fs/ext4/super.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/ext4/super.c
   4  *
   5  * Copyright (C) 1992, 1993, 1994, 1995
   6  * Remy Card (card@masi.ibp.fr)
   7  * Laboratoire MASI - Institut Blaise Pascal
   8  * Universite Pierre et Marie Curie (Paris VI)
   9  *
  10  *  from
  11  *
  12  *  linux/fs/minix/inode.c
  13  *
  14  *  Copyright (C) 1991, 1992  Linus Torvalds
  15  *
  16  *  Big-endian to little-endian byte-swapping/bitmaps by
  17  *        David S. Miller (davem@caip.rutgers.edu), 1995
  18  */
  19
  20 #include <linux/module.h>
  21 #include <linux/string.h>
  22 #include <linux/fs.h>
  23 #include <linux/time.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/backing-dev.h>
  29 #include <linux/parser.h>
  30 #include <linux/buffer_head.h>
  31 #include <linux/exportfs.h>
  32 #include <linux/vfs.h>
  33 #include <linux/random.h>
  34 #include <linux/mount.h>
  35 #include <linux/namei.h>
  36 #include <linux/quotaops.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/dax.h>
  42 #include <linux/uaccess.h>
  43 #include <linux/iversion.h>
  44 #include <linux/unicode.h>
  45 #include <linux/part_stat.h>
  46 #include <linux/kthread.h>
  47 #include <linux/freezer.h>
  48 #include <linux/fsnotify.h>
  49 #include <linux/fs_context.h>
  50 #include <linux/fs_parser.h>
  51
  52 #include "ext4.h"
  53 #include "ext4_extents.h"       /* Needed for trace points definition */
  54 #include "ext4_jbd2.h"
  55 #include "xattr.h"
  56 #include "acl.h"
  57 #include "mballoc.h"
  58 #include "fsmap.h"
  59
  60 #define CREATE_TRACE_POINTS
  61 #include <trace/events/ext4.h>
  62
  63 static struct ext4_lazy_init *ext4_li_info;
  64 static DEFINE_MUTEX(ext4_li_mtx);
  65 static struct ratelimit_state ext4_mount_msg_ratelimit;
  66
  67 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  68                              unsigned long journal_devnum);
  69 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  70 static void ext4_update_super(struct super_block *sb);
  71 static int ext4_commit_super(struct super_block *sb);
  72 static int ext4_mark_recovery_complete(struct super_block *sb,
  73                                         struct ext4_super_block *es);
  74 static int ext4_clear_journal_err(struct super_block *sb,
  75                                   struct ext4_super_block *es);
  76 static int ext4_sync_fs(struct super_block *sb, int wait);
  77 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  78 static int ext4_unfreeze(struct super_block *sb);
  79 static int ext4_freeze(struct super_block *sb);
  80 static inline int ext2_feature_set_ok(struct super_block *sb);
  81 static inline int ext3_feature_set_ok(struct super_block *sb);
  82 static void ext4_destroy_lazyinit_thread(void);
  83 static void ext4_unregister_li_request(struct super_block *sb);
  84 static void ext4_clear_request_list(void);
  85 static struct inode *ext4_get_journal_inode(struct super_block *sb,
  86                                             unsigned int journal_inum);
  87 static int ext4_validate_options(struct fs_context *fc);
  88 static int ext4_check_opt_consistency(struct fs_context *fc,
  89                                       struct super_block *sb);
  90 static 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         ext4_unregister_li_request(sb);
1203         ext4_quota_off_umount(sb);
1204
1205         flush_work(&sbi->s_error_work);
1206         destroy_workqueue(sbi->rsv_conversion_wq);
1207         ext4_release_orphan_info(sb);
1208
1209         /*
1210          * Unregister sysfs before destroying jbd2 journal.
1211          * Since we could still access attr_journal_task attribute via sysfs
1212          * path which could have sbi->s_journal->j_task as NULL
1213          */
1214         ext4_unregister_sysfs(sb);
1215
1216         if (sbi->s_journal) {
1217                 aborted = is_journal_aborted(sbi->s_journal);
1218                 err = jbd2_journal_destroy(sbi->s_journal);
1219                 sbi->s_journal = NULL;
1220                 if ((err < 0) && !aborted) {
1221                         ext4_abort(sb, -err, "Couldn't clean up the journal");
1222                 }
1223         }
1224
1225         ext4_es_unregister_shrinker(sbi);
1226         del_timer_sync(&sbi->s_err_report);
1227         ext4_release_system_zone(sb);
1228         ext4_mb_release(sb);
1229         ext4_ext_release(sb);
1230
1231         if (!sb_rdonly(sb) && !aborted) {
1232                 ext4_clear_feature_journal_needs_recovery(sb);
1233                 ext4_clear_feature_orphan_present(sb);
1234                 es->s_state = cpu_to_le16(sbi->s_mount_state);
1235         }
1236         if (!sb_rdonly(sb))
1237                 ext4_commit_super(sb);
1238
1239         rcu_read_lock();
1240         group_desc = rcu_dereference(sbi->s_group_desc);
1241         for (i = 0; i < sbi->s_gdb_count; i++)
1242                 brelse(group_desc[i]);
1243         kvfree(group_desc);
1244         flex_groups = rcu_dereference(sbi->s_flex_groups);
1245         if (flex_groups) {
1246                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
1247                         kvfree(flex_groups[i]);
1248                 kvfree(flex_groups);
1249         }
1250         rcu_read_unlock();
1251         percpu_counter_destroy(&sbi->s_freeclusters_counter);
1252         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1253         percpu_counter_destroy(&sbi->s_dirs_counter);
1254         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
1255         percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit);
1256         percpu_free_rwsem(&sbi->s_writepages_rwsem);
1257 #ifdef CONFIG_QUOTA
1258         for (i = 0; i < EXT4_MAXQUOTAS; i++)
1259                 kfree(get_qf_name(sb, sbi, i));
1260 #endif
1261
1262         /* Debugging code just in case the in-memory inode orphan list
1263          * isn't empty.  The on-disk one can be non-empty if we've
1264          * detected an error and taken the fs readonly, but the
1265          * in-memory list had better be clean by this point. */
1266         if (!list_empty(&sbi->s_orphan))
1267                 dump_orphan_list(sb, sbi);
1268         ASSERT(list_empty(&sbi->s_orphan));
1269
1270         sync_blockdev(sb->s_bdev);
1271         invalidate_bdev(sb->s_bdev);
1272         if (sbi->s_journal_bdev && sbi->s_journal_bdev != sb->s_bdev) {
1273                 /*
1274                  * Invalidate the journal device's buffers.  We don't want them
1275                  * floating about in memory - the physical journal device may
1276                  * hotswapped, and it breaks the `ro-after' testing code.
1277                  */
1278                 sync_blockdev(sbi->s_journal_bdev);
1279                 invalidate_bdev(sbi->s_journal_bdev);
1280                 ext4_blkdev_remove(sbi);
1281         }
1282
1283         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
1284         sbi->s_ea_inode_cache = NULL;
1285
1286         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
1287         sbi->s_ea_block_cache = NULL;
1288
1289         ext4_stop_mmpd(sbi);
1290
1291         brelse(sbi->s_sbh);
1292         sb->s_fs_info = NULL;
1293         /*
1294          * Now that we are completely done shutting down the
1295          * superblock, we need to actually destroy the kobject.
