct list_head *pos, *n;
3842         struct ext4_li_request *elr;
3843         unsigned long next_wakeup, cur;
3844
3845         BUG_ON(NULL == eli);
3846
3847 cont_thread:
3848         while (true) {
3849                 next_wakeup = MAX_JIFFY_OFFSET;
3850
3851                 mutex_lock(&eli->li_list_mtx);
3852                 if (list_empty(&eli->li_request_list)) {
3853                         mutex_unlock(&eli->li_list_mtx);
3854                         goto exit_thread;
3855                 }
3856                 list_for_each_safe(pos, n, &eli->li_request_list) {
3857                         int err = 0;
3858                         int progress = 0;
3859                         elr = list_entry(pos, struct ext4_li_request,
3860                                          lr_request);
3861
3862                         if (time_before(jiffies, elr->lr_next_sched)) {
3863                                 if (time_before(elr->lr_next_sched, next_wakeup))
3864                                         next_wakeup = elr->lr_next_sched;
3865                                 continue;
3866                         }
3867                         if (down_read_trylock(&elr->lr_super->s_umount)) {
3868                                 if (sb_start_write_trylock(elr->lr_super)) {
3869                                         progress = 1;
3870                                         /*
3871                                          * We hold sb->s_umount, sb can not
3872                                          * be removed from the list, it is
3873                                          * now safe to drop li_list_mtx
3874                                          */
3875                                         mutex_unlock(&eli->li_list_mtx);
3876                                         err = ext4_run_li_request(elr);
3877                                         sb_end_write(elr->lr_super);
3878                                         mutex_lock(&eli->li_list_mtx);
3879                                         n = pos->next;
3880                                 }
3881                                 up_read((&elr->lr_super->s_umount));
3882                         }
3883                         /* error, remove the lazy_init job */
3884                         if (err) {
3885                                 ext4_remove_li_request(elr);
3886                                 continue;
3887                         }
3888                         if (!progress) {
3889                                 elr->lr_next_sched = jiffies +
3890                                         (prandom_u32()
3891                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3892                         }
3893                         if (time_before(elr->lr_next_sched, next_wakeup))
3894                                 next_wakeup = elr->lr_next_sched;
3895                 }
3896                 mutex_unlock(&eli->li_list_mtx);
3897
3898                 try_to_freeze();
3899
3900                 cur = jiffies;
3901                 if ((time_after_eq(cur, next_wakeup)) ||
3902                     (MAX_JIFFY_OFFSET == next_wakeup)) {
3903                         cond_resched();
3904                         continue;
3905                 }
3906
3907                 schedule_timeout_interruptible(next_wakeup - cur);
3908
3909                 if (kthread_should_stop()) {
3910                         ext4_clear_request_list();
3911                         goto exit_thread;
3912                 }
3913         }
3914
3915 exit_thread:
3916         /*
3917          * It looks like the request list is empty, but we need
3918          * to check it under the li_list_mtx lock, to prevent any
3919          * additions into it, and of course we should lock ext4_li_mtx
3920          * to atomically free the list and ext4_li_info, because at
3921          * this point another ext4 filesystem could be registering
3922          * new one.
3923          */
3924         mutex_lock(&ext4_li_mtx);
3925         mutex_lock(&eli->li_list_mtx);
3926         if (!list_empty(&eli->li_request_list)) {
3927                 mutex_unlock(&eli->li_list_mtx);
3928                 mutex_unlock(&ext4_li_mtx);
3929                 goto cont_thread;
3930         }
3931         mutex_unlock(&eli->li_list_mtx);
3932         kfree(ext4_li_info);
3933         ext4_li_info = NULL;
3934         mutex_unlock(&ext4_li_mtx);
3935
3936         return 0;
3937 }
3938
3939 static void ext4_clear_request_list(void)
3940 {
3941         struct list_head *pos, *n;
3942         struct ext4_li_request *elr;
3943
3944         mutex_lock(&ext4_li_info->li_list_mtx);
3945         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3946                 elr = list_entry(pos, struct ext4_li_request,
3947                                  lr_request);
3948                 ext4_remove_li_request(elr);
3949         }
3950         mutex_unlock(&ext4_li_info->li_list_mtx);
3951 }
3952
3953 static int ext4_run_lazyinit_thread(void)
3954 {
3955         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3956                                          ext4_li_info, "ext4lazyinit");
3957         if (IS_ERR(ext4_lazyinit_task)) {
3958                 int err = PTR_ERR(ext4_lazyinit_task);
3959                 ext4_clear_request_list();
3960                 kfree(ext4_li_info);
3961                 ext4_li_info = NULL;
3962                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3963                                  "initialization thread\n",
3964                                  err);
3965                 return err;
3966         }
3967         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3968         return 0;
3969 }
3970
3971 /*
3972  * Check whether it make sense to run itable init. thread or not.
3973  * If there is at least one uninitialized inode table, return
3974  * corresponding group number, else the loop goes through all
3975  * groups and return total number of groups.
3976  */
3977 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3978 {
3979         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3980         struct ext4_group_desc *gdp = NULL;
3981
3982         if (!ext4_has_group_desc_csum(sb))
3983                 return ngroups;
3984
3985         for (group = 0; group < ngroups; group++) {
3986                 gdp = ext4_get_group_desc(sb, group, NULL);
3987                 if (!gdp)
3988                         continue;
3989
3990                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3991                         break;
3992         }
3993
3994         return group;
3995 }
3996
3997 static int ext4_li_info_new(void)
3998 {
3999         struct ext4_lazy_init *eli = NULL;
4000
4001         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
4002         if (!eli)
4003                 return -ENOMEM;
4004
4005         INIT_LIST_HEAD(&eli->li_request_list);
4006         mutex_init(&eli->li_list_mtx);
4007
4008         eli->li_state |= EXT4_LAZYINIT_QUIT;
4009
4010         ext4_li_info = eli;
4011
4012         return 0;
4013 }
4014
4015 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
4016                                             ext4_group_t start)
4017 {
4018         struct ext4_li_request *elr;
4019
4020         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
4021         if (!elr)
4022                 return NULL;
4023
4024         elr->lr_super = sb;
4025         elr->lr_first_not_zeroed = start;
4026         if (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS)) {
4027                 elr->lr_mode = EXT4_LI_MODE_ITABLE;
4028                 elr->lr_next_group = start;
4029         } else {
4030                 elr->lr_mode = EXT4_LI_MODE_PREFETCH_BBITMAP;
4031         }
4032
4033         /*
4034          * Randomize first schedule time of the request to
4035          * spread the inode table initialization requests
4036          * better.
4037          */
4038         elr->lr_next_sched = jiffies + (prandom_u32() %
4039                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
4040         return elr;
4041 }
4042
4043 int ext4_register_li_request(struct super_block *sb,
4044                              ext4_group_t first_not_zeroed)
4045 {
4046         struct ext4_sb_info *sbi = EXT4_SB(sb);
4047         struct ext4_li_request *elr = NULL;
4048         ext4_group_t ngroups = sbi->s_groups_count;
4049         int ret = 0;
4050
4051         mutex_lock(&ext4_li_mtx);
4052         if (sbi->s_li_request != NULL) {
4053                 /*
4054                  * Reset timeout so it can be computed again, because
4055                  * s_li_wait_mult might have changed.
4056                  */
4057                 sbi->s_li_request->lr_timeout = 0;
4058                 goto out;
4059         }
4060
4061         if (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS) &&
4062             (first_not_zeroed == ngroups || sb_rdonly(sb) ||
4063              !test_opt(sb, INIT_INODE_TABLE)))
4064                 goto out;
4065
4066         elr = ext4_li_request_new(sb, first_not_zeroed);
4067         if (!elr) {
4068                 ret = -ENOMEM;
4069                 goto out;
4070         }
4071
4072         if (NULL == ext4_li_info) {
4073                 ret = ext4_li_info_new();
4074                 if (ret)
4075                         goto out;
4076         }
4077
4078         mutex_lock(&ext4_li_info->li_list_mtx);
4079         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
4080         mutex_unlock(&ext4_li_info->li_list_mtx);
4081
4082         sbi->s_li_request = elr;
4083         /*
4084          * set elr to NULL here since it has been inserted to
4085          * the request_list and the removal and free of it is
4086          * handled by ext4_clear_request_list from now on.
4087          */
4088         elr = NULL;
4089
4090         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
4091                 ret = ext4_run_lazyinit_thread();
4092                 if (ret)
4093                         goto out;
4094         }
4095 out:
4096         mutex_unlock(&ext4_li_mtx);
4097         if (ret)
4098                 kfree(elr);
4099         return ret;
4100 }
4101
4102 /*
4103  * We do not need to lock anything since this is called on
4104  * module unload.
4105  */
4106 static void ext4_destroy_lazyinit_thread(void)
4107 {
4108         /*
4109          * If thread exited earlier
4110          * there's nothing to be done.
4111          */
4112         if (!ext4_li_info || !ext4_lazyinit_task)
4113                 return;
4114
4115         kthread_stop(ext4_lazyinit_task);
4116 }
4117
4118 static int set_journal_csum_feature_set(struct super_block *sb)
4119 {
4120         int ret = 1;
4121         int compat, incompat;
4122         struct ext4_sb_info *sbi = EXT4_SB(sb);
4123
4124         if (ext4_has_metadata_csum(sb)) {
4125                 /* journal checksum v3 */
4126                 compat = 0;
4127                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
4128         } else {
4129                 /* journal checksum v1 */
4130                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
4131                 incompat = 0;
4132         }
4133
4134         jbd2_journal_clear_features(sbi->s_journal,
4135                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
4136                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
4137                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
4138         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4139                 ret = jbd2_journal_set_features(sbi->s_journal,
4140                                 compat, 0,
4141                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
4142                                 incompat);
4143         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
4144                 ret = jbd2_journal_set_features(sbi->s_journal,
4145                                 compat, 0,
4146                                 incompat);
4147                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
4148                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
4149         } else {
4150                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
4151                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
4152         }
4153
4154         return ret;
4155 }
4156
4157 /*
4158  * Note: calculating the overhead so we can be compatible with
4159  * historical BSD practice is quite difficult in the face of
4160  * clusters/bigalloc.  This is because multiple metadata blocks from
4161  * different block group can end up in the same allocation cluster.
4162  * Calculating the exact overhead in the face of clustered allocation
4163  * requires either O(all block bitmaps) in memory or O(number of block
4164  * groups**2) in time.  We will still calculate the superblock for
4165  * older file systems --- and if we come across with a bigalloc file
4166  * system with zero in s_overhead_clusters the estimate will be close to
4167  * correct especially for very large cluster sizes --- but for newer
4168  * file systems, it's better to calculate this figure once at mkfs
4169  * time, and store it in the superblock.  If the superblock value is
4170  * present (even for non-bigalloc file systems), we will use it.
4171  */
4172 static int count_overhead(struct super_block *sb, ext4_group_t grp,
4173                           char *buf)
4174 {
4175         struct ext4_sb_info     *sbi = EXT4_SB(sb);
4176         struct ext4_group_desc  *gdp;
4177         ext4_fsblk_t            first_block, last_block, b;
4178         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
4179         int                     s, j, count = 0;
4180         int                     has_super = ext4_bg_has_super(sb, grp);
4181
4182         if (!ext4_has_feature_bigalloc(sb))
4183                 return (has_super + ext4_bg_num_gdb(sb, grp) +
4184                         (has_super ? le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0) +
4185                         sbi->s_itb_per_group + 2);
4186
4187         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
4188                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
4189         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
4190         for (i = 0; i < ngroups; i++) {
4191                 gdp = ext4_get_group_desc(sb, i, NULL);
4192                 b = ext4_block_bitmap(sb, gdp);
4193                 if (b >= first_block && b <= last_block) {
4194                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
4195                         count++;
4196                 }
4197                 b = ext4_inode_bitmap(sb, gdp);
4198                 if (b >= first_block && b <= last_block) {
4199                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
4200                         count++;
4201                 }
4202                 b = ext4_inode_table(sb, gdp);
4203                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
4204                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
4205                                 int c = EXT4_B2C(sbi, b - first_block);
4206                                 ext4_set_bit(c, buf);
4207                                 count++;
4208                         }
4209                 if (i != grp)
4210                         continue;
4211                 s = 0;
4212                 if (ext4_bg_has_super(sb, grp)) {
4213                         ext4_set_bit(s++, buf);
4214                         count++;
4215                 }
4216                 j = ext4_bg_num_gdb(sb, grp);
4217                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
4218                         ext4_error(sb, "Invalid number of block group "
4219                                    "descriptor blocks: %d", j);
4220                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
4221                 }
4222                 count += j;
4223                 for (; j > 0; j--)
4224                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
4225         }
4226         if (!count)
4227                 return 0;
4228         return EXT4_CLUSTERS_PER_GROUP(sb) -
4229                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
4230 }
4231
4232 /*
4233  * Compute the overhead and stash it in sbi->s_overhead
4234  */
4235 int ext4_calculate_overhead(struct super_block *sb)
4236 {
4237         struct ext4_sb_info *sbi = EXT4_SB(sb);
4238         struct ext4_super_block *es = sbi->s_es;
4239         struct inode *j_inode;
4240         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
4241         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4242         ext4_fsblk_t overhead = 0;
4243         char *buf = (char *) get_zeroed_page(GFP_NOFS);
4244
4245         if (!buf)
4246                 return -ENOMEM;
4247
4248         /*
4249          * Compute the overhead (FS structures).  This is constant
4250          * for a given filesystem unless the number of block groups
4251          * changes so we cache the previous value until it does.
4252          */
4253
4254         /*
4255          * All of the blocks before first_data_block are overhead
4256          */
4257         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
4258
4259         /*
4260          * Add the overhead found in each block group
4261          */
4262         for (i = 0; i < ngroups; i++) {
4263                 int blks;
4264
4265                 blks = count_overhead(sb, i, buf);
4266                 overhead += blks;
4267                 if (blks)
4268                         memset(buf, 0, PAGE_SIZE);
4269                 cond_resched();
4270         }
4271
4272         /*
4273          * Add the internal journal blocks whether the journal has been
4274          * loaded or not
4275          */
4276         if (sbi->s_journal && !sbi->s_journal_bdev)
4277                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_total_len);
4278         else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) {
4279                 /* j_inum for internal journal is non-zero */
4280                 j_inode = ext4_get_journal_inode(sb, j_inum);
4281                 if (j_inode) {
4282                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
4283                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
4284                         iput(j_inode);
4285                 } else {
4286                         ext4_msg(sb, KERN_ERR, "can't get journal size");
4287                 }
4288         }
4289         sbi->s_overhead = overhead;
4290         smp_wmb();
4291         free_page((unsigned long) buf);
4292         return 0;
4293 }
4294
4295 static void ext4_set_resv_clusters(struct super_block *sb)
4296 {
4297         ext4_fsblk_t resv_clusters;
4298         struct ext4_sb_info *sbi = EXT4_SB(sb);
4299
4300         /*
4301          * There's no need to reserve anything when we aren't using extents.