1296          */
1297         kobject_put(&sbi->s_kobj);
1298         wait_for_completion(&sbi->s_kobj_unregister);
1299         if (sbi->s_chksum_driver)
1300                 crypto_free_shash(sbi->s_chksum_driver);
1301         kfree(sbi->s_blockgroup_lock);
1302         fs_put_dax(sbi->s_daxdev);
1303         fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy);
1304 #if IS_ENABLED(CONFIG_UNICODE)
1305         utf8_unload(sb->s_encoding);
1306 #endif
1307         kfree(sbi);
1308 }
1309
1310 static struct kmem_cache *ext4_inode_cachep;
1311
1312 /*
1313  * Called inside transaction, so use GFP_NOFS
1314  */
1315 static struct inode *ext4_alloc_inode(struct super_block *sb)
1316 {
1317         struct ext4_inode_info *ei;
1318
1319         ei = alloc_inode_sb(sb, ext4_inode_cachep, GFP_NOFS);
1320         if (!ei)
1321                 return NULL;
1322
1323         inode_set_iversion(&ei->vfs_inode, 1);
1324         spin_lock_init(&ei->i_raw_lock);
1325         INIT_LIST_HEAD(&ei->i_prealloc_list);
1326         atomic_set(&ei->i_prealloc_active, 0);
1327         spin_lock_init(&ei->i_prealloc_lock);
1328         ext4_es_init_tree(&ei->i_es_tree);
1329         rwlock_init(&ei->i_es_lock);
1330         INIT_LIST_HEAD(&ei->i_es_list);
1331         ei->i_es_all_nr = 0;
1332         ei->i_es_shk_nr = 0;
1333         ei->i_es_shrink_lblk = 0;
1334         ei->i_reserved_data_blocks = 0;
1335         spin_lock_init(&(ei->i_block_reservation_lock));
1336         ext4_init_pending_tree(&ei->i_pending_tree);
1337 #ifdef CONFIG_QUOTA
1338         ei->i_reserved_quota = 0;
1339         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
1340 #endif
1341         ei->jinode = NULL;
1342         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
1343         spin_lock_init(&ei->i_completed_io_lock);
1344         ei->i_sync_tid = 0;
1345         ei->i_datasync_tid = 0;
1346         atomic_set(&ei->i_unwritten, 0);
1347         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
1348         ext4_fc_init_inode(&ei->vfs_inode);
1349         mutex_init(&ei->i_fc_lock);
1350         return &ei->vfs_inode;
1351 }
1352
1353 static int ext4_drop_inode(struct inode *inode)
1354 {
1355         int drop = generic_drop_inode(inode);
1356
1357         if (!drop)
1358                 drop = fscrypt_drop_inode(inode);
1359
1360         trace_ext4_drop_inode(inode, drop);
1361         return drop;
1362 }
1363
1364 static void ext4_free_in_core_inode(struct inode *inode)
1365 {
1366         fscrypt_free_inode(inode);
1367         if (!list_empty(&(EXT4_I(inode)->i_fc_list))) {
1368                 pr_warn("%s: inode %ld still in fc list",
1369                         __func__, inode->i_ino);
1370         }
1371         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1372 }
1373
1374 static void ext4_destroy_inode(struct inode *inode)
1375 {
1376         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1377                 ext4_msg(inode->i_sb, KERN_ERR,
1378                          "Inode %lu (%p): orphan list check failed!",
1379                          inode->i_ino, EXT4_I(inode));
1380                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1381                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1382                                 true);
1383                 dump_stack();
1384         }
1385
1386         if (EXT4_I(inode)->i_reserved_data_blocks)
1387                 ext4_msg(inode->i_sb, KERN_ERR,
1388                          "Inode %lu (%p): i_reserved_data_blocks (%u) not cleared!",
1389                          inode->i_ino, EXT4_I(inode),
1390                          EXT4_I(inode)->i_reserved_data_blocks);
1391 }
1392
1393 static void init_once(void *foo)
1394 {
1395         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1396
1397         INIT_LIST_HEAD(&ei->i_orphan);
1398         init_rwsem(&ei->xattr_sem);
1399         init_rwsem(&ei->i_data_sem);
1400         inode_init_once(&ei->vfs_inode);
1401         ext4_fc_init_inode(&ei->vfs_inode);
1402 }
1403
1404 static int __init init_inodecache(void)
1405 {
1406         ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache",
1407                                 sizeof(struct ext4_inode_info), 0,
1408                                 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
1409                                         SLAB_ACCOUNT),
1410                                 offsetof(struct ext4_inode_info, i_data),
1411                                 sizeof_field(struct ext4_inode_info, i_data),
1412                                 init_once);
1413         if (ext4_inode_cachep == NULL)
1414                 return -ENOMEM;
1415         return 0;
1416 }
1417
1418 static void destroy_inodecache(void)
1419 {
1420         /*
1421          * Make sure all delayed rcu free inodes are flushed before we
1422          * destroy cache.
1423          */
1424         rcu_barrier();
1425         kmem_cache_destroy(ext4_inode_cachep);
1426 }
1427
1428 void ext4_clear_inode(struct inode *inode)
1429 {
1430         ext4_fc_del(inode);
1431         invalidate_inode_buffers(inode);
1432         clear_inode(inode);
1433         ext4_discard_preallocations(inode, 0);
1434         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1435         dquot_drop(inode);
1436         if (EXT4_I(inode)->jinode) {
1437                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1438                                                EXT4_I(inode)->jinode);
1439                 jbd2_free_inode(EXT4_I(inode)->jinode);
1440                 EXT4_I(inode)->jinode = NULL;
1441         }
1442         fscrypt_put_encryption_info(inode);
1443         fsverity_cleanup_inode(inode);
1444 }
1445
1446 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1447                                         u64 ino, u32 generation)
1448 {
1449         struct inode *inode;
1450
1451         /*
1452          * Currently we don't know the generation for parent directory, so
1453          * a generation of 0 means "accept any"
1454          */
1455         inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE);
1456         if (IS_ERR(inode))
1457                 return ERR_CAST(inode);
1458         if (generation && inode->i_generation != generation) {
1459                 iput(inode);
1460                 return ERR_PTR(-ESTALE);
1461         }
1462
1463         return inode;
1464 }
1465
1466 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1467                                         int fh_len, int fh_type)
1468 {
1469         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1470                                     ext4_nfs_get_inode);
1471 }
1472
1473 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1474                                         int fh_len, int fh_type)
1475 {
1476         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1477                                     ext4_nfs_get_inode);
1478 }
1479
1480 static int ext4_nfs_commit_metadata(struct inode *inode)
1481 {
1482         struct writeback_control wbc = {
1483                 .sync_mode = WB_SYNC_ALL
1484         };
1485
1486         trace_ext4_nfs_commit_metadata(inode);
1487         return ext4_write_inode(inode, &wbc);
1488 }
1489
1490 #ifdef CONFIG_FS_ENCRYPTION
1491 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1492 {
1493         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1494                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1495 }
1496
1497 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1498                                                         void *fs_data)
1499 {
1500         handle_t *handle = fs_data;
1501         int res, res2, credits, retries = 0;
1502
1503         /*
1504          * Encrypting the root directory is not allowed because e2fsck expects
1505          * lost+found to exist and be unencrypted, and encrypting the root
1506          * directory would imply encrypting the lost+found directory as well as
1507          * the filename "lost+found" itself.
1508          */
1509         if (inode->i_ino == EXT4_ROOT_INO)
1510                 return -EPERM;
1511
1512         if (WARN_ON_ONCE(IS_DAX(inode) && i_size_read(inode)))
1513                 return -EINVAL;
1514
1515         if (ext4_test_inode_flag(inode, EXT4_INODE_DAX))
1516                 return -EOPNOTSUPP;
1517
1518         res = ext4_convert_inline_data(inode);
1519         if (res)
1520                 return res;
1521
1522         /*
1523          * If a journal handle was specified, then the encryption context is
1524          * being set on a new inode via inheritance and is part of a larger
1525          * transaction to create the inode.  Otherwise the encryption context is
1526          * being set on an existing inode in its own transaction.  Only in the
1527          * latter case should the "retry on ENOSPC" logic be used.
1528          */
1529
1530         if (handle) {
1531                 res = ext4_xattr_set_handle(handle, inode,
1532                                             EXT4_XATTR_INDEX_ENCRYPTION,
1533                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1534                                             ctx, len, 0);
1535                 if (!res) {
1536                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1537                         ext4_clear_inode_state(inode,
1538                                         EXT4_STATE_MAY_INLINE_DATA);
1539                         /*
1540                          * Update inode->i_flags - S_ENCRYPTED will be enabled,
1541                          * S_DAX may be disabled
1542                          */
1543                         ext4_set_inode_flags(inode, false);
1544                 }
1545                 return res;
1546         }
1547
1548         res = dquot_initialize(inode);
1549         if (res)
1550                 return res;
1551 retry:
1552         res = ext4_xattr_set_credits(inode, len, false /* is_create */,
1553                                      &credits);
1554         if (res)
1555                 return res;
1556
1557         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
1558         if (IS_ERR(handle))
1559                 return PTR_ERR(handle);
1560
1561         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1562                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1563                                     ctx, len, 0);
1564         if (!res) {
1565                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1566                 /*
1567                  * Update inode->i_flags - S_ENCRYPTED will be enabled,
1568                  * S_DAX may be disabled
1569                  */
1570                 ext4_set_inode_flags(inode, false);
1571                 res = ext4_mark_inode_dirty(handle, inode);
1572                 if (res)
1573                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1574         }
1575         res2 = ext4_journal_stop(handle);
1576
1577         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1578                 goto retry;
1579         if (!res)
1580                 res = res2;
1581         return res;
1582 }
1583
1584 static const union fscrypt_policy *ext4_get_dummy_policy(struct super_block *sb)
1585 {
1586         return EXT4_SB(sb)->s_dummy_enc_policy.policy;
1587 }
1588
1589 static bool ext4_has_stable_inodes(struct super_block *sb)
1590 {
1591         return ext4_has_feature_stable_inodes(sb);
1592 }
1593
1594 static void ext4_get_ino_and_lblk_bits(struct super_block *sb,
1595                                        int *ino_bits_ret, int *lblk_bits_ret)
1596 {
1597         *ino_bits_ret = 8 * sizeof(EXT4_SB(sb)->s_es->s_inodes_count);
1598         *lblk_bits_ret = 8 * sizeof(ext4_lblk_t);
1599 }
1600
1601 static const struct fscrypt_operations ext4_cryptops = {
1602         .key_prefix             = "ext4:",
1603         .get_context            = ext4_get_context,
1604         .set_context            = ext4_set_context,
1605         .get_dummy_policy       = ext4_get_dummy_policy,
1606         .empty_dir              = ext4_empty_dir,
1607         .has_stable_inodes      = ext4_has_stable_inodes,
1608         .get_ino_and_lblk_bits  = ext4_get_ino_and_lblk_bits,
1609 };
1610 #endif
1611
1612 #ifdef CONFIG_QUOTA
1613 static const char * const quotatypes[] = INITQFNAMES;
1614 #define QTYPE2NAME(t) (quotatypes[t])
1615
1616 static int ext4_write_dquot(struct dquot *dquot);
1617 static int ext4_acquire_dquot(struct dquot *dquot);
1618 static int ext4_release_dquot(struct dquot *dquot);
1619 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1620 static int ext4_write_info(struct super_block *sb, int type);
1621 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1622                          const struct path *path);
1623 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1624                                size_t len, loff_t off);
1625 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1626                                 const char *data, size_t len, loff_t off);
1627 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1628                              unsigned int flags);
1629
1630 static struct dquot **ext4_get_dquots(struct inode *inode)
1631 {
1632         return EXT4_I(inode)->i_dquot;
1633 }
1634
1635 static const struct dquot_operations ext4_quota_operations = {
1636         .get_reserved_space     = ext4_get_reserved_space,
1637         .write_dquot            = ext4_write_dquot,
1638         .acquire_dquot          = ext4_acquire_dquot,
1639         .release_dquot          = ext4_release_dquot,
1640         .mark_dirty             = ext4_mark_dquot_dirty,
1641         .write_info             = ext4_write_info,
1642         .alloc_dquot            = dquot_alloc,
1643         .destroy_dquot          = dquot_destroy,
1644         .get_projid             = ext4_get_projid,
1645         .get_inode_usage        = ext4_get_inode_usage,
1646         .get_next_id            = dquot_get_next_id,
1647 };
1648
1649 static const struct quotactl_ops ext4_qctl_operations = {
1650         .quota_on       = ext4_quota_on,
1651         .quota_off      = ext4_quota_off,
1652         .quota_sync     = dquot_quota_sync,
1653         .get_state      = dquot_get_state,
1654         .set_info       = dquot_set_dqinfo,
1655         .get_dqblk      = dquot_get_dqblk,
1656         .set_dqblk      = dquot_set_dqblk,
1657         .get_nextdqblk  = dquot_get_next_dqblk,
1658 };
1659 #endif
1660
1661 static const struct super_operations ext4_sops = {
1662         .alloc_inode    = ext4_alloc_inode,
1663         .free_inode     = ext4_free_in_core_inode,
1664         .destroy_inode  = ext4_destroy_inode,
1665         .write_inode    = ext4_write_inode,
1666         .dirty_inode    = ext4_dirty_inode,
1667         .drop_inode     = ext4_drop_inode,
1668         .evict_inode    = ext4_evict_inode,
1669         .put_super      = ext4_put_super,
1670         .sync_fs        = ext4_sync_fs,
1671         .freeze_fs      = ext4_freeze,
1672         .unfreeze_fs    = ext4_unfreeze,
1673         .statfs         = ext4_statfs,
1674         .show_options   = ext4_show_options,
1675 #ifdef CONFIG_QUOTA
1676         .quota_read     = ext4_quota_read,
1677         .quota_write    = ext4_quota_write,
1678         .get_dquots     = ext4_get_dquots,
1679 #endif
1680 };
1681
1682 static const struct export_operations ext4_export_ops = {
1683         .fh_to_dentry = ext4_fh_to_dentry,
1684         .fh_to_parent = ext4_fh_to_parent,
1685         .get_parent = ext4_get_parent,
1686         .commit_metadata = ext4_nfs_commit_metadata,
1687 };
1688
1689 enum {
1690         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1691         Opt_resgid, Opt_resuid, Opt_sb,
1692         Opt_nouid32, Opt_debug, Opt_removed,
1693         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1694         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1695         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1696         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1697         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1698         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1699         Opt_inlinecrypt,
1700         Opt_usrjquota, Opt_grpjquota, Opt_quota,
1701         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1702         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version,
1703         Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never,
1704         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
1705         Opt_nowarn_on_error, Opt_mblk_io_submit, Opt_debug_want_extra_isize,
1706         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1707         Opt_inode_readahead_blks, Opt_journal_ioprio,
1708         Opt_dioread_nolock, Opt_dioread_lock,
1709         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1710         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1711         Opt_no_prefetch_block_bitmaps, Opt_mb_optimize_scan,
1712         Opt_errors, Opt_data, Opt_data_err, Opt_jqfmt, Opt_dax_type,
1713 #ifdef CONFIG_EXT4_DEBUG
1714         Opt_fc_debug_max_replay, Opt_fc_debug_force
1715 #endif
1716 };
1717
1718 static const struct constant_table ext4_param_errors[] = {
1719         {"continue",    EXT4_MOUNT_ERRORS_CONT},
1720         {"panic",       EXT4_MOUNT_ERRORS_PANIC},
1721         {"remount-ro",  EXT4_MOUNT_ERRORS_RO},
1722         {}
1723 };
1724
1725 static const struct constant_table ext4_param_data[] = {
1726         {"journal",     EXT4_MOUNT_JOURNAL_DATA},
1727         {"ordered",     EXT4_MOUNT_ORDERED_DATA},
1728         {"writeback",   EXT4_MOUNT_WRITEBACK_DATA},
1729         {}
1730 };
1731
1732 static const struct constant_table ext4_param_data_err[] = {
1733         {"abort",       Opt_data_err_abort},
1734         {"ignore",      Opt_data_err_ignore},
1735         {}
1736 };
1737
1738 static const struct constant_table ext4_param_jqfmt[] = {
1739         {"vfsold",      QFMT_VFS_OLD},
1740         {"vfsv0",       QFMT_VFS_V0},
1741         {"vfsv1",       QFMT_VFS_V1},
1742         {}
1743 };
1744
1745 static const struct constant_table ext4_param_dax[] = {
1746         {"always",      Opt_dax_always},
1747         {"inode",       Opt_dax_inode},
1748         {"never",       Opt_dax_never},
1749         {}
1750 };
1751
1752 /* String parameter that allows empty argument */
1753 #define fsparam_string_empty(NAME, OPT) \
1754         __fsparam(fs_param_is_string, NAME, OPT, fs_param_can_be_empty, NULL)
1755
1756 /*
1757  * Mount option specification
1758  * We don't use fsparam_flag_no because of the way we set the
1759  * options and the way we show them in _ext4_show_options(). To
1760  * keep the changes to a minimum, let's keep the negative options
1761  * separate for now.