4302          * The space estimates are exact, there are no unwritten extents,
4303          * hole punching doesn't need new metadata... This is needed especially
4304          * to keep ext2/3 backward compatibility.
4305          */
4306         if (!ext4_has_feature_extents(sb))
4307                 return;
4308         /*
4309          * By default we reserve 2% or 4096 clusters, whichever is smaller.
4310          * This should cover the situations where we can not afford to run
4311          * out of space like for example punch hole, or converting
4312          * unwritten extents in delalloc path. In most cases such
4313          * allocation would require 1, or 2 blocks, higher numbers are
4314          * very rare.
4315          */
4316         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
4317                          sbi->s_cluster_bits);
4318
4319         do_div(resv_clusters, 50);
4320         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
4321
4322         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
4323 }
4324
4325 static const char *ext4_quota_mode(struct super_block *sb)
4326 {
4327 #ifdef CONFIG_QUOTA
4328         if (!ext4_quota_capable(sb))
4329                 return "none";
4330
4331         if (EXT4_SB(sb)->s_journal && ext4_is_quota_journalled(sb))
4332                 return "journalled";
4333         else
4334                 return "writeback";
4335 #else
4336         return "disabled";
4337 #endif
4338 }
4339
4340 static void ext4_setup_csum_trigger(struct super_block *sb,
4341                                     enum ext4_journal_trigger_type type,
4342                                     void (*trigger)(
4343                                         struct jbd2_buffer_trigger_type *type,
4344                                         struct buffer_head *bh,
4345                                         void *mapped_data,
4346                                         size_t size))
4347 {
4348         struct ext4_sb_info *sbi = EXT4_SB(sb);
4349
4350         sbi->s_journal_triggers[type].sb = sb;
4351         sbi->s_journal_triggers[type].tr_triggers.t_frozen = trigger;
4352 }
4353
4354 static void ext4_free_sbi(struct ext4_sb_info *sbi)
4355 {
4356         if (!sbi)
4357                 return;
4358
4359         kfree(sbi->s_blockgroup_lock);
4360         fs_put_dax(sbi->s_daxdev);
4361         kfree(sbi);
4362 }
4363
4364 static struct ext4_sb_info *ext4_alloc_sbi(struct super_block *sb)
4365 {
4366         struct ext4_sb_info *sbi;
4367
4368         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
4369         if (!sbi)
4370                 return NULL;
4371
4372         sbi->s_daxdev = fs_dax_get_by_bdev(sb->s_bdev, &sbi->s_dax_part_off);
4373
4374         sbi->s_blockgroup_lock =
4375                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
4376
4377         if (!sbi->s_blockgroup_lock)
4378                 goto err_out;
4379
4380         sb->s_fs_info = sbi;
4381         sbi->s_sb = sb;
4382         return sbi;
4383 err_out:
4384         fs_put_dax(sbi->s_daxdev);
4385         kfree(sbi);
4386         return NULL;
4387 }
4388
4389 static int __ext4_fill_super(struct fs_context *fc, struct super_block *sb)
4390 {
4391         struct buffer_head *bh, **group_desc;
4392         struct ext4_super_block *es = NULL;
4393         struct ext4_sb_info *sbi = EXT4_SB(sb);
4394         struct flex_groups **flex_groups;
4395         ext4_fsblk_t block;
4396         ext4_fsblk_t logical_sb_block;
4397         unsigned long offset = 0;
4398         unsigned long def_mount_opts;
4399         struct inode *root;
4400         int ret = -ENOMEM;
4401         int blocksize, clustersize;
4402         unsigned int db_count;
4403         unsigned int i;
4404         int needs_recovery, has_huge_files;
4405         __u64 blocks_count;
4406         int err = 0;
4407         ext4_group_t first_not_zeroed;
4408         struct ext4_fs_context *ctx = fc->fs_private;
4409         int silent = fc->sb_flags & SB_SILENT;
4410
4411         /* Set defaults for the variables that will be set during parsing */
4412         ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4413
4414         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
4415         sbi->s_sectors_written_start =
4416                 part_stat_read(sb->s_bdev, sectors[STAT_WRITE]);
4417
4418         /* -EINVAL is default */
4419         ret = -EINVAL;
4420         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
4421         if (!blocksize) {
4422                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
4423                 goto out_fail;
4424         }
4425
4426         /*
4427          * The ext4 superblock will not be buffer aligned for other than 1kB
4428          * block sizes.  We need to calculate the offset from buffer start.
4429          */
4430         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
4431                 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE;
4432                 offset = do_div(logical_sb_block, blocksize);
4433         } else {
4434                 logical_sb_block = sbi->s_sb_block;
4435         }
4436
4437         bh = ext4_sb_bread_unmovable(sb, logical_sb_block);
4438         if (IS_ERR(bh)) {
4439                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
4440                 ret = PTR_ERR(bh);
4441                 goto out_fail;
4442         }
4443         /*
4444          * Note: s_es must be initialized as soon as possible because
4445          *       some ext4 macro-instructions depend on its value
4446          */
4447         es = (struct ext4_super_block *) (bh->b_data + offset);
4448         sbi->s_es = es;
4449         sb->s_magic = le16_to_cpu(es->s_magic);
4450         if (sb->s_magic != EXT4_SUPER_MAGIC)
4451                 goto cantfind_ext4;
4452         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
4453
4454         /* Warn if metadata_csum and gdt_csum are both set. */
4455         if (ext4_has_feature_metadata_csum(sb) &&
4456             ext4_has_feature_gdt_csum(sb))
4457                 ext4_warning(sb, "metadata_csum and uninit_bg are "
4458                              "redundant flags; please run fsck.");
4459
4460         /* Check for a known checksum algorithm */
4461         if (!ext4_verify_csum_type(sb, es)) {
4462                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
4463                          "unknown checksum algorithm.");
4464                 silent = 1;
4465                 goto cantfind_ext4;
4466         }
4467         ext4_setup_csum_trigger(sb, EXT4_JTR_ORPHAN_FILE,
4468                                 ext4_orphan_file_block_trigger);
4469
4470         /* Load the checksum driver */
4471         sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
4472         if (IS_ERR(sbi->s_chksum_driver)) {
4473                 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
4474                 ret = PTR_ERR(sbi->s_chksum_driver);
4475                 sbi->s_chksum_driver = NULL;
4476                 goto failed_mount;
4477         }
4478
4479         /* Check superblock checksum */
4480         if (!ext4_superblock_csum_verify(sb, es)) {
4481                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
4482                          "invalid superblock checksum.  Run e2fsck?");
4483                 silent = 1;
4484                 ret = -EFSBADCRC;
4485                 goto cantfind_ext4;
4486         }
4487
4488         /* Precompute checksum seed for all metadata */
4489         if (ext4_has_feature_csum_seed(sb))
4490                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
4491         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
4492                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
4493                                                sizeof(es->s_uuid));
4494
4495         /* Set defaults before we parse the mount options */
4496         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
4497         set_opt(sb, INIT_INODE_TABLE);
4498         if (def_mount_opts & EXT4_DEFM_DEBUG)
4499                 set_opt(sb, DEBUG);
4500         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
4501                 set_opt(sb, GRPID);
4502         if (def_mount_opts & EXT4_DEFM_UID16)
4503                 set_opt(sb, NO_UID32);
4504         /* xattr user namespace & acls are now defaulted on */
4505         set_opt(sb, XATTR_USER);
4506 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4507         set_opt(sb, POSIX_ACL);
4508 #endif
4509         if (ext4_has_feature_fast_commit(sb))
4510                 set_opt2(sb, JOURNAL_FAST_COMMIT);
4511         /* don't forget to enable journal_csum when metadata_csum is enabled. */
4512         if (ext4_has_metadata_csum(sb))
4513                 set_opt(sb, JOURNAL_CHECKSUM);
4514
4515         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
4516                 set_opt(sb, JOURNAL_DATA);
4517         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
4518                 set_opt(sb, ORDERED_DATA);
4519         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
4520                 set_opt(sb, WRITEBACK_DATA);
4521
4522         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
4523                 set_opt(sb, ERRORS_PANIC);
4524         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
4525                 set_opt(sb, ERRORS_CONT);
4526         else
4527                 set_opt(sb, ERRORS_RO);
4528         /* block_validity enabled by default; disable with noblock_validity */
4529         set_opt(sb, BLOCK_VALIDITY);
4530         if (def_mount_opts & EXT4_DEFM_DISCARD)
4531                 set_opt(sb, DISCARD);
4532
4533         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
4534         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
4535         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
4536         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
4537         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
4538
4539         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
4540                 set_opt(sb, BARRIER);
4541
4542         /*
4543          * enable delayed allocation by default
4544          * Use -o nodelalloc to turn it off
4545          */
4546         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
4547             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
4548                 set_opt(sb, DELALLOC);
4549
4550         /*
4551          * set default s_li_wait_mult for lazyinit, for the case there is
4552          * no mount option specified.
4553          */
4554         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
4555
4556         if (le32_to_cpu(es->s_log_block_size) >
4557             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4558                 ext4_msg(sb, KERN_ERR,
4559                          "Invalid log block size: %u",
4560                          le32_to_cpu(es->s_log_block_size));
4561                 goto failed_mount;
4562         }
4563         if (le32_to_cpu(es->s_log_cluster_size) >
4564             (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4565                 ext4_msg(sb, KERN_ERR,
4566                          "Invalid log cluster size: %u",
4567                          le32_to_cpu(es->s_log_cluster_size));
4568                 goto failed_mount;
4569         }
4570
4571         blocksize = EXT4_MIN_BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
4572
4573         if (blocksize == PAGE_SIZE)
4574                 set_opt(sb, DIOREAD_NOLOCK);
4575
4576         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
4577                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
4578                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
4579         } else {
4580                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
4581                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
4582                 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) {
4583                         ext4_msg(sb, KERN_ERR, "invalid first ino: %u",
4584                                  sbi->s_first_ino);
4585                         goto failed_mount;
4586                 }
4587                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
4588                     (!is_power_of_2(sbi->s_inode_size)) ||
4589                     (sbi->s_inode_size > blocksize)) {
4590                         ext4_msg(sb, KERN_ERR,
4591                                "unsupported inode size: %d",
4592                                sbi->s_inode_size);
4593                         ext4_msg(sb, KERN_ERR, "blocksize: %d", blocksize);
4594                         goto failed_mount;
4595                 }
4596                 /*
4597                  * i_atime_extra is the last extra field available for
4598                  * [acm]times in struct ext4_inode. Checking for that
4599                  * field should suffice to ensure we have extra space
4600                  * for all three.
4601                  */
4602                 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) +
4603                         sizeof(((struct ext4_inode *)0)->i_atime_extra)) {
4604                         sb->s_time_gran = 1;
4605                         sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX;
4606                 } else {
4607                         sb->s_time_gran = NSEC_PER_SEC;
4608                         sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX;
4609                 }
4610                 sb->s_time_min = EXT4_TIMESTAMP_MIN;
4611         }
4612         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
4613                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4614                         EXT4_GOOD_OLD_INODE_SIZE;
4615                 if (ext4_has_feature_extra_isize(sb)) {
4616                         unsigned v, max = (sbi->s_inode_size -
4617                                            EXT4_GOOD_OLD_INODE_SIZE);
4618
4619                         v = le16_to_cpu(es->s_want_extra_isize);
4620                         if (v > max) {
4621                                 ext4_msg(sb, KERN_ERR,
4622                                          "bad s_want_extra_isize: %d", v);
4623                                 goto failed_mount;
4624                         }
4625                         if (sbi->s_want_extra_isize < v)
4626                                 sbi->s_want_extra_isize = v;
4627
4628                         v = le16_to_cpu(es->s_min_extra_isize);
4629                         if (v > max) {
4630                                 ext4_msg(sb, KERN_ERR,
4631                                          "bad s_min_extra_isize: %d", v);
4632                                 goto failed_mount;
4633                         }
4634                         if (sbi->s_want_extra_isize < v)
4635                                 sbi->s_want_extra_isize = v;
4636                 }
4637         }
4638
4639         err = parse_apply_sb_mount_options(sb, ctx);
4640         if (err < 0)
4641                 goto failed_mount;
4642
4643         sbi->s_def_mount_opt = sbi->s_mount_opt;
4644
4645         err = ext4_check_opt_consistency(fc, sb);
4646         if (err < 0)
4647                 goto failed_mount;
4648
4649         err = ext4_apply_options(fc, sb);
4650         if (err < 0)
4651                 goto failed_mount;
4652
4653 #if IS_ENABLED(CONFIG_UNICODE)
4654         if (ext4_has_feature_casefold(sb) && !sb->s_encoding) {
4655                 const struct ext4_sb_encodings *encoding_info;
4656                 struct unicode_map *encoding;
4657                 __u16 encoding_flags = le16_to_cpu(es->s_encoding_flags);
4658
4659                 encoding_info = ext4_sb_read_encoding(es);
4660                 if (!encoding_info) {
4661                         ext4_msg(sb, KERN_ERR,
4662                                  "Encoding requested by superblock is unknown");
4663                         goto failed_mount;
4664                 }
4665
4666                 encoding = utf8_load(encoding_info->version);
4667                 if (IS_ERR(encoding)) {
4668                         ext4_msg(sb, KERN_ERR,
4669                                  "can't mount with superblock charset: %s-%u.%u.%u "
4670                                  "not supported by the kernel. flags: 0x%x.",
4671                                  encoding_info->name,
4672                                  unicode_major(encoding_info->version),
4673                                  unicode_minor(encoding_info->version),
4674                                  unicode_rev(encoding_info->version),
4675                                  encoding_flags);
4676                         goto failed_mount;
4677                 }
4678                 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: "
4679                          "%s-%u.%u.%u with flags 0x%hx", encoding_info->name,
4680                          unicode_major(encoding_info->version),
4681                          unicode_minor(encoding_info->version),
4682                          unicode_rev(encoding_info->version),
4683                          encoding_flags);
4684
4685                 sb->s_encoding = encoding;
4686                 sb->s_encoding_flags = encoding_flags;
4687         }
4688 #endif
4689
4690         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4691                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with data=journal disables delayed allocation, dioread_nolock, O_DIRECT and fast_commit support!\n");
4692                 /* can't mount with both data=journal and dioread_nolock. */
4693                 clear_opt(sb, DIOREAD_NOLOCK);
4694                 clear_opt2(sb, JOURNAL_FAST_COMMIT);
4695                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4696                         ext4_msg(sb, KERN_ERR, "can't mount with "
4697                                  "both data=journal and delalloc");
4698                         goto failed_mount;
4699                 }
4700                 if (test_opt(sb, DAX_ALWAYS)) {
4701                         ext4_msg(sb, KERN_ERR, "can't mount with "
4702                                  "both data=journal and dax");
4703                         goto failed_mount;
4704                 }
4705                 if (ext4_has_feature_encrypt(sb)) {
4706                         ext4_msg(sb, KERN_WARNING,
4707                                  "encrypted files will use data=ordered "
4708                                  "instead of data journaling mode");
4709                 }
4710                 if (test_opt(sb, DELALLOC))
4711                         clear_opt(sb, DELALLOC);
4712         } else {
4713                 sb->s_iflags |= SB_I_CGROUPWB;
4714         }
4715
4716         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
4717                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
4718
4719         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
4720             (ext4_has_compat_features(sb) ||
4721              ext4_has_ro_compat_features(sb) ||
4722              ext4_has_incompat_features(sb)))
4723                 ext4_msg(sb, KERN_WARNING,
4724                        "feature flags set on rev 0 fs, "
4725                        "running e2fsck is recommended");
4726
4727         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
4728                 set_opt2(sb, HURD_COMPAT);
4729                 if (ext4_has_feature_64bit(sb)) {
4730                         ext4_msg(sb, KERN_ERR,
4731                                  "The Hurd can't support 64-bit file systems");
4732                         goto failed_mount;
4733                 }
4734
4735                 /*
4736                  * ea_inode feature uses l_i_version field which is not
4737                  * available in HURD_COMPAT mode.
4738                  */
4739                 if (ext4_has_feature_ea_inode(sb)) {
4740                         ext4_msg(sb, KERN_ERR,
4741                                  "ea_inode feature is not supported for Hurd");
4742                         goto failed_mount;
4743                 }
4744         }
4745
4746         if (IS_EXT2_SB(sb)) {
4747                 if (ext2_feature_set_ok(sb))
4748                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
4749                                  "using the ext4 subsystem");
4750                 else {
4751                         /*
4752                          * If we're probing be silent, if this looks like
4753                          * it's actually an ext[34] filesystem.
4754                          */
4755                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4756                                 goto failed_mount;
4757                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
4758                                  "to feature incompatibilities");
4759                         goto failed_mount;
4760                 }
4761         }
4762
4763         if (IS_EXT3_SB(sb)) {
4764                 if (ext3_feature_set_ok(sb))
4765                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
4766                                  "using the ext4 subsystem");
4767                 else {
4768                         /*
4769                          * If we're probing be silent, if this looks like
4770                          * it's actually an ext4 filesystem.
4771                          */
4772                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4773                                 goto failed_mount;
4774                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
4775                                  "to feature incompatibilities");
4776                         goto failed_mount;
4777                 }
4778         }
4779
4780         /*
4781          * Check feature flags regardless of the revision level, since we
4782          * previously didn't change the revision level when setting the flags,
4783          * so there is a chance incompat flags are set on a rev 0 filesystem.
4784          */
4785         if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
4786                 goto failed_mount;
4787
4788         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
4789                 ext4_msg(sb, KERN_ERR,
4790                          "Number of reserved GDT blocks insanely large: %d",
4791                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
4792                 goto failed_mount;
4793         }
4794
4795         if (sbi->s_daxdev) {
4796                 if (blocksize == PAGE_SIZE)
4797                         set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags);
4798                 else
4799                         ext4_msg(sb, KERN_ERR, "unsupported blocksize for DAX\n");
4800         }
4801
4802         if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) {
4803                 if (ext4_has_feature_inline_data(sb)) {
4804                         ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
4805                                         " that may contain inline data");
4806                         goto failed_mount;
4807                 }
4808                 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) {
4809                         ext4_msg(sb, KERN_ERR,
4810                                 "DAX unsupported by block device.");
4811                         goto failed_mount;
4812                 }
4813         }
4814
4815         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
4816                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
4817                          es->s_encryption_level);
4818                 goto failed_mount;
4819         }
4820
4821         if (sb->s_blocksize != blocksize) {
4822                 /*
4823                  * bh must be released before kill_bdev(), otherwise
4824                  * it won't be freed and its page also. kill_bdev()
4825                  * is called by sb_set_blocksize().
4826                  */
4827                 brelse(bh);
4828                 /* Validate the filesystem blocksize */
4829                 if (!sb_set_blocksize(sb, blocksize)) {
4830                         ext4_msg(sb, KERN_ERR, "bad block size %d",
4831                                         blocksize);
4832                         bh = NULL;
4833                         goto failed_mount;
4834                 }
4835
4836                 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE;
4837                 offset = do_div(logical_sb_block, blocksize);
4838                 bh = ext4_sb_bread_unmovable(sb, logical_sb_block);
4839                 if (IS_ERR(bh)) {
4840                         ext4_msg(sb, KERN_ERR,
4841                                "Can't read superblock on 2nd try");
4842                         ret = PTR_ERR(bh);
4843                         bh = NULL;
4844                         goto failed_mount;
4845                 }
4846                 es = (struct ext4_super_block *)(bh->b_data + offset);
4847                 sbi->s_es = es;
4848                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
4849                         ext4_msg(sb, KERN_ERR,
4850                                "Magic mismatch, very weird!");
4851                         goto failed_mount;
4852                 }
4853         }
4854
4855         has_huge_files = ext4_has_feature_huge_file(sb);
4856         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
4857                                                       has_huge_files);
4858         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
4859
4860         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
4861         if (ext4_has_feature_64bit(sb)) {
4862                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
4863                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
4864                     !is_power_of_2(sbi->s_desc_size)) {
4865                         ext4_msg(sb, KERN_ERR,
4866                                "unsupported descriptor size %lu",
4867                                sbi->s_desc_size);
4868                         goto failed_mount;
4869                 }
4870         } else
4871                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
4872
4873         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
4874         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
4875
4876         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
4877         if (sbi->s_inodes_per_block == 0)
4878                 goto cantfind_ext4;
4879         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
4880             sbi->s_inodes_per_group > blocksize * 8) {
4881                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
4882                          sbi->s_inodes_per_group);
4883                 goto failed_mount;
4884         }
4885         sbi->s_itb_per_group = sbi->s_inodes_per_group /
4886                                         sbi->s_inodes_per_block;
4887         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
4888         sbi->s_sbh = bh;
4889         sbi->s_mount_state = le16_to_cpu(es->s_state);
4890         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
4891         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
4892
4893         for (i = 0; i < 4; i++)
4894                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
4895         sbi->s_def_hash_version = es->s_def_hash_version;
4896         if (ext4_has_feature_dir_index(sb)) {
4897                 i = le32_to_cpu(es->s_flags);
4898                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
4899                         sbi->s_hash_unsigned = 3;
4900                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
4901 #ifdef __CHAR_UNSIGNED__
4902                         if (!sb_rdonly(sb))
4903                                 es->s_flags |=
4904                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
4905                         sbi->s_hash_unsigned = 3;
4906 #else
4907                         if (!sb_rdonly(sb))
4908                                 es->s_flags |=
4909                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
4910 #endif
4911                 }
4912         }
4913
4914         /* Handle clustersize */
4915         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
4916         if (ext4_has_feature_bigalloc(sb)) {
4917                 if (clustersize < blocksize) {
4918                         ext4_msg(sb, KERN_ERR,
4919                                  "cluster size (%d) smaller than "
4920                                  "block size (%d)", clustersize, blocksize);
4921                         goto failed_mount;
4922                 }
4923                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
4924                         le32_to_cpu(es->s_log_block_size);
4925                 sbi->s_clusters_per_group =
4926                         le32_to_cpu(es->s_clusters_per_group);
4927                 if (sbi->s_clusters_per_group > blocksize * 8) {
4928                         ext4_msg(sb, KERN_ERR,
4929                                  "#clusters per group too big: %lu",
4930                                  sbi->s_clusters_per_group);
4931                         goto failed_mount;
4932                 }
4933                 if (sbi->s_blocks_per_group !=
4934                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
4935                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
4936                                  "clusters per group (%lu) inconsistent",
4937                                  sbi->s_blocks_per_group,
4938                                  sbi->s_clusters_per_group);
4939                         goto failed_mount;
4940                 }
4941         } else {
4942                 if (clustersize != blocksize) {
4943                         ext4_msg(sb, KERN_ERR,
4944                                  "fragment/cluster size (%d) != "
4945                                  "block size (%d)", clustersize, blocksize);
4946                         goto failed_mount;
4947                 }
4948                 if (sbi->s_blocks_per_group > blocksize * 8) {
4949                         ext4_msg(sb, KERN_ERR,
4950                                  "#blocks per group too big: %lu",
4951                                  sbi->s_blocks_per_group);
4952                         goto failed_mount;
4953                 }
4954                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
4955                 sbi->s_cluster_bits = 0;
4956         }
4957         sbi->s_cluster_ratio = clustersize / blocksize;
4958
4959         /* Do we have standard group size of clustersize * 8 blocks ? */
4960         if (sbi->s_blocks_per_group == clustersize << 3)
4961                 set_opt2(sb, STD_GROUP_SIZE);
4962
4963         /*
4964          * Test whether we have more sectors than will fit in sector_t,
4965          * and whether the max offset is addressable by the page cache.
4966          */
4967         err = generic_check_addressable(sb->s_blocksize_bits,
4968                                         ext4_blocks_count(es));
4969         if (err) {
4970                 ext4_msg(sb, KERN_ERR, "filesystem"
4971                          " too large to mount safely on this system");
4972                 goto failed_mount;
4973         }
4974
4975         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
4976                 goto cantfind_ext4;
4977
4978         /* check blocks count against device size */
4979         blocks_count = sb_bdev_nr_blocks(sb);
4980         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
4981                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
4982                        "exceeds size of device (%llu blocks)",
4983                        ext4_blocks_count(es), blocks_count);
4984                 goto failed_mount;
4985         }
4986
4987         /*
4988          * It makes no sense for the first data block to be beyond the end
4989          * of the filesystem.
4990          */
4991         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
4992                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4993                          "block %u is beyond end of filesystem (%llu)",
4994                          le32_to_cpu(es->s_first_data_block),
4995                          ext4_blocks_count(es));
4996                 goto failed_mount;
4997         }
4998         if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) &&
4999             (sbi->s_cluster_ratio == 1)) {
5000                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
5001                          "block is 0 with a 1k block and cluster size");
5002                 goto failed_mount;
5003         }
5004
5005         blocks_count = (ext4_blocks_count(es) -
5006                         le32_to_cpu(es->s_first_data_block) +
5007                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
5008         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
5009         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
5010                 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu "
5011                        "(block count %llu, first data block %u, "
5012                        "blocks per group %lu)", blocks_count,
5013                        ext4_blocks_count(es),
5014                        le32_to_cpu(es->s_first_data_block),
5015                        EXT4_BLOCKS_PER_GROUP(sb));
5016                 goto failed_mount;
5017         }
5018         sbi->s_groups_count = blocks_count;
5019         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
5020                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
5021         if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) !=
5022             le32_to_cpu(es->s_inodes_count)) {
5023                 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu",
5024                          le32_to_cpu(es->s_inodes_count),
5025                          ((u64)sbi->s_groups_count * sbi->s_inodes_per_group));
5026                 ret = -EINVAL;
5027                 goto failed_mount;
5028         }
5029         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
5030                    EXT4_DESC_PER_BLOCK(sb);
5031         if (ext4_has_feature_meta_bg(sb)) {
5032                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
5033                         ext4_msg(sb, KERN_WARNING,
5034                                  "first meta block group too large: %u "
5035                                  "(group descriptor block count %u)",
5036                                  le32_to_cpu(es->s_first_meta_bg), db_count);
5037                         goto failed_mount;
5038                 }
5039         }
5040         rcu_assign_pointer(sbi->s_group_desc,
5041                            kvmalloc_array(db_count,
5042                                           sizeof(struct buffer_head *),
5043                                           GFP_KERNEL));
5044         if (sbi->s_group_desc == NULL) {
5045                 ext4_msg(sb, KERN_ERR, "not enough memory");
5046                 ret = -ENOMEM;
5047                 goto failed_mount;
5048         }
5049
5050         bgl_lock_init(sbi->s_blockgroup_lock);
5051
5052         /* Pre-read the descriptors into the buffer cache */
5053         for (i = 0; i < db_count; i++) {
5054                 block = descriptor_loc(sb, logical_sb_block, i);
5055                 ext4_sb_breadahead_unmovable(sb, block);
5056         }
5057
5058         for (i = 0; i < db_count; i++) {
5059                 struct buffer_head *bh;
5060
5061                 block = descriptor_loc(sb, logical_sb_block, i);
5062                 bh = ext4_sb_bread_unmovable(sb, block);
5063                 if (IS_ERR(bh)) {
5064                         ext4_msg(sb, KERN_ERR,
5065                                "can't read group descriptor %d", i);
5066                         db_count = i;
5067                         ret = PTR_ERR(bh);
5068                         goto failed_mount2;
5069                 }
5070                 rcu_read_lock();
5071                 rcu_dereference(sbi->s_group_desc)[i] = bh;
5072                 rcu_read_unlock();
5073         }
5074         sbi->s_gdb_count = db_count;
5075         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
5076                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
5077                 ret = -EFSCORRUPTED;
5078                 goto failed_mount2;
5079         }
5080
5081         timer_setup(&sbi->s_err_report, print_daily_error_info, 0);
5082         spin_lock_init(&sbi->s_error_lock);
5083         INIT_WORK(&sbi->s_error_work, flush_stashed_error_work);
5084
5085         /* Register extent status tree shrinker */
5086         if (ext4_es_register_shrinker(sbi))
5087                 goto failed_mount3;
5088
5089         sbi->s_stripe = ext4_get_stripe_size(sbi);
5090         sbi->s_extent_max_zeroout_kb = 32;
5091
5092         /*
5093          * set up enough so that it can read an inode
5094          */
5095         sb->s_op = &ext4_sops;
5096         sb->s_export_op = &ext4_export_ops;
5097         sb->s_xattr = ext4_xattr_handlers;
5098 #ifdef CONFIG_FS_ENCRYPTION
5099         sb->s_cop = &ext4_cryptops;
5100 #endif
5101 #ifdef CONFIG_FS_VERITY
5102         sb->s_vop = &ext4_verityops;
5103 #endif
5104 #ifdef CONFIG_QUOTA
5105         sb->dq_op = &ext4_quota_operations;
5106         if (ext4_has_feature_quota(sb))
5107                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
5108         else
5109                 sb->s_qcop = &ext4_qctl_operations;
5110         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
5111 #endif
5112         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
5113
5114         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
5115         mutex_init(&sbi->s_orphan_lock);
5116
5117         /* Initialize fast commit stuff */
5118         atomic_set(&sbi->s_fc_subtid, 0);
5119         INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_MAIN]);
5120         INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_STAGING]);
5121         INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_MAIN]);
5122         INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_STAGING]);
5123         sbi->s_fc_bytes = 0;
5124         ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE);
5125         sbi->s_fc_ineligible_tid = 0;
5126         spin_lock_init(&sbi->s_fc_lock);
5127         memset(&sbi->s_fc_stats, 0, sizeof(sbi->s_fc_stats));
5128         sbi->s_fc_replay_state.fc_regions = NULL;
5129         sbi->s_fc_replay_state.fc_regions_size = 0;
5130         sbi->s_fc_replay_state.fc_regions_used = 0;
5131         sbi->s_fc_replay_state.fc_regions_valid = 0;
5132         sbi->s_fc_replay_state.fc_modified_inodes = NULL;
5133         sbi->s_fc_replay_state.fc_modified_inodes_size = 0;
5134         sbi->s_fc_replay_state.fc_modified_inodes_used = 0;
5135
5136         sb->s_root = NULL;
5137
5138         needs_recovery = (es->s_last_orphan != 0 ||
5139                           ext4_has_feature_orphan_present(sb) ||
5140                           ext4_has_feature_journal_needs_recovery(sb));
5141
5142         if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb))
5143                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
5144                         goto failed_mount3a;
5145
5146         /*
5147          * The first inode we look at is the journal inode.  Don't try
5148          * root first: it may be modified in the journal!