1762  */
1763 static const struct fs_parameter_spec ext4_param_specs[] = {
1764         fsparam_flag    ("bsddf",               Opt_bsd_df),
1765         fsparam_flag    ("minixdf",             Opt_minix_df),
1766         fsparam_flag    ("grpid",               Opt_grpid),
1767         fsparam_flag    ("bsdgroups",           Opt_grpid),
1768         fsparam_flag    ("nogrpid",             Opt_nogrpid),
1769         fsparam_flag    ("sysvgroups",          Opt_nogrpid),
1770         fsparam_u32     ("resgid",              Opt_resgid),
1771         fsparam_u32     ("resuid",              Opt_resuid),
1772         fsparam_u32     ("sb",                  Opt_sb),
1773         fsparam_enum    ("errors",              Opt_errors, ext4_param_errors),
1774         fsparam_flag    ("nouid32",             Opt_nouid32),
1775         fsparam_flag    ("debug",               Opt_debug),
1776         fsparam_flag    ("oldalloc",            Opt_removed),
1777         fsparam_flag    ("orlov",               Opt_removed),
1778         fsparam_flag    ("user_xattr",          Opt_user_xattr),
1779         fsparam_flag    ("nouser_xattr",        Opt_nouser_xattr),
1780         fsparam_flag    ("acl",                 Opt_acl),
1781         fsparam_flag    ("noacl",               Opt_noacl),
1782         fsparam_flag    ("norecovery",          Opt_noload),
1783         fsparam_flag    ("noload",              Opt_noload),
1784         fsparam_flag    ("bh",                  Opt_removed),
1785         fsparam_flag    ("nobh",                Opt_removed),
1786         fsparam_u32     ("commit",              Opt_commit),
1787         fsparam_u32     ("min_batch_time",      Opt_min_batch_time),
1788         fsparam_u32     ("max_batch_time",      Opt_max_batch_time),
1789         fsparam_u32     ("journal_dev",         Opt_journal_dev),
1790         fsparam_bdev    ("journal_path",        Opt_journal_path),
1791         fsparam_flag    ("journal_checksum",    Opt_journal_checksum),
1792         fsparam_flag    ("nojournal_checksum",  Opt_nojournal_checksum),
1793         fsparam_flag    ("journal_async_commit",Opt_journal_async_commit),
1794         fsparam_flag    ("abort",               Opt_abort),
1795         fsparam_enum    ("data",                Opt_data, ext4_param_data),
1796         fsparam_enum    ("data_err",            Opt_data_err,
1797                                                 ext4_param_data_err),
1798         fsparam_string_empty
1799                         ("usrjquota",           Opt_usrjquota),
1800         fsparam_string_empty
1801                         ("grpjquota",           Opt_grpjquota),
1802         fsparam_enum    ("jqfmt",               Opt_jqfmt, ext4_param_jqfmt),
1803         fsparam_flag    ("grpquota",            Opt_grpquota),
1804         fsparam_flag    ("quota",               Opt_quota),
1805         fsparam_flag    ("noquota",             Opt_noquota),
1806         fsparam_flag    ("usrquota",            Opt_usrquota),
1807         fsparam_flag    ("prjquota",            Opt_prjquota),
1808         fsparam_flag    ("barrier",             Opt_barrier),
1809         fsparam_u32     ("barrier",             Opt_barrier),
1810         fsparam_flag    ("nobarrier",           Opt_nobarrier),
1811         fsparam_flag    ("i_version",           Opt_i_version),
1812         fsparam_flag    ("dax",                 Opt_dax),
1813         fsparam_enum    ("dax",                 Opt_dax_type, ext4_param_dax),
1814         fsparam_u32     ("stripe",              Opt_stripe),
1815         fsparam_flag    ("delalloc",            Opt_delalloc),
1816         fsparam_flag    ("nodelalloc",          Opt_nodelalloc),
1817         fsparam_flag    ("warn_on_error",       Opt_warn_on_error),
1818         fsparam_flag    ("nowarn_on_error",     Opt_nowarn_on_error),
1819         fsparam_u32     ("debug_want_extra_isize",
1820                                                 Opt_debug_want_extra_isize),
1821         fsparam_flag    ("mblk_io_submit",      Opt_removed),
1822         fsparam_flag    ("nomblk_io_submit",    Opt_removed),
1823         fsparam_flag    ("block_validity",      Opt_block_validity),
1824         fsparam_flag    ("noblock_validity",    Opt_noblock_validity),
1825         fsparam_u32     ("inode_readahead_blks",
1826                                                 Opt_inode_readahead_blks),
1827         fsparam_u32     ("journal_ioprio",      Opt_journal_ioprio),
1828         fsparam_u32     ("auto_da_alloc",       Opt_auto_da_alloc),
1829         fsparam_flag    ("auto_da_alloc",       Opt_auto_da_alloc),
1830         fsparam_flag    ("noauto_da_alloc",     Opt_noauto_da_alloc),
1831         fsparam_flag    ("dioread_nolock",      Opt_dioread_nolock),
1832         fsparam_flag    ("nodioread_nolock",    Opt_dioread_lock),
1833         fsparam_flag    ("dioread_lock",        Opt_dioread_lock),
1834         fsparam_flag    ("discard",             Opt_discard),
1835         fsparam_flag    ("nodiscard",           Opt_nodiscard),
1836         fsparam_u32     ("init_itable",         Opt_init_itable),
1837         fsparam_flag    ("init_itable",         Opt_init_itable),
1838         fsparam_flag    ("noinit_itable",       Opt_noinit_itable),
1839 #ifdef CONFIG_EXT4_DEBUG
1840         fsparam_flag    ("fc_debug_force",      Opt_fc_debug_force),
1841         fsparam_u32     ("fc_debug_max_replay", Opt_fc_debug_max_replay),
1842 #endif
1843         fsparam_u32     ("max_dir_size_kb",     Opt_max_dir_size_kb),
1844         fsparam_flag    ("test_dummy_encryption",
1845                                                 Opt_test_dummy_encryption),
1846         fsparam_string  ("test_dummy_encryption",
1847                                                 Opt_test_dummy_encryption),
1848         fsparam_flag    ("inlinecrypt",         Opt_inlinecrypt),
1849         fsparam_flag    ("nombcache",           Opt_nombcache),
1850         fsparam_flag    ("no_mbcache",          Opt_nombcache), /* for backward compatibility */
1851         fsparam_flag    ("prefetch_block_bitmaps",
1852                                                 Opt_removed),
1853         fsparam_flag    ("no_prefetch_block_bitmaps",
1854                                                 Opt_no_prefetch_block_bitmaps),
1855         fsparam_s32     ("mb_optimize_scan",    Opt_mb_optimize_scan),
1856         fsparam_string  ("check",               Opt_removed),   /* mount option from ext2/3 */
1857         fsparam_flag    ("nocheck",             Opt_removed),   /* mount option from ext2/3 */
1858         fsparam_flag    ("reservation",         Opt_removed),   /* mount option from ext2/3 */
1859         fsparam_flag    ("noreservation",       Opt_removed),   /* mount option from ext2/3 */
1860         fsparam_u32     ("journal",             Opt_removed),   /* mount option from ext2/3 */
1861         {}
1862 };
1863
1864 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1865 #define DEFAULT_MB_OPTIMIZE_SCAN        (-1)
1866
1867 static const char deprecated_msg[] =
1868         "Mount option \"%s\" will be removed by %s\n"
1869         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1870
1871 #define MOPT_SET        0x0001
1872 #define MOPT_CLEAR      0x0002
1873 #define MOPT_NOSUPPORT  0x0004
1874 #define MOPT_EXPLICIT   0x0008
1875 #ifdef CONFIG_QUOTA
1876 #define MOPT_Q          0
1877 #define MOPT_QFMT       0x0010
1878 #else
1879 #define MOPT_Q          MOPT_NOSUPPORT
1880 #define MOPT_QFMT       MOPT_NOSUPPORT
1881 #endif
1882 #define MOPT_NO_EXT2    0x0020
1883 #define MOPT_NO_EXT3    0x0040
1884 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1885 #define MOPT_SKIP       0x0080
1886 #define MOPT_2          0x0100
1887
1888 static const struct mount_opts {
1889         int     token;
1890         int     mount_opt;
1891         int     flags;
1892 } ext4_mount_opts[] = {
1893         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1894         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1895         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1896         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1897         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1898         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1899         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1900          MOPT_EXT4_ONLY | MOPT_SET},
1901         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1902          MOPT_EXT4_ONLY | MOPT_CLEAR},
1903         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1904         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1905         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1906          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1907         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1908          MOPT_EXT4_ONLY | MOPT_CLEAR},
1909         {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET},
1910         {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR},
1911         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1912          MOPT_EXT4_ONLY | MOPT_CLEAR},