5149          */
5150         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
5151                 err = ext4_load_journal(sb, es, ctx->journal_devnum);
5152                 if (err)
5153                         goto failed_mount3a;
5154         } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
5155                    ext4_has_feature_journal_needs_recovery(sb)) {
5156                 ext4_msg(sb, KERN_ERR, "required journal recovery "
5157                        "suppressed and not mounted read-only");
5158                 goto failed_mount_wq;
5159         } else {
5160                 /* Nojournal mode, all journal mount options are illegal */
5161                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
5162                         ext4_msg(sb, KERN_ERR, "can't mount with "
5163                                  "journal_checksum, fs mounted w/o journal");
5164                         goto failed_mount_wq;
5165                 }
5166                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5167                         ext4_msg(sb, KERN_ERR, "can't mount with "
5168                                  "journal_async_commit, fs mounted w/o journal");
5169                         goto failed_mount_wq;
5170                 }
5171                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
5172                         ext4_msg(sb, KERN_ERR, "can't mount with "
5173                                  "commit=%lu, fs mounted w/o journal",
5174                                  sbi->s_commit_interval / HZ);
5175                         goto failed_mount_wq;
5176                 }
5177                 if (EXT4_MOUNT_DATA_FLAGS &
5178                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
5179                         ext4_msg(sb, KERN_ERR, "can't mount with "
5180                                  "data=, fs mounted w/o journal");
5181                         goto failed_mount_wq;
5182                 }
5183                 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM;
5184                 clear_opt(sb, JOURNAL_CHECKSUM);
5185                 clear_opt(sb, DATA_FLAGS);
5186                 clear_opt2(sb, JOURNAL_FAST_COMMIT);
5187                 sbi->s_journal = NULL;
5188                 needs_recovery = 0;
5189                 goto no_journal;
5190         }
5191
5192         if (ext4_has_feature_64bit(sb) &&
5193             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
5194                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
5195                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
5196                 goto failed_mount_wq;
5197         }
5198
5199         if (!set_journal_csum_feature_set(sb)) {
5200                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
5201                          "feature set");
5202                 goto failed_mount_wq;
5203         }
5204
5205         if (test_opt2(sb, JOURNAL_FAST_COMMIT) &&
5206                 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
5207                                           JBD2_FEATURE_INCOMPAT_FAST_COMMIT)) {
5208                 ext4_msg(sb, KERN_ERR,
5209                         "Failed to set fast commit journal feature");
5210                 goto failed_mount_wq;
5211         }
5212
5213         /* We have now updated the journal if required, so we can
5214          * validate the data journaling mode. */
5215         switch (test_opt(sb, DATA_FLAGS)) {
5216         case 0:
5217                 /* No mode set, assume a default based on the journal
5218                  * capabilities: ORDERED_DATA if the journal can
5219                  * cope, else JOURNAL_DATA
5220                  */
5221                 if (jbd2_journal_check_available_features
5222                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
5223                         set_opt(sb, ORDERED_DATA);
5224                         sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
5225                 } else {
5226                         set_opt(sb, JOURNAL_DATA);
5227                         sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
5228                 }
5229                 break;
5230
5231         case EXT4_MOUNT_ORDERED_DATA:
5232         case EXT4_MOUNT_WRITEBACK_DATA:
5233                 if (!jbd2_journal_check_available_features
5234                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
5235                         ext4_msg(sb, KERN_ERR, "Journal does not support "
5236                                "requested data journaling mode");
5237                         goto failed_mount_wq;
5238                 }
5239                 break;
5240         default:
5241                 break;
5242         }
5243
5244         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
5245             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5246                 ext4_msg(sb, KERN_ERR, "can't mount with "
5247                         "journal_async_commit in data=ordered mode");
5248                 goto failed_mount_wq;
5249         }
5250
5251         set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio);
5252
5253         sbi->s_journal->j_submit_inode_data_buffers =
5254                 ext4_journal_submit_inode_data_buffers;
5255         sbi->s_journal->j_finish_inode_data_buffers =
5256                 ext4_journal_finish_inode_data_buffers;
5257
5258 no_journal:
5259         if (!test_opt(sb, NO_MBCACHE)) {
5260                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
5261                 if (!sbi->s_ea_block_cache) {
5262                         ext4_msg(sb, KERN_ERR,
5263                                  "Failed to create ea_block_cache");
5264                         goto failed_mount_wq;
5265                 }
5266
5267                 if (ext4_has_feature_ea_inode(sb)) {
5268                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
5269                         if (!sbi->s_ea_inode_cache) {
5270                                 ext4_msg(sb, KERN_ERR,
5271                                          "Failed to create ea_inode_cache");
5272                                 goto failed_mount_wq;
5273                         }
5274                 }
5275         }
5276
5277         if (ext4_has_feature_verity(sb) && blocksize != PAGE_SIZE) {
5278                 ext4_msg(sb, KERN_ERR, "Unsupported blocksize for fs-verity");
5279                 goto failed_mount_wq;
5280         }
5281
5282         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) &&
5283             !ext4_has_feature_encrypt(sb)) {
5284                 ext4_set_feature_encrypt(sb);
5285                 ext4_commit_super(sb);
5286         }
5287
5288         /*
5289          * Get the # of file system overhead blocks from the
5290          * superblock if present.
5291          */
5292         sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
5293         /* ignore the precalculated value if it is ridiculous */
5294         if (sbi->s_overhead > ext4_blocks_count(es))
5295                 sbi->s_overhead = 0;
5296         /*
5297          * If the bigalloc feature is not enabled recalculating the
5298          * overhead doesn't take long, so we might as well just redo
5299          * it to make sure we are using the correct value.
5300          */
5301         if (!ext4_has_feature_bigalloc(sb))
5302                 sbi->s_overhead = 0;
5303         if (sbi->s_overhead == 0) {
5304                 err = ext4_calculate_overhead(sb);
5305                 if (err)
5306                         goto failed_mount_wq;
5307         }
5308
5309         /*
5310          * The maximum number of concurrent works can be high and
5311          * concurrency isn't really necessary.  Limit it to 1.
5312          */
5313         EXT4_SB(sb)->rsv_conversion_wq =
5314                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
5315         if (!EXT4_SB(sb)->rsv_conversion_wq) {
5316                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
5317                 ret = -ENOMEM;
5318                 goto failed_mount4;
5319         }
5320
5321         /*
5322          * The jbd2_journal_load will have done any necessary log recovery,
5323          * so we can safely mount the rest of the filesystem now.
5324          */
5325
5326         root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL);
5327         if (IS_ERR(root)) {
5328                 ext4_msg(sb, KERN_ERR, "get root inode failed");
5329                 ret = PTR_ERR(root);
5330                 root = NULL;
5331                 goto failed_mount4;
5332         }
5333         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
5334                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
5335                 iput(root);
5336                 goto failed_mount4;
5337         }
5338
5339         sb->s_root = d_make_root(root);
5340         if (!sb->s_root) {
5341                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
5342                 ret = -ENOMEM;
5343                 goto failed_mount4;
5344         }
5345
5346         ret = ext4_setup_super(sb, es, sb_rdonly(sb));
5347         if (ret == -EROFS) {
5348                 sb->s_flags |= SB_RDONLY;
5349                 ret = 0;
5350         } else if (ret)
5351                 goto failed_mount4a;
5352
5353         ext4_set_resv_clusters(sb);
5354
5355         if (test_opt(sb, BLOCK_VALIDITY)) {
5356                 err = ext4_setup_system_zone(sb);
5357                 if (err) {
5358                         ext4_msg(sb, KERN_ERR, "failed to initialize system "
5359                                  "zone (%d)", err);
5360                         goto failed_mount4a;
5361                 }
5362         }
5363         ext4_fc_replay_cleanup(sb);
5364
5365         ext4_ext_init(sb);
5366
5367         /*
5368          * Enable optimize_scan if number of groups is > threshold. This can be
5369          * turned off by passing "mb_optimize_scan=0". This can also be
5370          * turned on forcefully by passing "mb_optimize_scan=1".
5371          */
5372         if (!(ctx->spec & EXT4_SPEC_mb_optimize_scan)) {
5373                 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD)
5374                         set_opt2(sb, MB_OPTIMIZE_SCAN);
5375                 else
5376                         clear_opt2(sb, MB_OPTIMIZE_SCAN);
5377         }
5378
5379         err = ext4_mb_init(sb);
5380         if (err) {
5381                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
5382                          err);
5383                 goto failed_mount5;
5384         }
5385
5386         /*
5387          * We can only set up the journal commit callback once
5388          * mballoc is initialized
5389          */
5390         if (sbi->s_journal)
5391                 sbi->s_journal->j_commit_callback =
5392                         ext4_journal_commit_callback;
5393
5394         block = ext4_count_free_clusters(sb);
5395         ext4_free_blocks_count_set(sbi->s_es,
5396                                    EXT4_C2B(sbi, block));
5397         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
5398                                   GFP_KERNEL);
5399         if (!err) {
5400                 unsigned long freei = ext4_count_free_inodes(sb);
5401                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
5402                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
5403                                           GFP_KERNEL);
5404         }
5405         /*
5406          * Update the checksum after updating free space/inode
5407          * counters.  Otherwise the superblock can have an incorrect
5408          * checksum in the buffer cache until it is written out and
5409          * e2fsprogs programs trying to open a file system immediately
5410          * after it is mounted can fail.
5411          */
5412         ext4_superblock_csum_set(sb);
5413         if (!err)
5414                 err = percpu_counter_init(&sbi->s_dirs_counter,
5415                                           ext4_count_dirs(sb), GFP_KERNEL);
5416         if (!err)
5417                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
5418                                           GFP_KERNEL);
5419         if (!err)
5420                 err = percpu_counter_init(&sbi->s_sra_exceeded_retry_limit, 0,
5421                                           GFP_KERNEL);
5422         if (!err)
5423                 err = percpu_init_rwsem(&sbi->s_writepages_rwsem);
5424
5425         if (err) {
5426                 ext4_msg(sb, KERN_ERR, "insufficient memory");
5427                 goto failed_mount6;
5428         }
5429
5430         if (ext4_has_feature_flex_bg(sb))
5431                 if (!ext4_fill_flex_info(sb)) {
5432                         ext4_msg(sb, KERN_ERR,
5433                                "unable to initialize "
5434                                "flex_bg meta info!");
5435                         ret = -ENOMEM;
5436                         goto failed_mount6;
5437                 }
5438
5439         err = ext4_register_li_request(sb, first_not_zeroed);
5440         if (err)
5441                 goto failed_mount6;
5442
5443         err = ext4_register_sysfs(sb);
5444         if (err)
5445                 goto failed_mount7;
5446
5447         err = ext4_init_orphan_info(sb);
5448         if (err)
5449                 goto failed_mount8;
5450 #ifdef CONFIG_QUOTA
5451         /* Enable quota usage during mount. */
5452         if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
5453                 err = ext4_enable_quotas(sb);
5454                 if (err)
5455                         goto failed_mount9;
5456         }
5457 #endif  /* CONFIG_QUOTA */
5458
5459         /*
5460          * Save the original bdev mapping's wb_err value which could be
5461          * used to detect the metadata async write error.