1913         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1914          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1915         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1916                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1917          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1918         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1919         {Opt_data_err, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_NO_EXT2},
1920         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1921         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1922         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1923         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1924         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1925         {Opt_dax_type, 0, MOPT_EXT4_ONLY},
1926         {Opt_journal_dev, 0, MOPT_NO_EXT2},
1927         {Opt_journal_path, 0, MOPT_NO_EXT2},
1928         {Opt_journal_ioprio, 0, MOPT_NO_EXT2},
1929         {Opt_data, 0, MOPT_NO_EXT2},
1930         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1931         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1932 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1933         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1934         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1935 #else
1936         {Opt_acl, 0, MOPT_NOSUPPORT},
1937         {Opt_noacl, 0, MOPT_NOSUPPORT},
1938 #endif
1939         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1940         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1941         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1942         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1943                                                         MOPT_SET | MOPT_Q},
1944         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1945                                                         MOPT_SET | MOPT_Q},
1946         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1947                                                         MOPT_SET | MOPT_Q},
1948         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1949                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1950                                                         MOPT_CLEAR | MOPT_Q},
1951         {Opt_usrjquota, 0, MOPT_Q},
1952         {Opt_grpjquota, 0, MOPT_Q},
1953         {Opt_jqfmt, 0, MOPT_QFMT},
1954         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1955         {Opt_no_prefetch_block_bitmaps, EXT4_MOUNT_NO_PREFETCH_BLOCK_BITMAPS,
1956          MOPT_SET},
1957 #ifdef CONFIG_EXT4_DEBUG
1958         {Opt_fc_debug_force, EXT4_MOUNT2_JOURNAL_FAST_COMMIT,
1959          MOPT_SET | MOPT_2 | MOPT_EXT4_ONLY},
1960 #endif
1961         {Opt_err, 0, 0}
1962 };
1963
1964 #if IS_ENABLED(CONFIG_UNICODE)
1965 static const struct ext4_sb_encodings {
1966         __u16 magic;
1967         char *name;
1968         unsigned int version;
1969 } ext4_sb_encoding_map[] = {
1970         {EXT4_ENC_UTF8_12_1, "utf8", UNICODE_AGE(12, 1, 0)},
1971 };
1972
1973 static const struct ext4_sb_encodings *
1974 ext4_sb_read_encoding(const struct ext4_super_block *es)
1975 {
1976         __u16 magic = le16_to_cpu(es->s_encoding);
1977         int i;
1978
1979         for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++)
1980                 if (magic == ext4_sb_encoding_map[i].magic)
1981                         return &ext4_sb_encoding_map[i];
1982
1983         return NULL;
1984 }
1985 #endif
1986
1987 static int ext4_set_test_dummy_encryption(struct super_block *sb, char *arg)
1988 {
1989 #ifdef CONFIG_FS_ENCRYPTION
1990         struct ext4_sb_info *sbi = EXT4_SB(sb);
1991         int err;
1992
1993         err = fscrypt_set_test_dummy_encryption(sb, arg,
1994                                                 &sbi->s_dummy_enc_policy);
1995         if (err) {
1996                 ext4_msg(sb, KERN_WARNING,
1997                          "Error while setting test dummy encryption [%d]", err);
1998                 return err;
1999         }
2000         ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled");
2001 #endif
2002         return 0;
2003 }
2004
2005 #define EXT4_SPEC_JQUOTA                        (1 <<  0)
2006 #define EXT4_SPEC_JQFMT                         (1 <<  1)
2007 #define EXT4_SPEC_DATAJ                         (1 <<  2)
2008 #define EXT4_SPEC_SB_BLOCK                      (1 <<  3)
2009 #define EXT4_SPEC_JOURNAL_DEV                   (1 <<  4)
2010 #define EXT4_SPEC_JOURNAL_IOPRIO                (1 <<  5)
2011 #define EXT4_SPEC_DUMMY_ENCRYPTION              (1 <<  6)
2012 #define EXT4_SPEC_s_want_extra_isize            (1 <<  7)
2013 #define EXT4_SPEC_s_max_batch_time              (1 <<  8)
2014 #define EXT4_SPEC_s_min_batch_time              (1 <<  9)
2015 #define EXT4_SPEC_s_inode_readahead_blks        (1 << 10)
2016 #define EXT4_SPEC_s_li_wait_mult                (1 << 11)
2017 #define EXT4_SPEC_s_max_dir_size_kb             (1 << 12)
2018 #define EXT4_SPEC_s_stripe                      (1 << 13)
2019 #define EXT4_SPEC_s_resuid                      (1 << 14)
2020 #define EXT4_SPEC_s_resgid                      (1 << 15)
2021 #define EXT4_SPEC_s_commit_interval             (1 << 16)
2022 #define EXT4_SPEC_s_fc_debug_max_replay         (1 << 17)
2023 #define EXT4_SPEC_s_sb_block                    (1 << 18)
2024 #define EXT4_SPEC_mb_optimize_scan              (1 << 19)
2025
2026 struct ext4_fs_context {
2027         char            *s_qf_names[EXT4_MAXQUOTAS];
2028         char            *test_dummy_enc_arg;
2029         int             s_jquota_fmt;   /* Format of quota to use */
2030 #ifdef CONFIG_EXT4_DEBUG
2031         int s_fc_debug_max_replay;
2032 #endif
2033         unsigned short  qname_spec;
2034         unsigned long   vals_s_flags;   /* Bits to set in s_flags */
2035         unsigned long   mask_s_flags;   /* Bits changed in s_flags */
2036         unsigned long   journal_devnum;
2037         unsigned long   s_commit_interval;
2038         unsigned long   s_stripe;
2039         unsigned int    s_inode_readahead_blks;
2040         unsigned int    s_want_extra_isize;
2041         unsigned int    s_li_wait_mult;
2042         unsigned int    s_max_dir_size_kb;
2043         unsigned int    journal_ioprio;
2044         unsigned int    vals_s_mount_opt;
2045         unsigned int    mask_s_mount_opt;
2046         unsigned int    vals_s_mount_opt2;
2047         unsigned int    mask_s_mount_opt2;
2048         unsigned long   vals_s_mount_flags;
2049         unsigned long   mask_s_mount_flags;
2050         unsigned int    opt_flags;      /* MOPT flags */
2051         unsigned int    spec;
2052         u32             s_max_batch_time;
2053         u32             s_min_batch_time;
2054         kuid_t          s_resuid;
2055         kgid_t          s_resgid;
2056         ext4_fsblk_t    s_sb_block;
2057 };
2058
2059 static void ext4_fc_free(struct fs_context *fc)
2060 {
2061         struct ext4_fs_context *ctx = fc->fs_private;
2062         int i;
2063
2064         if (!ctx)
2065                 return;
2066
2067         for (i = 0; i < EXT4_MAXQUOTAS; i++)
2068                 kfree(ctx->s_qf_names[i]);
2069
2070         kfree(ctx->test_dummy_enc_arg);
2071         kfree(ctx);
2072 }
2073
2074 int ext4_init_fs_context(struct fs_context *fc)
2075 {
2076         struct ext4_fs_context *ctx;
2077
2078         ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL);
2079         if (!ctx)
2080                 return -ENOMEM;
2081
2082         fc->fs_private = ctx;
2083         fc->ops = &ext4_context_ops;
2084
2085         return 0;
2086 }
2087
2088 #ifdef CONFIG_QUOTA
2089 /*
2090  * Note the name of the specified quota file.
2091  */
2092 static int note_qf_name(struct fs_context *fc, int qtype,
2093                        struct fs_parameter *param)
2094 {
2095         struct ext4_fs_context *ctx = fc->fs_private;
2096         char *qname;
2097
2098         if (param->size < 1) {
2099                 ext4_msg(NULL, KERN_ERR, "Missing quota name");
2100                 return -EINVAL;
2101         }
2102         if (strchr(param->string, '/')) {
2103                 ext4_msg(NULL, KERN_ERR,
2104                          "quotafile must be on filesystem root");
2105                 return -EINVAL;
2106         }
2107         if (ctx->s_qf_names[qtype]) {
2108                 if (strcmp(ctx->s_qf_names[qtype], param->string) != 0) {
2109                         ext4_msg(NULL, KERN_ERR,
2110                                  "%s quota file already specified",
2111                                  QTYPE2NAME(qtype));
2112                         return -EINVAL;
2113                 }
2114                 return 0;
2115         }
2116
2117         qname = kmemdup_nul(param->string, param->size, GFP_KERNEL);
2118         if (!qname) {
2119                 ext4_msg(NULL, KERN_ERR,
2120                          "Not enough memory for storing quotafile name");
2121                 return -ENOMEM;
2122         }
2123         ctx->s_qf_names[qtype] = qname;
2124         ctx->qname_spec |= 1 << qtype;
2125         ctx->spec |= EXT4_SPEC_JQUOTA;
2126         return 0;
2127 }
2128
2129 /*
2130  * Clear the name of the specified quota file.