5462          */
5463         spin_lock_init(&sbi->s_bdev_wb_lock);
5464         errseq_check_and_advance(&sb->s_bdev->bd_inode->i_mapping->wb_err,
5465                                  &sbi->s_bdev_wb_err);
5466         sb->s_bdev->bd_super = sb;
5467         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
5468         ext4_orphan_cleanup(sb, es);
5469         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
5470         if (needs_recovery) {
5471                 ext4_msg(sb, KERN_INFO, "recovery complete");
5472                 err = ext4_mark_recovery_complete(sb, es);
5473                 if (err)
5474                         goto failed_mount9;
5475         }
5476
5477         if (test_opt(sb, DISCARD)) {
5478                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
5479                 if (!blk_queue_discard(q))
5480                         ext4_msg(sb, KERN_WARNING,
5481                                  "mounting with \"discard\" option, but "
5482                                  "the device does not support discard");
5483         }
5484
5485         if (es->s_error_count)
5486                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
5487
5488         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
5489         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
5490         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
5491         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
5492         atomic_set(&sbi->s_warning_count, 0);
5493         atomic_set(&sbi->s_msg_count, 0);
5494
5495         return 0;
5496
5497 cantfind_ext4:
5498         if (!silent)
5499                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
5500         goto failed_mount;
5501
5502 failed_mount9:
5503         ext4_release_orphan_info(sb);
5504 failed_mount8:
5505         ext4_unregister_sysfs(sb);
5506         kobject_put(&sbi->s_kobj);
5507 failed_mount7:
5508         ext4_unregister_li_request(sb);
5509 failed_mount6:
5510         ext4_mb_release(sb);
5511         rcu_read_lock();
5512         flex_groups = rcu_dereference(sbi->s_flex_groups);
5513         if (flex_groups) {
5514                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
5515                         kvfree(flex_groups[i]);
5516                 kvfree(flex_groups);
5517         }
5518         rcu_read_unlock();
5519         percpu_counter_destroy(&sbi->s_freeclusters_counter);
5520         percpu_counter_destroy(&sbi->s_freeinodes_counter);
5521         percpu_counter_destroy(&sbi->s_dirs_counter);
5522         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
5523         percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit);
5524         percpu_free_rwsem(&sbi->s_writepages_rwsem);
5525 failed_mount5:
5526         ext4_ext_release(sb);
5527         ext4_release_system_zone(sb);
5528 failed_mount4a:
5529         dput(sb->s_root);
5530         sb->s_root = NULL;
5531 failed_mount4:
5532         ext4_msg(sb, KERN_ERR, "mount failed");
5533         if (EXT4_SB(sb)->rsv_conversion_wq)
5534                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
5535 failed_mount_wq:
5536         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
5537         sbi->s_ea_inode_cache = NULL;
5538
5539         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
5540         sbi->s_ea_block_cache = NULL;
5541
5542         if (sbi->s_journal) {
5543                 /* flush s_error_work before journal destroy. */
5544                 flush_work(&sbi->s_error_work);
5545                 jbd2_journal_destroy(sbi->s_journal);
5546                 sbi->s_journal = NULL;
5547         }
5548 failed_mount3a:
5549         ext4_es_unregister_shrinker(sbi);
5550 failed_mount3:
5551         /* flush s_error_work before sbi destroy */
5552         flush_work(&sbi->s_error_work);
5553         del_timer_sync(&sbi->s_err_report);
5554         ext4_stop_mmpd(sbi);
5555 failed_mount2:
5556         rcu_read_lock();
5557         group_desc = rcu_dereference(sbi->s_group_desc);
5558         for (i = 0; i < db_count; i++)
5559                 brelse(group_desc[i]);
5560         kvfree(group_desc);
5561         rcu_read_unlock();
5562 failed_mount:
5563         if (sbi->s_chksum_driver)
5564                 crypto_free_shash(sbi->s_chksum_driver);
5565
5566 #if IS_ENABLED(CONFIG_UNICODE)
5567         utf8_unload(sb->s_encoding);
5568 #endif
5569
5570 #ifdef CONFIG_QUOTA
5571         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5572                 kfree(get_qf_name(sb, sbi, i));
5573 #endif
5574         fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy);
5575         /* ext4_blkdev_remove() calls kill_bdev(), release bh before it. */
5576         brelse(bh);
5577         ext4_blkdev_remove(sbi);
5578 out_fail:
5579         sb->s_fs_info = NULL;
5580         return err ? err : ret;
5581 }
5582
5583 static int ext4_fill_super(struct super_block *sb, struct fs_context *fc)
5584 {
5585         struct ext4_fs_context *ctx = fc->fs_private;
5586         struct ext4_sb_info *sbi;
5587         const char *descr;
5588         int ret;
5589
5590         sbi = ext4_alloc_sbi(sb);
5591         if (!sbi)
5592                 return -ENOMEM;
5593
5594         fc->s_fs_info = sbi;
5595
5596         /* Cleanup superblock name */
5597         strreplace(sb->s_id, '/', '!');
5598
5599         sbi->s_sb_block = 1;    /* Default super block location */
5600         if (ctx->spec & EXT4_SPEC_s_sb_block)
5601                 sbi->s_sb_block = ctx->s_sb_block;
5602
5603         ret = __ext4_fill_super(fc, sb);
5604         if (ret < 0)
5605                 goto free_sbi;
5606
5607         if (sbi->s_journal) {
5608                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
5609                         descr = " journalled data mode";
5610                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
5611                         descr = " ordered data mode";
5612                 else
5613                         descr = " writeback data mode";
5614         } else
5615                 descr = "out journal";
5616
5617         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
5618                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
5619                          "Quota mode: %s.", descr, ext4_quota_mode(sb));
5620
5621         /* Update the s_overhead_clusters if necessary */
5622         ext4_update_overhead(sb);
5623         return 0;
5624
5625 free_sbi:
5626         ext4_free_sbi(sbi);
5627         fc->s_fs_info = NULL;
5628         return ret;
5629 }
5630
5631 static int ext4_get_tree(struct fs_context *fc)
5632 {
5633         return get_tree_bdev(fc, ext4_fill_super);
5634 }
5635
5636 /*
5637  * Setup any per-fs journal parameters now.  We'll do this both on
5638  * initial mount, once the journal has been initialised but before we've
5639  * done any recovery; and again on any subsequent remount.
5640  */
5641 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
5642 {
5643         struct ext4_sb_info *sbi = EXT4_SB(sb);
5644
5645         journal->j_commit_interval = sbi->s_commit_interval;
5646         journal->j_min_batch_time = sbi->s_min_batch_time;
5647         journal->j_max_batch_time = sbi->s_max_batch_time;
5648         ext4_fc_init(sb, journal);
5649
5650         write_lock(&journal->j_state_lock);
5651         if (test_opt(sb, BARRIER))
5652                 journal->j_flags |= JBD2_BARRIER;
5653         else
5654                 journal->j_flags &= ~JBD2_BARRIER;
5655         if (test_opt(sb, DATA_ERR_ABORT))
5656                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
5657         else
5658                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
5659         write_unlock(&journal->j_state_lock);
5660 }
5661
5662 static struct inode *ext4_get_journal_inode(struct super_block *sb,
5663                                              unsigned int journal_inum)
5664 {
5665         struct inode *journal_inode;
5666
5667         /*
5668          * Test for the existence of a valid inode on disk.  Bad things
5669          * happen if we iget() an unused inode, as the subsequent iput()
5670          * will try to delete it.
5671          */
5672         journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL);
5673         if (IS_ERR(journal_inode)) {
5674                 ext4_msg(sb, KERN_ERR, "no journal found");
5675                 return NULL;
5676         }
5677         if (!journal_inode->i_nlink) {
5678                 make_bad_inode(journal_inode);
5679                 iput(journal_inode);
5680                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
5681                 return NULL;
5682         }
5683
5684         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
5685                   journal_inode, journal_inode->i_size);
5686         if (!S_ISREG(journal_inode->i_mode)) {
5687                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
5688                 iput(journal_inode);
5689                 return NULL;
5690         }
5691         return journal_inode;
5692 }
5693
5694 static journal_t *ext4_get_journal(struct super_block *sb,
5695                                    unsigned int journal_inum)
5696 {
5697         struct inode *journal_inode;
5698         journal_t *journal;
5699
5700         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5701                 return NULL;
5702
5703         journal_inode = ext4_get_journal_inode(sb, journal_inum);
5704         if (!journal_inode)
5705                 return NULL;
5706
5707         journal = jbd2_journal_init_inode(journal_inode);
5708         if (!journal) {
5709                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
5710                 iput(journal_inode);
5711                 return NULL;
5712         }
5713         journal->j_private = sb;
5714         ext4_init_journal_params(sb, journal);
5715         return journal;
5716 }
5717
5718 static journal_t *ext4_get_dev_journal(struct super_block *sb,
5719                                        dev_t j_dev)
5720 {
5721         struct buffer_head *bh;
5722         journal_t *journal;
5723         ext4_fsblk_t start;
5724         ext4_fsblk_t len;
5725         int hblock, blocksize;
5726         ext4_fsblk_t sb_block;
5727         unsigned long offset;
5728         struct ext4_super_block *es;
5729         struct block_device *bdev;
5730
5731         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5732                 return NULL;
5733
5734         bdev = ext4_blkdev_get(j_dev, sb);
5735         if (bdev == NULL)
5736                 return NULL;
5737
5738         blocksize = sb->s_blocksize;
5739         hblock = bdev_logical_block_size(bdev);
5740         if (blocksize < hblock) {
5741                 ext4_msg(sb, KERN_ERR,
5742                         "blocksize too small for journal device");
5743                 goto out_bdev;
5744         }
5745
5746         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
5747         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
5748         set_blocksize(bdev, blocksize);
5749         if (!(bh = __bread(bdev, sb_block, blocksize))) {
5750                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
5751                        "external journal");
5752                 goto out_bdev;
5753         }
5754
5755         es = (struct ext4_super_block *) (bh->b_data + offset);
5756         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
5757             !(le32_to_cpu(es->s_feature_incompat) &
5758               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
5759                 ext4_msg(sb, KERN_ERR, "external journal has "
5760                                         "bad superblock");
5761                 brelse(bh);
5762                 goto out_bdev;
5763         }
5764
5765         if ((le32_to_cpu(es->s_feature_ro_compat) &
5766              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
5767             es->s_checksum != ext4_superblock_csum(sb, es)) {
5768                 ext4_msg(sb, KERN_ERR, "external journal has "
5769                                        "corrupt superblock");
5770                 brelse(bh);
5771                 goto out_bdev;
5772         }
5773
5774         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
5775                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
5776                 brelse(bh);
5777                 goto out_bdev;
5778         }
5779
5780         len = ext4_blocks_count(es);
5781         start = sb_block + 1;
5782         brelse(bh);     /* we're done with the superblock */
5783
5784         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
5785                                         start, len, blocksize);
5786         if (!journal) {
5787                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
5788                 goto out_bdev;
5789         }
5790         journal->j_private = sb;
5791         if (ext4_read_bh_lock(journal->j_sb_buffer, REQ_META | REQ_PRIO, true)) {
5792                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
5793                 goto out_journal;
5794         }
5795         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
5796                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
5797                                         "user (unsupported) - %d",
5798                         be32_to_cpu(journal->j_superblock->s_nr_users));
5799                 goto out_journal;
5800         }
5801         EXT4_SB(sb)->s_journal_bdev = bdev;
5802         ext4_init_journal_params(sb, journal);
5803         return journal;
5804
5805 out_journal:
5806         jbd2_journal_destroy(journal);
5807 out_bdev:
5808         ext4_blkdev_put(bdev);
5809         return NULL;
5810 }
5811
5812 static int ext4_load_journal(struct super_block *sb,
5813                              struct ext4_super_block *es,
5814                              unsigned long journal_devnum)
5815 {
5816         journal_t *journal;
5817         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
5818         dev_t journal_dev;
5819         int err = 0;
5820         int really_read_only;
5821         int journal_dev_ro;
5822
5823         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5824                 return -EFSCORRUPTED;
5825
5826         if (journal_devnum &&
5827             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5828                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
5829                         "numbers have changed");
5830                 journal_dev = new_decode_dev(journal_devnum);
5831         } else
5832                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
5833
5834         if (journal_inum && journal_dev) {
5835                 ext4_msg(sb, KERN_ERR,
5836                          "filesystem has both journal inode and journal device!");
5837                 return -EINVAL;
5838         }
5839
5840         if (journal_inum) {
5841                 journal = ext4_get_journal(sb, journal_inum);
5842                 if (!journal)
5843                         return -EINVAL;
5844         } else {
5845                 journal = ext4_get_dev_journal(sb, journal_dev);
5846                 if (!journal)
5847                         return -EINVAL;
5848         }
5849
5850         journal_dev_ro = bdev_read_only(journal->j_dev);
5851         really_read_only = bdev_read_only(sb->s_bdev) | journal_dev_ro;
5852
5853         if (journal_dev_ro && !sb_rdonly(sb)) {
5854                 ext4_msg(sb, KERN_ERR,
5855                          "journal device read-only, try mounting with '-o ro'");
5856                 err = -EROFS;
5857                 goto err_out;
5858         }
5859
5860         /*
5861          * Are we loading a blank journal or performing recovery after a
5862          * crash?  For recovery, we need to check in advance whether we
5863          * can get read-write access to the device.
5864          */
5865         if (ext4_has_feature_journal_needs_recovery(sb)) {
5866                 if (sb_rdonly(sb)) {
5867                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
5868                                         "required on readonly filesystem");
5869                         if (really_read_only) {
5870                                 ext4_msg(sb, KERN_ERR, "write access "
5871                                         "unavailable, cannot proceed "
5872                                         "(try mounting with noload)");
5873                                 err = -EROFS;
5874                                 goto err_out;
5875                         }
5876                         ext4_msg(sb, KERN_INFO, "write access will "
5877                                "be enabled during recovery");
5878                 }
5879         }
5880
5881         if (!(journal->j_flags & JBD2_BARRIER))
5882                 ext4_msg(sb, KERN_INFO, "barriers disabled");
5883
5884         if (!ext4_has_feature_journal_needs_recovery(sb))
5885                 err = jbd2_journal_wipe(journal, !really_read_only);
5886         if (!err) {
5887                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
5888                 if (save)
5889                         memcpy(save, ((char *) es) +
5890                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
5891                 err = jbd2_journal_load(journal);
5892                 if (save)
5893                         memcpy(((char *) es) + EXT4_S_ERR_START,
5894                                save, EXT4_S_ERR_LEN);
5895                 kfree(save);
5896         }
5897
5898         if (err) {
5899                 ext4_msg(sb, KERN_ERR, "error loading journal");
5900                 goto err_out;
5901         }
5902
5903         EXT4_SB(sb)->s_journal = journal;
5904         err = ext4_clear_journal_err(sb, es);
5905         if (err) {
5906                 EXT4_SB(sb)->s_journal = NULL;
5907                 jbd2_journal_destroy(journal);
5908                 return err;
5909         }
5910
5911         if (!really_read_only && journal_devnum &&
5912             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5913                 es->s_journal_dev = cpu_to_le32(journal_devnum);
5914
5915                 /* Make sure we flush the recovery flag to disk. */
5916                 ext4_commit_super(sb);
5917         }
5918
5919         return 0;
5920
5921 err_out:
5922         jbd2_journal_destroy(journal);
5923         return err;
5924 }
5925
5926 /* Copy state of EXT4_SB(sb) into buffer for on-disk superblock */
5927 static void ext4_update_super(struct super_block *sb)
5928 {
5929         struct ext4_sb_info *sbi = EXT4_SB(sb);
5930         struct ext4_super_block *es = sbi->s_es;
5931         struct buffer_head *sbh = sbi->s_sbh;
5932
5933         lock_buffer(sbh);
5934         /*
5935          * If the file system is mounted read-only, don't update the
5936          * superblock write time.  This avoids updating the superblock
5937          * write time when we are mounting the root file system
5938          * read/only but we need to replay the journal; at that point,
5939          * for people who are east of GMT and who make their clock
5940          * tick in localtime for Windows bug-for-bug compatibility,
5941          * the clock is set in the future, and this will cause e2fsck
5942          * to complain and force a full file system check.
5943          */
5944         if (!(sb->s_flags & SB_RDONLY))
5945                 ext4_update_tstamp(es, s_wtime);
5946         es->s_kbytes_written =
5947                 cpu_to_le64(sbi->s_kbytes_written +
5948                     ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) -
5949                       sbi->s_sectors_written_start) >> 1));
5950         if (percpu_counter_initialized(&sbi->s_freeclusters_counter))
5951                 ext4_free_blocks_count_set(es,
5952                         EXT4_C2B(sbi, percpu_counter_sum_positive(
5953                                 &sbi->s_freeclusters_counter)));
5954         if (percpu_counter_initialized(&sbi->s_freeinodes_counter))
5955                 es->s_free_inodes_count =
5956                         cpu_to_le32(percpu_counter_sum_positive(
5957                                 &sbi->s_freeinodes_counter));
5958         /* Copy error information to the on-disk superblock */
5959         spin_lock(&sbi->s_error_lock);
5960         if (sbi->s_add_error_count > 0) {
5961                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
5962                 if (!es->s_first_error_time && !es->s_first_error_time_hi) {
5963                         __ext4_update_tstamp(&es->s_first_error_time,
5964                                              &es->s_first_error_time_hi,
5965                                              sbi->s_first_error_time);
5966                         strncpy(es->s_first_error_func, sbi->s_first_error_func,
5967                                 sizeof(es->s_first_error_func));
5968                         es->s_first_error_line =
5969                                 cpu_to_le32(sbi->s_first_error_line);
5970                         es->s_first_error_ino =
5971                                 cpu_to_le32(sbi->s_first_error_ino);
5972                         es->s_first_error_block =
5973                                 cpu_to_le64(sbi->s_first_error_block);
5974                         es->s_first_error_errcode =
5975                                 ext4_errno_to_code(sbi->s_first_error_code);
5976                 }
5977                 __ext4_update_tstamp(&es->s_last_error_time,
5978                                      &es->s_last_error_time_hi,
5979                                      sbi->s_last_error_time);
5980                 strncpy(es->s_last_error_func, sbi->s_last_error_func,
5981                         sizeof(es->s_last_error_func));
5982                 es->s_last_error_line = cpu_to_le32(sbi->s_last_error_line);
5983                 es->s_last_error_ino = cpu_to_le32(sbi->s_last_error_ino);
5984                 es->s_last_error_block = cpu_to_le64(sbi->s_last_error_block);
5985                 es->s_last_error_errcode =
5986                                 ext4_errno_to_code(sbi->s_last_error_code);
5987                 /*
5988                  * Start the daily error reporting function if it hasn't been
5989                  * started already
5990                  */
5991                 if (!es->s_error_count)
5992                         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);
5993                 le32_add_cpu(&es->s_error_count, sbi->s_add_error_count);
5994                 sbi->s_add_error_count = 0;
5995         }
5996         spin_unlock(&sbi->s_error_lock);
5997
5998         ext4_superblock_csum_set(sb);
5999         unlock_buffer(sbh);
6000 }
6001
6002 static int ext4_commit_super(struct super_block *sb)
6003 {
6004         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
6005         int error = 0;
6006
6007         if (!sbh)
6008                 return -EINVAL;
6009         if (block_device_ejected(sb))
6010                 return -ENODEV;
6011
6012         ext4_update_super(sb);
6013
6014         if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
6015                 /*
6016                  * Oh, dear.  A previous attempt to write the
6017                  * superblock failed.  This could happen because the
6018                  * USB device was yanked out.  Or it could happen to
6019                  * be a transient write error and maybe the block will
6020                  * be remapped.  Nothing we can do but to retry the
6021                  * write and hope for the best.
6022                  */
6023                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
6024                        "superblock detected");
6025                 clear_buffer_write_io_error(sbh);
6026                 set_buffer_uptodate(sbh);
6027         }
6028         BUFFER_TRACE(sbh, "marking dirty");
6029         mark_buffer_dirty(sbh);
6030         error = __sync_dirty_buffer(sbh,
6031                 REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
6032         if (buffer_write_io_error(sbh)) {
6033                 ext4_msg(sb, KERN_ERR, "I/O error while writing "
6034                        "superblock");
6035                 clear_buffer_write_io_error(sbh);
6036                 set_buffer_uptodate(sbh);
6037         }
6038         return error;
6039 }
6040
6041 /*
6042  * Have we just finished recovery?  If so, and if we are mounting (or
6043  * remounting) the filesystem readonly, then we will end up with a
6044  * consistent fs on disk.  Record that fact.
6045  */
6046 static int ext4_mark_recovery_complete(struct super_block *sb,
6047                                        struct ext4_super_block *es)
6048 {
6049         int err;
6050         journal_t *journal = EXT4_SB(sb)->s_journal;
6051
6052         if (!ext4_has_feature_journal(sb)) {
6053                 if (journal != NULL) {
6054                         ext4_error(sb, "Journal got removed while the fs was "
6055                                    "mounted!");
6056                         return -EFSCORRUPTED;
6057                 }
6058                 return 0;
6059         }
6060         jbd2_journal_lock_updates(journal);
6061         err = jbd2_journal_flush(journal, 0);
6062         if (err < 0)
6063                 goto out;
6064
6065         if (sb_rdonly(sb) && (ext4_has_feature_journal_needs_recovery(sb) ||
6066             ext4_has_feature_orphan_present(sb))) {
6067                 if (!ext4_orphan_file_empty(sb)) {
6068                         ext4_error(sb, "Orphan file not empty on read-only fs.");
6069                         err = -EFSCORRUPTED;
6070                         goto out;
6071                 }
6072                 ext4_clear_feature_journal_needs_recovery(sb);
6073                 ext4_clear_feature_orphan_present(sb);
6074                 ext4_commit_super(sb);
6075         }
6076 out:
6077         jbd2_journal_unlock_updates(journal);
6078         return err;
6079 }
6080
6081 /*
6082  * If we are mounting (or read-write remounting) a filesystem whose journal
6083  * has recorded an error from a previous lifetime, move that error to the
6084  * main filesystem now.
6085  */
6086 static int ext4_clear_journal_err(struct super_block *sb,
6087                                    struct ext4_super_block *es)
6088 {
6089         journal_t *journal;
6090         int j_errno;
6091         const char *errstr;
6092
6093         if (!ext4_has_feature_journal(sb)) {
6094                 ext4_error(sb, "Journal got removed while the fs was mounted!");
6095                 return -EFSCORRUPTED;
6096         }
6097
6098         journal = EXT4_SB(sb)->s_journal;
6099
6100         /*
6101          * Now check for any error status which may have been recorded in the
6102          * journal by a prior ext4_error() or ext4_abort()
6103          */
6104
6105         j_errno = jbd2_journal_errno(journal);
6106         if (j_errno) {
6107                 char nbuf[16];
6108
6109                 errstr = ext4_decode_error(sb, j_errno, nbuf);
6110                 ext4_warning(sb, "Filesystem error recorded "
6111                              "from previous mount: %s", errstr);
6112                 ext4_warning(sb, "Marking fs in need of filesystem check.");
6113
6114                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
6115                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
6116                 ext4_commit_super(sb);
6117
6118                 jbd2_journal_clear_err(journal);
6119                 jbd2_journal_update_sb_errno(journal);
6120         }
6121         return 0;
6122 }
6123
6124 /*
6125  * Force the running and committing transactions to commit,
6126  * and wait on the commit.
6127  */
6128 int ext4_force_commit(struct super_block *sb)
6129 {
6130         journal_t *journal;
6131
6132         if (sb_rdonly(sb))
6133                 return 0;
6134
6135         journal = EXT4_SB(sb)->s_journal;
6136         return ext4_journal_force_commit(journal);
6137 }
6138
6139 static int ext4_sync_fs(struct super_block *sb, int wait)
6140 {
6141         int ret = 0;
6142         tid_t target;
6143         bool needs_barrier = false;
6144         struct ext4_sb_info *sbi = EXT4_SB(sb);
6145
6146         if (unlikely(ext4_forced_shutdown(sbi)))
6147                 return 0;
6148
6149         trace_ext4_sync_fs(sb, wait);
6150         flush_workqueue(sbi->rsv_conversion_wq);
6151         /*
6152          * Writeback quota in non-journalled quota case - journalled quota has
6153          * no dirty dquots
6154          */
6155         dquot_writeback_dquots(sb, -1);
6156         /*
6157          * Data writeback is possible w/o journal transaction, so barrier must
6158          * being sent at the end of the function. But we can skip it if
6159          * transaction_commit will do it for us.
6160          */
6161         if (sbi->s_journal) {
6162                 target = jbd2_get_latest_transaction(sbi->s_journal);
6163                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
6164                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
6165                         needs_barrier = true;
6166
6167                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
6168                         if (wait)
6169                                 ret = jbd2_log_wait_commit(sbi->s_journal,
6170                                                            target);
6171                 }
6172         } else if (wait && test_opt(sb, BARRIER))
6173                 needs_barrier = true;
6174         if (needs_barrier) {
6175                 int err;
6176                 err = blkdev_issue_flush(sb->s_bdev);
6177                 if (!ret)
6178                         ret = err;
6179         }
6180
6181         return ret;
6182 }
6183
6184 /*
6185  * LVM calls this function before a (read-only) snapshot is created.  This
6186  * gives us a chance to flush the journal completely and mark the fs clean.
6187  *
6188  * Note that only this function cannot bring a filesystem to be in a clean
6189  * state independently. It relies on upper layer to stop all data & metadata
6190  * modifications.
6191  */
6192 static int ext4_freeze(struct super_block *sb)
6193 {
6194         int error = 0;
6195         journal_t *journal;
6196
6197         if (sb_rdonly(sb))
6198                 return 0;
6199
6200         journal = EXT4_SB(sb)->s_journal;
6201
6202         if (journal) {
6203                 /* Now we set up the journal barrier. */
6204                 jbd2_journal_lock_updates(journal);
6205
6206                 /*
6207                  * Don't clear the needs_recovery flag if we failed to
6208                  * flush the journal.
6209                  */
6210                 error = jbd2_journal_flush(journal, 0);
6211                 if (error < 0)
6212                         goto out;
6213
6214                 /* Journal blocked and flushed, clear needs_recovery flag. */
6215                 ext4_clear_feature_journal_needs_recovery(sb);
6216                 if (ext4_orphan_file_empty(sb))
6217                         ext4_clear_feature_orphan_present(sb);
6218         }
6219
6220         error = ext4_commit_super(sb);
6221 out:
6222         if (journal)
6223                 /* we rely on upper layer to stop further updates */
6224                 jbd2_journal_unlock_updates(journal);
6225         return error;
6226 }
6227
6228 /*
6229  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
6230  * flag here, even though the filesystem is not technically dirty yet.
6231  */
6232 static int ext4_unfreeze(struct super_block *sb)
6233 {
6234         if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
6235                 return 0;
6236
6237         if (EXT4_SB(sb)->s_journal) {
6238                 /* Reset the needs_recovery flag before the fs is unlocked. */
6239                 ext4_set_feature_journal_needs_recovery(sb);
6240                 if (ext4_has_feature_orphan_file(sb))
6241                         ext4_set_feature_orphan_present(sb);
6242         }
6243
6244         ext4_commit_super(sb);
6245         return 0;
6246 }
6247
6248 /*
6249  * Structure to save mount options for ext4_remount's benefit
6250  */
6251 struct ext4_mount_options {
6252         unsigned long s_mount_opt;
6253         unsigned long s_mount_opt2;
6254         kuid_t s_resuid;
6255         kgid_t s_resgid;
6256         unsigned long s_commit_interval;
6257         u32 s_min_batch_time, s_max_batch_time;
6258 #ifdef CONFIG_QUOTA
6259         int s_jquota_fmt;
6260         char *s_qf_names[EXT4_MAXQUOTAS];
6261 #endif
6262 };
6263
6264 static int __ext4_remount(struct fs_context *fc, struct super_block *sb)
6265 {
6266         struct ext4_fs_context *ctx = fc->fs_private;
6267         struct ext4_super_block *es;
6268         struct ext4_sb_info *sbi = EXT4_SB(sb);
6269         unsigned long old_sb_flags;
6270         struct ext4_mount_options old_opts;
6271         ext4_group_t g;
6272         int err = 0;
6273 #ifdef CONFIG_QUOTA
6274         int enable_quota = 0;
6275         int i, j;
6276         char *to_free[EXT4_MAXQUOTAS];
6277 #endif
6278
6279         ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
6280
6281         /* Store the original options */
6282         old_sb_flags = sb->s_flags;
6283         old_opts.s_mount_opt = sbi->s_mount_opt;
6284         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
6285         old_opts.s_resuid = sbi->s_resuid;
6286         old_opts.s_resgid = sbi->s_resgid;
6287         old_opts.s_commit_interval = sbi->s_commit_interval;
6288         old_opts.s_min_batch_time = sbi->s_min_batch_time;
6289         old_opts.s_max_batch_time = sbi->s_max_batch_time;
6290 #ifdef CONFIG_QUOTA
6291         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
6292         for (i = 0; i < EXT4_MAXQUOTAS; i++)
6293                 if (sbi->s_qf_names[i]) {
6294                         char *qf_name = get_qf_name(sb, sbi, i);
6295
6296                         old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL);
6297                         if (!old_opts.s_qf_names[i]) {
6298                                 for (j = 0; j < i; j++)
6299                                         kfree(old_opts.s_qf_names[j]);
6300                                 return -ENOMEM;
6301                         }
6302                 } else
6303                         old_opts.s_qf_names[i] = NULL;
6304 #endif
6305         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
6306                 ctx->journal_ioprio =
6307                         sbi->s_journal->j_task->io_context->ioprio;
6308
6309         ext4_apply_options(fc, sb);
6310
6311         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
6312             test_opt(sb, JOURNAL_CHECKSUM)) {
6313                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
6314                          "during remount not supported; ignoring");
6315                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
6316         }
6317
6318         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
6319                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
6320                         ext4_msg(sb, KERN_ERR, "can't mount with "
6321                                  "both data=journal and delalloc");
6322                         err = -EINVAL;
6323                         goto restore_opts;
6324                 }
6325                 if (test_opt(sb, DIOREAD_NOLOCK)) {
6326                         ext4_msg(sb, KERN_ERR, "can't mount with "
6327                                  "both data=journal and dioread_nolock");
6328                         err = -EINVAL;
6329                         goto restore_opts;
6330                 }
6331         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
6332                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
6333                         ext4_msg(sb, KERN_ERR, "can't mount with "
6334                                 "journal_async_commit in data=ordered mode");
6335                         err = -EINVAL;
6336                         goto restore_opts;
6337                 }
6338         }
6339
6340         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
6341                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
6342                 err = -EINVAL;
6343                 goto restore_opts;
6344         }
6345
6346         if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED))
6347                 ext4_abort(sb, ESHUTDOWN, "Abort forced by user");
6348
6349         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
6350                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
6351
6352         es = sbi->s_es;
6353
6354         if (sbi->s_journal) {
6355                 ext4_init_journal_params(sb, sbi->s_journal);
6356                 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio);
6357         }
6358
6359         /* Flush outstanding errors before changing fs state */
6360         flush_work(&sbi->s_error_work);
6361
6362         if ((bool)(fc->sb_flags & SB_RDONLY) != sb_rdonly(sb)) {
6363                 if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED)) {
6364                         err = -EROFS;
6365                         goto restore_opts;
6366                 }
6367
6368                 if (fc->sb_flags & SB_RDONLY) {
6369                         err = sync_filesystem(sb);
6370                         if (err < 0)
6371                                 goto restore_opts;
6372                         err = dquot_suspend(sb, -1);
6373                         if (err < 0)
6374                                 goto restore_opts;
6375
6376                         /*
6377                          * First of all, the unconditional stuff we have to do
6378                          * to disable replay of the journal when we next remount
6379                          */
6380                         sb->s_flags |= SB_RDONLY;
6381
6382                         /*
6383                          * OK, test if we are remounting a valid rw partition
6384                          * readonly, and if so set the rdonly flag and then
6385                          * mark the partition as valid again.