2131  */
2132 static int unnote_qf_name(struct fs_context *fc, int qtype)
2133 {
2134         struct ext4_fs_context *ctx = fc->fs_private;
2135
2136         if (ctx->s_qf_names[qtype])
2137                 kfree(ctx->s_qf_names[qtype]);
2138
2139         ctx->s_qf_names[qtype] = NULL;
2140         ctx->qname_spec |= 1 << qtype;
2141         ctx->spec |= EXT4_SPEC_JQUOTA;
2142         return 0;
2143 }
2144 #endif
2145
2146 #define EXT4_SET_CTX(name)                                              \
2147 static inline void ctx_set_##name(struct ext4_fs_context *ctx,          \
2148                                   unsigned long flag)                   \
2149 {                                                                       \
2150         ctx->mask_s_##name |= flag;                                     \
2151         ctx->vals_s_##name |= flag;                                     \
2152 }
2153
2154 #define EXT4_CLEAR_CTX(name)                                            \
2155 static inline void ctx_clear_##name(struct ext4_fs_context *ctx,        \
2156                                     unsigned long flag)                 \
2157 {                                                                       \
2158         ctx->mask_s_##name |= flag;                                     \
2159         ctx->vals_s_##name &= ~flag;                                    \
2160 }
2161
2162 #define EXT4_TEST_CTX(name)                                             \
2163 static inline unsigned long                                             \
2164 ctx_test_##name(struct ext4_fs_context *ctx, unsigned long flag)        \
2165 {                                                                       \
2166         return (ctx->vals_s_##name & flag);                             \
2167 }
2168
2169 EXT4_SET_CTX(flags); /* set only */
2170 EXT4_SET_CTX(mount_opt);
2171 EXT4_CLEAR_CTX(mount_opt);
2172 EXT4_TEST_CTX(mount_opt);
2173 EXT4_SET_CTX(mount_opt2);
2174 EXT4_CLEAR_CTX(mount_opt2);
2175 EXT4_TEST_CTX(mount_opt2);
2176
2177 static inline void ctx_set_mount_flag(struct ext4_fs_context *ctx, int bit)
2178 {
2179         set_bit(bit, &ctx->mask_s_mount_flags);
2180         set_bit(bit, &ctx->vals_s_mount_flags);
2181 }
2182
2183 static int ext4_parse_param(struct fs_context *fc, struct fs_parameter *param)
2184 {
2185         struct ext4_fs_context *ctx = fc->fs_private;
2186         struct fs_parse_result result;
2187         const struct mount_opts *m;
2188         int is_remount;
2189         kuid_t uid;
2190         kgid_t gid;
2191         int token;
2192
2193         token = fs_parse(fc, ext4_param_specs, param, &result);
2194         if (token < 0)
2195                 return token;
2196         is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE;
2197
2198         for (m = ext4_mount_opts; m->token != Opt_err; m++)
2199                 if (token == m->token)
2200                         break;
2201
2202         ctx->opt_flags |= m->flags;
2203
2204         if (m->flags & MOPT_EXPLICIT) {
2205                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
2206                         ctx_set_mount_opt2(ctx, EXT4_MOUNT2_EXPLICIT_DELALLOC);
2207                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
2208                         ctx_set_mount_opt2(ctx,
2209                                        EXT4_MOUNT2_EXPLICIT_JOURNAL_CHECKSUM);
2210                 } else
2211                         return -EINVAL;
2212         }
2213
2214         if (m->flags & MOPT_NOSUPPORT) {
2215                 ext4_msg(NULL, KERN_ERR, "%s option not supported",
2216                          param->key);
2217                 return 0;
2218         }
2219
2220         switch (token) {
2221 #ifdef CONFIG_QUOTA
2222         case Opt_usrjquota:
2223                 if (!*param->string)
2224                         return unnote_qf_name(fc, USRQUOTA);
2225                 else
2226                         return note_qf_name(fc, USRQUOTA, param);
2227         case Opt_grpjquota:
2228                 if (!*param->string)
2229                         return unnote_qf_name(fc, GRPQUOTA);
2230                 else
2231                         return note_qf_name(fc, GRPQUOTA, param);
2232 #endif
2233         case Opt_noacl:
2234         case Opt_nouser_xattr:
2235                 ext4_msg(NULL, KERN_WARNING, deprecated_msg, param->key, "3.5");
2236                 break;
2237         case Opt_sb:
2238                 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
2239                         ext4_msg(NULL, KERN_WARNING,
2240                                  "Ignoring %s option on remount", param->key);
2241                 } else {
2242                         ctx->s_sb_block = result.uint_32;
2243                         ctx->spec |= EXT4_SPEC_s_sb_block;
2244                 }
2245                 return 0;
2246         case Opt_removed:
2247                 ext4_msg(NULL, KERN_WARNING, "Ignoring removed %s option",
2248                          param->key);
2249                 return 0;
2250         case Opt_abort:
2251                 ctx_set_mount_flag(ctx, EXT4_MF_FS_ABORTED);
2252                 return 0;
2253         case Opt_i_version:
2254                 ext4_msg(NULL, KERN_WARNING, deprecated_msg, param->key, "5.20");
2255                 ext4_msg(NULL, KERN_WARNING, "Use iversion instead\n");
2256                 ctx_set_flags(ctx, SB_I_VERSION);
2257                 return 0;
2258         case Opt_inlinecrypt:
2259 #ifdef CONFIG_FS_ENCRYPTION_INLINE_CRYPT
2260                 ctx_set_flags(ctx, SB_INLINECRYPT);
2261 #else
2262                 ext4_msg(NULL, KERN_ERR, "inline encryption not supported");
2263 #endif
2264                 return 0;
2265         case Opt_errors:
2266                 ctx_clear_mount_opt(ctx, EXT4_MOUNT_ERRORS_MASK);
2267                 ctx_set_mount_opt(ctx, result.uint_32);
2268                 return 0;
2269 #ifdef CONFIG_QUOTA
2270         case Opt_jqfmt:
2271                 ctx->s_jquota_fmt = result.uint_32;
2272                 ctx->spec |= EXT4_SPEC_JQFMT;
2273                 return 0;
2274 #endif
2275         case Opt_data:
2276                 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS);
2277                 ctx_set_mount_opt(ctx, result.uint_32);
2278                 ctx->spec |= EXT4_SPEC_DATAJ;
2279                 return 0;
2280         case Opt_commit:
2281                 if (result.uint_32 == 0)
2282                         ctx->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE;
2283                 else if (result.uint_32 > INT_MAX / HZ) {
2284                         ext4_msg(NULL, KERN_ERR,
2285                                  "Invalid commit interval %d, "
2286                                  "must be smaller than %d",
2287                                  result.uint_32, INT_MAX / HZ);
2288                         return -EINVAL;
2289                 }
2290                 ctx->s_commit_interval = HZ * result.uint_32;
2291                 ctx->spec |= EXT4_SPEC_s_commit_interval;
2292                 return 0;
2293         case Opt_debug_want_extra_isize:
2294                 if ((result.uint_32 & 1) || (result.uint_32 < 4)) {
2295                         ext4_msg(NULL, KERN_ERR,
2296                                  "Invalid want_extra_isize %d", result.uint_32);
2297                         return -EINVAL;
2298                 }
2299                 ctx->s_want_extra_isize = result.uint_32;
2300                 ctx->spec |= EXT4_SPEC_s_want_extra_isize;
2301                 return 0;
2302         case Opt_max_batch_time:
2303                 ctx->s_max_batch_time = result.uint_32;
2304                 ctx->spec |= EXT4_SPEC_s_max_batch_time;
2305                 return 0;
2306         case Opt_min_batch_time:
2307                 ctx->s_min_batch_time = result.uint_32;
2308                 ctx->spec |= EXT4_SPEC_s_min_batch_time;
2309                 return 0;
2310         case Opt_inode_readahead_blks:
2311                 if (result.uint_32 &&
2312                     (result.uint_32 > (1 << 30) ||
2313                      !is_power_of_2(result.uint_32))) {
2314                         ext4_msg(NULL, KERN_ERR,
2315                                  "EXT4-fs: inode_readahead_blks must be "
2316                                  "0 or a power of 2 smaller than 2^31");
2317                         return -EINVAL;
2318                 }
2319                 ctx->s_inode_readahead_blks = result.uint_32;
2320                 ctx->spec |= EXT4_SPEC_s_inode_readahead_blks;
2321                 return 0;
2322         case Opt_init_itable:
2323                 ctx_set_mount_opt(ctx, EXT4_MOUNT_INIT_INODE_TABLE);
2324                 ctx->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
2325                 if (param->type == fs_value_is_string)