6386                          */
6387                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
6388                             (sbi->s_mount_state & EXT4_VALID_FS))
6389                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
6390
6391                         if (sbi->s_journal) {
6392                                 /*
6393                                  * We let remount-ro finish even if marking fs
6394                                  * as clean failed...
6395                                  */
6396                                 ext4_mark_recovery_complete(sb, es);
6397                         }
6398                 } else {
6399                         /* Make sure we can mount this feature set readwrite */
6400                         if (ext4_has_feature_readonly(sb) ||
6401                             !ext4_feature_set_ok(sb, 0)) {
6402                                 err = -EROFS;
6403                                 goto restore_opts;
6404                         }
6405                         /*
6406                          * Make sure the group descriptor checksums
6407                          * are sane.  If they aren't, refuse to remount r/w.
6408                          */
6409                         for (g = 0; g < sbi->s_groups_count; g++) {
6410                                 struct ext4_group_desc *gdp =
6411                                         ext4_get_group_desc(sb, g, NULL);
6412
6413                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
6414                                         ext4_msg(sb, KERN_ERR,
6415                "ext4_remount: Checksum for group %u failed (%u!=%u)",
6416                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
6417                                                le16_to_cpu(gdp->bg_checksum));
6418                                         err = -EFSBADCRC;
6419                                         goto restore_opts;
6420                                 }
6421                         }
6422
6423                         /*
6424                          * If we have an unprocessed orphan list hanging
6425                          * around from a previously readonly bdev mount,
6426                          * require a full umount/remount for now.
6427                          */
6428                         if (es->s_last_orphan || !ext4_orphan_file_empty(sb)) {
6429                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
6430                                        "remount RDWR because of unprocessed "
6431                                        "orphan inode list.  Please "
6432                                        "umount/remount instead");
6433                                 err = -EINVAL;
6434                                 goto restore_opts;
6435                         }
6436
6437                         /*
6438                          * Mounting a RDONLY partition read-write, so reread
6439                          * and store the current valid flag.  (It may have
6440                          * been changed by e2fsck since we originally mounted
6441                          * the partition.)
6442                          */
6443                         if (sbi->s_journal) {
6444                                 err = ext4_clear_journal_err(sb, es);
6445                                 if (err)
6446                                         goto restore_opts;
6447                         }
6448                         sbi->s_mount_state = le16_to_cpu(es->s_state);
6449
6450                         err = ext4_setup_super(sb, es, 0);
6451                         if (err)
6452                                 goto restore_opts;
6453
6454                         sb->s_flags &= ~SB_RDONLY;
6455                         if (ext4_has_feature_mmp(sb))
6456                                 if (ext4_multi_mount_protect(sb,
6457                                                 le64_to_cpu(es->s_mmp_block))) {
6458                                         err = -EROFS;
6459                                         goto restore_opts;
6460                                 }
6461 #ifdef CONFIG_QUOTA
6462                         enable_quota = 1;
6463 #endif
6464                 }
6465         }
6466
6467         /*
6468          * Reinitialize lazy itable initialization thread based on
6469          * current settings
6470          */
6471         if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
6472                 ext4_unregister_li_request(sb);
6473         else {
6474                 ext4_group_t first_not_zeroed;
6475                 first_not_zeroed = ext4_has_uninit_itable(sb);
6476                 ext4_register_li_request(sb, first_not_zeroed);
6477         }
6478
6479         /*
6480          * Handle creation of system zone data early because it can fail.
6481          * Releasing of existing data is done when we are sure remount will
6482          * succeed.
6483          */
6484         if (test_opt(sb, BLOCK_VALIDITY) && !sbi->s_system_blks) {
6485                 err = ext4_setup_system_zone(sb);
6486                 if (err)
6487                         goto restore_opts;
6488         }
6489
6490         if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) {
6491                 err = ext4_commit_super(sb);
6492                 if (err)
6493                         goto restore_opts;
6494         }
6495
6496 #ifdef CONFIG_QUOTA
6497         /* Release old quota file names */
6498         for (i = 0; i < EXT4_MAXQUOTAS; i++)
6499                 kfree(old_opts.s_qf_names[i]);
6500         if (enable_quota) {
6501                 if (sb_any_quota_suspended(sb))
6502                         dquot_resume(sb, -1);
6503                 else if (ext4_has_feature_quota(sb)) {
6504                         err = ext4_enable_quotas(sb);
6505                         if (err)
6506                                 goto restore_opts;
6507                 }
6508         }
6509 #endif
6510         if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks)
6511                 ext4_release_system_zone(sb);
6512
6513         if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb))
6514                 ext4_stop_mmpd(sbi);
6515
6516         return 0;
6517
6518 restore_opts:
6519         sb->s_flags = old_sb_flags;
6520         sbi->s_mount_opt = old_opts.s_mount_opt;
6521         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
6522         sbi->s_resuid = old_opts.s_resuid;
6523         sbi->s_resgid = old_opts.s_resgid;
6524         sbi->s_commit_interval = old_opts.s_commit_interval;
6525         sbi->s_min_batch_time = old_opts.s_min_batch_time;
6526         sbi->s_max_batch_time = old_opts.s_max_batch_time;
6527         if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks)
6528                 ext4_release_system_zone(sb);
6529 #ifdef CONFIG_QUOTA
6530         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
6531         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
6532                 to_free[i] = get_qf_name(sb, sbi, i);
6533                 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]);
6534         }
6535         synchronize_rcu();
6536         for (i = 0; i < EXT4_MAXQUOTAS; i++)
6537                 kfree(to_free[i]);
6538 #endif
6539         if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb))
6540                 ext4_stop_mmpd(sbi);
6541         return err;
6542 }
6543
6544 static int ext4_reconfigure(struct fs_context *fc)
6545 {
6546         struct super_block *sb = fc->root->d_sb;
6547         int ret;
6548
6549         fc->s_fs_info = EXT4_SB(sb);
6550
6551         ret = ext4_check_opt_consistency(fc, sb);
6552         if (ret < 0)
6553                 return ret;
6554
6555         ret = __ext4_remount(fc, sb);
6556         if (ret < 0)
6557                 return ret;
6558
6559         ext4_msg(sb, KERN_INFO, "re-mounted. Quota mode: %s.",
6560                  ext4_quota_mode(sb));
6561
6562         return 0;
6563 }
6564
6565 #ifdef CONFIG_QUOTA
6566 static int ext4_statfs_project(struct super_block *sb,
6567                                kprojid_t projid, struct kstatfs *buf)
6568 {
6569         struct kqid qid;
6570         struct dquot *dquot;
6571         u64 limit;
6572         u64 curblock;
6573
6574         qid = make_kqid_projid(projid);
6575         dquot = dqget(sb, qid);
6576         if (IS_ERR(dquot))
6577                 return PTR_ERR(dquot);
6578         spin_lock(&dquot->dq_dqb_lock);
6579
6580         limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit,
6581                              dquot->dq_dqb.dqb_bhardlimit);
6582         limit >>= sb->s_blocksize_bits;
6583
6584         if (limit && buf->f_blocks > limit) {
6585                 curblock = (dquot->dq_dqb.dqb_curspace +
6586                             dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits;
6587                 buf->f_blocks = limit;
6588                 buf->f_bfree = buf->f_bavail =
6589                         (buf->f_blocks > curblock) ?
6590                          (buf->f_blocks - curblock) : 0;
6591         }
6592
6593         limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit,
6594                              dquot->dq_dqb.dqb_ihardlimit);
6595         if (limit && buf->f_files > limit) {
6596                 buf->f_files = limit;
6597                 buf->f_ffree =
6598                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
6599                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
6600         }
6601
6602         spin_unlock(&dquot->dq_dqb_lock);
6603         dqput(dquot);
6604         return 0;
6605 }
6606 #endif
6607
6608 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
6609 {
6610         struct super_block *sb = dentry->d_sb;
6611         struct ext4_sb_info *sbi = EXT4_SB(sb);
6612         struct ext4_super_block *es = sbi->s_es;
6613         ext4_fsblk_t overhead = 0, resv_blocks;
6614         s64 bfree;
6615         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
6616
6617         if (!test_opt(sb, MINIX_DF))
6618                 overhead = sbi->s_overhead;
6619
6620         buf->f_type = EXT4_SUPER_MAGIC;
6621         buf->f_bsize = sb->s_blocksize;
6622         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
6623         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
6624                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
6625         /* prevent underflow in case that few free space is available */
6626         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
6627         buf->f_bavail = buf->f_bfree -
6628                         (ext4_r_blocks_count(es) + resv_blocks);
6629         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
6630                 buf->f_bavail = 0;
6631         buf->f_files = le32_to_cpu(es->s_inodes_count);
6632         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
6633         buf->f_namelen = EXT4_NAME_LEN;
6634         buf->f_fsid = uuid_to_fsid(es->s_uuid);
6635
6636 #ifdef CONFIG_QUOTA
6637         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
6638             sb_has_quota_limits_enabled(sb, PRJQUOTA))
6639                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
6640 #endif
6641         return 0;
6642 }
6643
6644
6645 #ifdef CONFIG_QUOTA
6646
6647 /*
6648  * Helper functions so that transaction is started before we acquire dqio_sem
6649  * to keep correct lock ordering of transaction > dqio_sem
6650  */
6651 static inline struct inode *dquot_to_inode(struct dquot *dquot)
6652 {
6653         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
6654 }
6655
6656 static int ext4_write_dquot(struct dquot *dquot)
6657 {
6658         int ret, err;
6659         handle_t *handle;
6660         struct inode *inode;
6661
6662         inode = dquot_to_inode(dquot);
6663         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
6664                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
6665         if (IS_ERR(handle))
6666                 return PTR_ERR(handle);
6667         ret = dquot_commit(dquot);
6668         err = ext4_journal_stop(handle);
6669         if (!ret)
6670                 ret = err;
6671         return ret;
6672 }
6673
6674 static int ext4_acquire_dquot(struct dquot *dquot)
6675 {
6676         int ret, err;
6677         handle_t *handle;
6678
6679         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
6680                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
6681         if (IS_ERR(handle))
6682                 return PTR_ERR(handle);
6683         ret = dquot_acquire(dquot);
6684         err = ext4_journal_stop(handle);
6685         if (!ret)
6686                 ret = err;
6687         return ret;
6688 }
6689
6690 static int ext4_release_dquot(struct dquot *dquot)
6691 {
6692         int ret, err;
6693         handle_t *handle;
6694
6695         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
6696                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
6697         if (IS_ERR(handle)) {
6698                 /* Release dquot anyway to avoid endless cycle in dqput() */
6699                 dquot_release(dquot);
6700                 return PTR_ERR(handle);
6701         }
6702         ret = dquot_release(dquot);
6703         err = ext4_journal_stop(handle);
6704         if (!ret)
6705                 ret = err;
6706         return ret;
6707 }
6708
6709 static int ext4_mark_dquot_dirty(struct dquot *dquot)
6710 {
6711         struct super_block *sb = dquot->dq_sb;
6712
6713         if (ext4_is_quota_journalled(sb)) {
6714                 dquot_mark_dquot_dirty(dquot);
6715                 return ext4_write_dquot(dquot);
6716         } else {
6717                 return dquot_mark_dquot_dirty(dquot);
6718         }
6719 }
6720
6721 static int ext4_write_info(struct super_block *sb, int type)
6722 {
6723         int ret, err;
6724         handle_t *handle;
6725
6726         /* Data block + inode block */
6727         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
6728         if (IS_ERR(handle))
6729                 return PTR_ERR(handle);
6730         ret = dquot_commit_info(sb, type);
6731         err = ext4_journal_stop(handle);
6732         if (!ret)
6733                 ret = err;
6734         return ret;
6735 }
6736
6737 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
6738 {
6739         struct ext4_inode_info *ei = EXT4_I(inode);
6740
6741         /* The first argument of lockdep_set_subclass has to be
6742          * *exactly* the same as the argument to init_rwsem() --- in
6743          * this case, in init_once() --- or lockdep gets unhappy
6744          * because the name of the lock is set using the
6745          * stringification of the argument to init_rwsem().
6746          */
6747         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
6748         lockdep_set_subclass(&ei->i_data_sem, subclass);
6749 }
6750
6751 /*
6752  * Standard function to be called on quota_on
6753  */
6754 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
6755                          const struct path *path)
6756 {
6757         int err;
6758
6759         if (!test_opt(sb, QUOTA))
6760                 return -EINVAL;
6761
6762         /* Quotafile not on the same filesystem? */
6763         if (path->dentry->d_sb != sb)
6764                 return -EXDEV;
6765
6766         /* Quota already enabled for this file? */
6767         if (IS_NOQUOTA(d_inode(path->dentry)))
6768                 return -EBUSY;
6769
6770         /* Journaling quota? */
6771         if (EXT4_SB(sb)->s_qf_names[type]) {
6772                 /* Quotafile not in fs root? */
6773                 if (path->dentry->d_parent != sb->s_root)
6774                         ext4_msg(sb, KERN_WARNING,
6775                                 "Quota file not on filesystem root. "
6776                                 "Journaled quota will not work");
6777                 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
6778         } else {
6779                 /*
6780                  * Clear the flag just in case mount options changed since
6781                  * last time.
6782                  */
6783                 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
6784         }
6785
6786         /*
6787          * When we journal data on quota file, we have to flush journal to see
6788          * all updates to the file when we bypass pagecache...
6789          */
6790         if (EXT4_SB(sb)->s_journal &&
6791             ext4_should_journal_data(d_inode(path->dentry))) {
6792                 /*
6793                  * We don't need to lock updates but journal_flush() could
6794                  * otherwise be livelocked...
6795                  */
6796                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
6797                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal, 0);
6798                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
6799                 if (err)
6800                         return err;
6801         }
6802
6803         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
6804         err = dquot_quota_on(sb, type, format_id, path);
6805         if (!err) {
6806                 struct inode *inode = d_inode(path->dentry);
6807                 handle_t *handle;
6808
6809                 /*
6810                  * Set inode flags to prevent userspace from messing with quota
6811                  * files. If this fails, we return success anyway since quotas
6812                  * are already enabled and this is not a hard failure.
6813                  */
6814                 inode_lock(inode);
6815                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
6816                 if (IS_ERR(handle))
6817                         goto unlock_inode;
6818                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
6819                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
6820                                 S_NOATIME | S_IMMUTABLE);
6821                 err = ext4_mark_inode_dirty(handle, inode);
6822                 ext4_journal_stop(handle);
6823         unlock_inode:
6824                 inode_unlock(inode);
6825                 if (err)
6826                         dquot_quota_off(sb, type);
6827         }
6828         if (err)
6829                 lockdep_set_quota_inode(path->dentry->d_inode,
6830                                              I_DATA_SEM_NORMAL);
6831         return err;
6832 }
6833
6834 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
6835                              unsigned int flags)
6836 {
6837         int err;
6838         struct inode *qf_inode;
6839         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
6840                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
6841                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
6842                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
6843         };
6844
6845         BUG_ON(!ext4_has_feature_quota(sb));
6846
6847         if (!qf_inums[type])
6848                 return -EPERM;
6849
6850         qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL);
6851         if (IS_ERR(qf_inode)) {
6852                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
6853                 return PTR_ERR(qf_inode);
6854         }
6855
6856         /* Don't account quota for quota files to avoid recursion */
6857         qf_inode->i_flags |= S_NOQUOTA;
6858         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
6859         err = dquot_load_quota_inode(qf_inode, type, format_id, flags);
6860         if (err)
6861                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
6862         iput(qf_inode);
6863
6864         return err;
6865 }
6866
6867 /* Enable usage tracking for all quota types. */
6868 int ext4_enable_quotas(struct super_block *sb)
6869 {
6870         int type, err = 0;
6871         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
6872                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
6873                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
6874                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
6875         };
6876         bool quota_mopt[EXT4_MAXQUOTAS] = {
6877                 test_opt(sb, USRQUOTA),
6878                 test_opt(sb, GRPQUOTA),
6879                 test_opt(sb, PRJQUOTA),
6880         };
6881
6882         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
6883         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
6884                 if (qf_inums[type]) {
6885                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
6886                                 DQUOT_USAGE_ENABLED |
6887                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
6888                         if (err) {
6889                                 ext4_warning(sb,
6890                                         "Failed to enable quota tracking "
6891                                         "(type=%d, err=%d). Please run "
6892                                         "e2fsck to fix.", type, err);
6893                                 for (type--; type >= 0; type--) {
6894                                         struct inode *inode;
6895
6896                                         inode = sb_dqopt(sb)->files[type];
6897                                         if (inode)
6898                                                 inode = igrab(inode);
6899                                         dquot_quota_off(sb, type);
6900                                         if (inode) {
6901                                                 lockdep_set_quota_inode(inode,
6902                                                         I_DATA_SEM_NORMAL);
6903                                                 iput(inode);
6904                                         }
6905                                 }
6906
6907                                 return err;
6908                         }
6909                 }
6910         }
6911         return 0;
6912 }
6913
6914 static int ext4_quota_off(struct super_block *sb, int type)
6915 {
6916         struct inode *inode = sb_dqopt(sb)->files[type];
6917         handle_t *handle;
6918         int err;
6919
6920         /* Force all delayed allocation blocks to be allocated.
6921          * Caller already holds s_umount sem */
6922         if (test_opt(sb, DELALLOC))
6923                 sync_filesystem(sb);
6924
6925         if (!inode || !igrab(inode))
6926                 goto out;
6927
6928         err = dquot_quota_off(sb, type);
6929         if (err || ext4_has_feature_quota(sb))
6930                 goto out_put;
6931
6932         inode_lock(inode);
6933         /*
6934          * Update modification times of quota files when userspace can
6935          * start looking at them. If we fail, we return success anyway since
6936          * this is not a hard failure and quotas are already disabled.
6937          */
6938         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
6939         if (IS_ERR(handle)) {
6940                 err = PTR_ERR(handle);
6941                 goto out_unlock;
6942         }
6943         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
6944         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
6945         inode->i_mtime = inode->i_ctime = current_time(inode);
6946         err = ext4_mark_inode_dirty(handle, inode);
6947         ext4_journal_stop(handle);
6948 out_unlock:
6949         inode_unlock(inode);
6950 out_put:
6951         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
6952         iput(inode);
6953         return err;
6954 out:
6955         return dquot_quota_off(sb, type);
6956 }
6957
6958 /* Read data from quotafile - avoid pagecache and such because we cannot afford
6959  * acquiring the locks... As quota files are never truncated and quota code
6960  * itself serializes the operations (and no one else should touch the files)
6961  * we don't have to be afraid of races */
6962 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
6963                                size_t len, loff_t off)
6964 {
6965         struct inode *inode = sb_dqopt(sb)->files[type];
6966         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
6967         int offset = off & (sb->s_blocksize - 1);
6968         int tocopy;
6969         size_t toread;
6970         struct buffer_head *bh;
6971         loff_t i_size = i_size_read(inode);
6972
6973         if (off > i_size)
6974                 return 0;
6975         if (off+len > i_size)
6976                 len = i_size-off;
6977         toread = len;
6978         while (toread > 0) {
6979                 tocopy = sb->s_blocksize - offset < toread ?
6980                                 sb->s_blocksize - offset : toread;
6981                 bh = ext4_bread(NULL, inode, blk, 0);
6982                 if (IS_ERR(bh))
6983                         return PTR_ERR(bh);
6984                 if (!bh)        /* A hole? */
6985                         memset(data, 0, tocopy);
6986                 else
6987                         memcpy(data, bh->b_data+offset, tocopy);
6988                 brelse(bh);
6989                 offset = 0;
6990                 toread -= tocopy;
6991                 data += tocopy;
6992                 blk++;
6993         }
6994         return len;
6995 }
6996
6997 /* Write to quotafile (we know the transaction is already started and has
6998  * enough credits) */
6999 static ssize_t ext4_quota_write(struct super_block *sb, int type,
7000                                 const char *data, size_t len, loff_t off)
7001 {
7002         struct inode *inode = sb_dqopt(sb)->files[type];
7003         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
7004         int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1);
7005         int retries = 0;
7006         struct buffer_head *bh;
7007         handle_t *handle = journal_current_handle();
7008
7009         if (!handle) {
7010                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
7011                         " cancelled because transaction is not started",
7012                         (unsigned long long)off, (unsigned long long)len);
7013                 return -EIO;
7014         }
7015         /*
7016          * Since we account only one data block in transaction credits,
7017          * then it is impossible to cross a block boundary.
7018          */
7019         if (sb->s_blocksize - offset < len) {
7020                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
7021                         " cancelled because not block aligned",
7022                         (unsigned long long)off, (unsigned long long)len);
7023                 return -EIO;
7024         }
7025
7026         do {
7027                 bh = ext4_bread(handle, inode, blk,
7028                                 EXT4_GET_BLOCKS_CREATE |
7029                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
7030         } while (PTR_ERR(bh) == -ENOSPC &&
7031                  ext4_should_retry_alloc(inode->i_sb, &retries));
7032         if (IS_ERR(bh))
7033                 return PTR_ERR(bh);
7034         if (!bh)
7035                 goto out;
7036         BUFFER_TRACE(bh, "get write access");
7037         err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
7038         if (err) {
7039                 brelse(bh);
7040                 return err;
7041         }
7042         lock_buffer(bh);
7043         memcpy(bh->b_data+offset, data, len);
7044         flush_dcache_page(bh->b_page);
7045         unlock_buffer(bh);
7046         err = ext4_handle_dirty_metadata(handle, NULL, bh);
7047         brelse(bh);
7048 out:
7049         if (inode->i_size < off + len) {
7050                 i_size_write(inode, off + len);
7051                 EXT4_I(inode)->i_disksize = inode->i_size;
7052                 err2 = ext4_mark_inode_dirty(handle, inode);
7053                 if (unlikely(err2 && !err))
7054                         err = err2;
7055         }
7056         return err ? err : len;
7057 }
7058 #endif
7059
7060 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
7061 static inline void register_as_ext2(void)
7062 {
7063         int err = register_filesystem(&ext2_fs_type);
7064         if (err)
7065                 printk(KERN_WARNING
7066                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
7067 }
7068
7069 static inline void unregister_as_ext2(void)
7070 {
7071         unregister_filesystem(&ext2_fs_type);
7072 }
7073
7074 static inline int ext2_feature_set_ok(struct super_block *sb)
7075 {
7076         if (ext4_has_unknown_ext2_incompat_features(sb))
7077                 return 0;
7078         if (sb_rdonly(sb))
7079                 return 1;
7080         if (ext4_has_unknown_ext2_ro_compat_features(sb))
7081                 return 0;
7082         return 1;
7083 }
7084 #else
7085 static inline void register_as_ext2(void) { }
7086 static inline void unregister_as_ext2(void) { }
7087 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
7088 #endif
7089
7090 static inline void register_as_ext3(void)
7091 {
7092         int err = register_filesystem(&ext3_fs_type);
7093         if (err)
7094                 printk(KERN_WARNING
7095                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
7096 }
7097
7098 static inline void unregister_as_ext3(void)
7099 {
7100         unregister_filesystem(&ext3_fs_type);
7101 }
7102
7103 static inline int ext3_feature_set_ok(struct super_block *sb)
7104 {
7105         if (ext4_has_unknown_ext3_incompat_features(sb))
7106                 return 0;
7107         if (!ext4_has_feature_journal(sb))
7108                 return 0;
7109         if (sb_rdonly(sb))
7110                 return 1;
7111         if (ext4_has_unknown_ext3_ro_compat_features(sb))
7112                 return 0;
7113         return 1;
7114 }
7115
7116 static struct file_system_type ext4_fs_type = {
7117         .owner                  = THIS_MODULE,
7118         .name                   = "ext4",
7119         .init_fs_context        = ext4_init_fs_context,
7120         .parameters             = ext4_param_specs,
7121         .kill_sb                = kill_block_super,
7122         .fs_flags               = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
7123 };
7124 MODULE_ALIAS_FS("ext4");
7125
7126 /* Shared across all ext4 file systems */
7127 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
7128
7129 static int __init ext4_init_fs(void)
7130 {
7131         int i, err;
7132
7133         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
7134         ext4_li_info = NULL;
7135
7136         /* Build-time check for flags consistency */
7137         ext4_check_flag_values();
7138
7139         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
7140                 init_waitqueue_head(&ext4__ioend_wq[i]);
7141
7142         err = ext4_init_es();
7143         if (err)
7144                 return err;
7145
7146         err = ext4_init_pending();
7147         if (err)
7148                 goto out7;
7149
7150         err = ext4_init_post_read_processing();
7151         if (err)
7152                 goto out6;
7153
7154         err = ext4_init_pageio();
7155         if (err)
7156                 goto out5;
7157
7158         err = ext4_init_system_zone();
7159         if (err)
7160                 goto out4;
7161
7162         err = ext4_init_sysfs();
7163         if (err)
7164                 goto out3;
7165
7166         err = ext4_init_mballoc();
7167         if (err)
7168                 goto out2;
7169         err = init_inodecache();
7170         if (err)
7171                 goto out1;
7172
7173         err = ext4_fc_init_dentry_cache();
7174         if (err)
7175                 goto out05;
7176
7177         register_as_ext3();
7178         register_as_ext2();
7179         err = register_filesystem(&ext4_fs_type);
7180         if (err)
7181                 goto out;
7182
7183         return 0;
7184 out:
7185         unregister_as_ext2();
7186         unregister_as_ext3();
7187         ext4_fc_destroy_dentry_cache();
7188 out05:
7189         destroy_inodecache();
7190 out1:
7191         ext4_exit_mballoc();
7192 out2:
7193         ext4_exit_sysfs();
7194 out3:
7195         ext4_exit_system_zone();
7196 out4:
7197         ext4_exit_pageio();
7198 out5:
7199         ext4_exit_post_read_processing();
7200 out6:
7201         ext4_exit_pending();
7202 out7:
7203         ext4_exit_es();
7204
7205         return err;
7206 }
7207
7208 static void __exit ext4_exit_fs(void)
7209 {
7210         ext4_destroy_lazyinit_thread();
7211         unregister_as_ext2();
7212         unregister_as_ext3();
7213         unregister_filesystem(&ext4_fs_type);
7214         ext4_fc_destroy_dentry_cache();
7215         destroy_inodecache();
7216         ext4_exit_mballoc();
7217         ext4_exit_sysfs();
7218         ext4_exit_system_zone();
7219         ext4_exit_pageio();
7220         ext4_exit_post_read_processing();
7221         ext4_exit_es();
7222         ext4_exit_pending();
7223 }
7224
7225 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
7226 MODULE_DESCRIPTION("Fourth Extended Filesystem");
7227 MODULE_LICENSE("GPL");
7228 MODULE_SOFTDEP("pre: crc32c");
7229 module_init(ext4_init_fs)
7230 module_exit(ext4_exit_fs)