2326                         ctx->s_li_wait_mult = result.uint_32;
2327                 ctx->spec |= EXT4_SPEC_s_li_wait_mult;
2328                 return 0;
2329         case Opt_max_dir_size_kb:
2330                 ctx->s_max_dir_size_kb = result.uint_32;
2331                 ctx->spec |= EXT4_SPEC_s_max_dir_size_kb;
2332                 return 0;
2333 #ifdef CONFIG_EXT4_DEBUG
2334         case Opt_fc_debug_max_replay:
2335                 ctx->s_fc_debug_max_replay = result.uint_32;
2336                 ctx->spec |= EXT4_SPEC_s_fc_debug_max_replay;
2337                 return 0;
2338 #endif
2339         case Opt_stripe:
2340                 ctx->s_stripe = result.uint_32;
2341                 ctx->spec |= EXT4_SPEC_s_stripe;
2342                 return 0;
2343         case Opt_resuid:
2344                 uid = make_kuid(current_user_ns(), result.uint_32);
2345                 if (!uid_valid(uid)) {
2346                         ext4_msg(NULL, KERN_ERR, "Invalid uid value %d",
2347                                  result.uint_32);
2348                         return -EINVAL;
2349                 }
2350                 ctx->s_resuid = uid;
2351                 ctx->spec |= EXT4_SPEC_s_resuid;
2352                 return 0;
2353         case Opt_resgid:
2354                 gid = make_kgid(current_user_ns(), result.uint_32);
2355                 if (!gid_valid(gid)) {
2356                         ext4_msg(NULL, KERN_ERR, "Invalid gid value %d",
2357                                  result.uint_32);
2358                         return -EINVAL;
2359                 }
2360                 ctx->s_resgid = gid;
2361                 ctx->spec |= EXT4_SPEC_s_resgid;
2362                 return 0;
2363         case Opt_journal_dev:
2364                 if (is_remount) {
2365                         ext4_msg(NULL, KERN_ERR,
2366                                  "Cannot specify journal on remount");
2367                         return -EINVAL;
2368                 }
2369                 ctx->journal_devnum = result.uint_32;
2370                 ctx->spec |= EXT4_SPEC_JOURNAL_DEV;
2371                 return 0;
2372         case Opt_journal_path:
2373         {
2374                 struct inode *journal_inode;
2375                 struct path path;
2376                 int error;
2377
2378                 if (is_remount) {
2379                         ext4_msg(NULL, KERN_ERR,
2380                                  "Cannot specify journal on remount");
2381                         return -EINVAL;
2382                 }
2383
2384                 error = fs_lookup_param(fc, param, 1, &path);
2385                 if (error) {
2386                         ext4_msg(NULL, KERN_ERR, "error: could not find "
2387                                  "journal device path");
2388                         return -EINVAL;
2389                 }
2390
2391                 journal_inode = d_inode(path.dentry);
2392                 ctx->journal_devnum = new_encode_dev(journal_inode->i_rdev);
2393                 ctx->spec |= EXT4_SPEC_JOURNAL_DEV;
2394                 path_put(&path);
2395                 return 0;
2396         }
2397         case Opt_journal_ioprio:
2398                 if (result.uint_32 > 7) {
2399                         ext4_msg(NULL, KERN_ERR, "Invalid journal IO priority"
2400                                  " (must be 0-7)");
2401                         return -EINVAL;
2402                 }
2403                 ctx->journal_ioprio =
2404                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, result.uint_32);
2405                 ctx->spec |= EXT4_SPEC_JOURNAL_IOPRIO;
2406                 return 0;
2407         case Opt_test_dummy_encryption:
2408 #ifdef CONFIG_FS_ENCRYPTION
2409                 if (param->type == fs_value_is_flag) {
2410                         ctx->spec |= EXT4_SPEC_DUMMY_ENCRYPTION;
2411                         ctx->test_dummy_enc_arg = NULL;
2412                         return 0;
2413                 }
2414                 if (*param->string &&
2415                     !(!strcmp(param->string, "v1") ||
2416                       !strcmp(param->string, "v2"))) {
2417                         ext4_msg(NULL, KERN_WARNING,
2418                                  "Value of option \"%s\" is unrecognized",
2419                                  param->key);
2420                         return -EINVAL;
2421                 }
2422                 ctx->spec |= EXT4_SPEC_DUMMY_ENCRYPTION;
2423                 ctx->test_dummy_enc_arg = kmemdup_nul(param->string, param->size,
2424                                                       GFP_KERNEL);
2425 #else
2426                 ext4_msg(NULL, KERN_WARNING,
2427                          "Test dummy encryption mount option ignored");
2428 #endif
2429                 return 0;
2430         case Opt_dax:
2431         case Opt_dax_type:
2432 #ifdef CONFIG_FS_DAX
2433         {
2434                 int type = (token == Opt_dax) ?
2435                            Opt_dax : result.uint_32;
2436
2437                 switch (type) {
2438                 case Opt_dax:
2439                 case Opt_dax_always:
2440                         ctx_set_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS);
2441                         ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER);
2442                         break;
2443                 case Opt_dax_never:
2444                         ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER);
2445                         ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS);
2446                         break;
2447                 case Opt_dax_inode:
2448                         ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS);
2449                         ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER);
2450                         /* Strictly for printing options */
2451                         ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE);
2452                         break;
2453                 }
2454                 return 0;
2455         }
2456 #else
2457                 ext4_msg(NULL, KERN_INFO, "dax option not supported");
2458                 return -EINVAL;
2459 #endif
2460         case Opt_data_err:
2461                 if (result.uint_32 == Opt_data_err_abort)
2462                         ctx_set_mount_opt(ctx, m->mount_opt);
2463                 else if (result.uint_32 == Opt_data_err_ignore)
2464                         ctx_clear_mount_opt(ctx, m->mount_opt);
2465                 return 0;
2466         case Opt_mb_optimize_scan:
2467                 if (result.int_32 == 1) {
2468                         ctx_set_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN);
2469                         ctx->spec |= EXT4_SPEC_mb_optimize_scan;
2470                 } else if (result.int_32 == 0) {
2471                         ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN);
2472                         ctx->spec |= EXT4_SPEC_mb_optimize_scan;
2473                 } else {
2474                         ext4_msg(NULL, KERN_WARNING,
2475                                  "mb_optimize_scan should be set to 0 or 1.");
2476                         return -EINVAL;
2477                 }
2478                 return 0;
2479         }
2480
2481         /*
2482          * At this point we should only be getting options requiring MOPT_SET,
2483          * or MOPT_CLEAR. Anything else is a bug
2484          */
2485         if (m->token == Opt_err) {
2486                 ext4_msg(NULL, KERN_WARNING, "buggy handling of option %s",
2487                          param->key);
2488                 WARN_ON(1);
2489                 return -EINVAL;
2490         }
2491
2492         else {
2493                 unsigned int set = 0;
2494
2495                 if ((param->type == fs_value_is_flag) ||
2496                     result.uint_32 > 0)
2497                         set = 1;
2498
2499                 if (m->flags & MOPT_CLEAR)
2500                         set = !set;
2501                 else if (unlikely(!(m->flags & MOPT_SET))) {
2502                         ext4_msg(NULL, KERN_WARNING,
2503                                  "buggy handling of option %s",
2504                                  param->key);
2505                         WARN_ON(1);
2506                         return -EINVAL;
2507                 }
2508                 if (m->flags & MOPT_2) {
2509                         if (set != 0)
2510                                 ctx_set_mount_opt2(ctx, m->mount_opt);
2511                         else
2512                                 ctx_clear_mount_opt2(ctx, m->mount_opt);
2513                 } else {
2514                         if (set != 0)
2515                                 ctx_set_mount_opt(ctx, m->mount_opt);
2516                         else
2517                                 ctx_clear_mount_opt(ctx, m->mount_opt);
2518                 }
2519         }
2520
2521         return 0;
2522 }
2523
2524 static int parse_options(struct fs_context *fc, char *options)
2525 {
2526         struct fs_parameter param;
2527         int ret;
2528         char *key;
2529
2530         if (!options)
2531                 return 0;
2532
2533         while ((key = strsep(&options, ",")) != NULL) {
2534                 if (*key) {
2535                         size_t v_len = 0;
2536                         char *value = strchr(key, '=');
2537
2538                         param.type = fs_value_is_flag;
2539                         param.string = NULL;
2540
2541                         if (value) {
2542                                 if (value == key)
2543                                         continue;
2544
2545                                 *value++ = 0;
2546                                 v_len = strlen(value);
2547                                 param.string = kmemdup_nul(value, v_len,
2548                                                            GFP_KERNEL);
2549                                 if (!param.string)
2550                                         return -ENOMEM;
2551                                 param.type = fs_value_is_string;
2552                         }
2553
2554                         param.key = key;
2555                         param.size = v_len;
2556
2557                         ret = ext4_parse_param(fc, &param);
2558                         if (param.string)
2559                                 kfree(param.string);
2560                         if (ret < 0)
2561                                 return ret;
2562                 }
2563         }
2564
2565         ret = ext4_validate_options(fc);
2566         if (ret < 0)
2567                 return ret;
2568
2569         return 0;
2570 }
2571
2572 static int parse_apply_sb_mount_options(struct super_block *sb,
2573                                         struct ext4_fs_context *m_ctx)
2574 {
2575         struct ext4_sb_info *sbi = EXT4_SB(sb);
2576         char *s_mount_opts = NULL;
2577         struct ext4_fs_context *s_ctx = NULL;
2578         struct fs_context *fc = NULL;
2579         int ret = -ENOMEM;
2580
2581         if (!sbi->s_es->s_mount_opts[0])
2582                 return 0;
2583
2584         s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
2585                                 sizeof(sbi->s_es->s_mount_opts),
2586                                 GFP_KERNEL);
2587         if (!s_mount_opts)
2588                 return ret;
2589
2590         fc = kzalloc(sizeof(struct fs_context), GFP_KERNEL);
2591         if (!fc)
2592                 goto out_free;
2593
2594         s_ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL);
2595         if (!s_ctx)
2596                 goto out_free;
2597
2598         fc->fs_private = s_ctx;
2599         fc->s_fs_info = sbi;
2600
2601         ret = parse_options(fc, s_mount_opts);
2602         if (ret < 0)
2603                 goto parse_failed;
2604
2605         ret = ext4_check_opt_consistency(fc, sb);
2606         if (ret < 0) {
2607 parse_failed:
2608                 ext4_msg(sb, KERN_WARNING,
2609                          "failed to parse options in superblock: %s",
2610                          s_mount_opts);
2611                 ret = 0;
2612                 goto out_free;
2613         }
2614
2615         if (s_ctx->spec & EXT4_SPEC_JOURNAL_DEV)
2616                 m_ctx->journal_devnum = s_ctx->journal_devnum;
2617         if (s_ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)
2618                 m_ctx->journal_ioprio = s_ctx->journal_ioprio;
2619
2620         ret = ext4_apply_options(fc, sb);
2621
2622 out_free:
2623         kfree(s_ctx);
2624         kfree(fc);
2625         kfree(s_mount_opts);
2626         return ret;
2627 }
2628
2629 static void ext4_apply_quota_options(struct fs_context *fc,
2630                                      struct super_block *sb)
2631 {
2632 #ifdef CONFIG_QUOTA
2633         bool quota_feature = ext4_has_feature_quota(sb);
2634         struct ext4_fs_context *ctx = fc->fs_private;
2635         struct ext4_sb_info *sbi = EXT4_SB(sb);
2636         char *qname;
2637         int i;
2638
2639         if (quota_feature)
2640                 return;
2641
2642         if (ctx->spec & EXT4_SPEC_JQUOTA) {
2643                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2644                         if (!(ctx->qname_spec & (1 << i)))
2645                                 continue;
2646
2647                         qname = ctx->s_qf_names[i]; /* May be NULL */
2648                         if (qname)
2649                                 set_opt(sb, QUOTA);
2650                         ctx->s_qf_names[i] = NULL;
2651                         qname = rcu_replace_pointer(sbi->s_qf_names[i], qname,
2652                                                 lockdep_is_held(&sb->s_umount));
2653                         if (qname)
2654                                 kfree_rcu(qname);
2655                 }
2656         }
2657
2658         if (ctx->spec & EXT4_SPEC_JQFMT)
2659                 sbi->s_jquota_fmt = ctx->s_jquota_fmt;
2660 #endif
2661 }
2662
2663 /*
2664  * Check quota settings consistency.
2665  */
2666 static int ext4_check_quota_consistency(struct fs_context *fc,
2667                                         struct super_block *sb)
2668 {
2669 #ifdef CONFIG_QUOTA
2670         struct ext4_fs_context *ctx = fc->fs_private;
2671         struct ext4_sb_info *sbi = EXT4_SB(sb);
2672         bool quota_feature = ext4_has_feature_quota(sb);
2673         bool quota_loaded = sb_any_quota_loaded(sb);
2674         bool usr_qf_name, grp_qf_name, usrquota, grpquota;
2675         int quota_flags, i;
2676
2677         /*
2678          * We do the test below only for project quotas. 'usrquota' and
2679          * 'grpquota' mount options are allowed even without quota feature
2680          * to support legacy quotas in quota files.
2681          */
2682         if (ctx_test_mount_opt(ctx, EXT4_MOUNT_PRJQUOTA) &&
2683             !ext4_has_feature_project(sb)) {
2684                 ext4_msg(NULL, KERN_ERR, "Project quota feature not enabled. "
2685                          "Cannot enable project quota enforcement.");
2686                 return -EINVAL;
2687         }
2688
2689         quota_flags = EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
2690                       EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA;
2691         if (quota_loaded &&
2692             ctx->mask_s_mount_opt & quota_flags &&
2693             !ctx_test_mount_opt(ctx, quota_flags))
2694                 goto err_quota_change;
2695
2696         if (ctx->spec & EXT4_SPEC_JQUOTA) {
2697
2698                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2699                         if (!(ctx->qname_spec & (1 << i)))
2700                                 continue;
2701
2702                         if (quota_loaded &&
2703                             !!sbi->s_qf_names[i] != !!ctx->s_qf_names[i])
2704                                 goto err_jquota_change;
2705
2706                         if (sbi->s_qf_names[i] && ctx->s_qf_names[i] &&
2707                             strcmp(get_qf_name(sb, sbi, i),
2708                                    ctx->s_qf_names[i]) != 0)
2709                                 goto err_jquota_specified;
2710                 }
2711
2712                 if (quota_feature) {
2713                         ext4_msg(NULL, KERN_INFO,
2714                                  "Journaled quota options ignored when "
2715                                  "QUOTA feature is enabled");
2716                         return 0;
2717                 }
2718         }
2719
2720         if (ctx->spec & EXT4_SPEC_JQFMT) {
2721                 if (sbi->s_jquota_fmt != ctx->s_jquota_fmt && quota_loaded)
2722                         goto err_jquota_change;
2723                 if (quota_feature) {
2724                         ext4_msg(NULL, KERN_INFO, "Quota format mount options "
2725                                  "ignored when QUOTA feature is enabled");
2726                         return 0;
2727                 }
2728         }
2729
2730         /* Make sure we don't mix old and new quota format */
2731         usr_qf_name = (get_qf_name(sb, sbi, USRQUOTA) ||
2732                        ctx->s_qf_names[USRQUOTA]);
2733         grp_qf_name = (get_qf_name(sb, sbi, GRPQUOTA) ||
2734                        ctx->s_qf_names[GRPQUOTA]);
2735
2736         usrquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) ||
2737                     test_opt(sb, USRQUOTA));
2738
2739         grpquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) ||
2740                     test_opt(sb, GRPQUOTA));
2741
2742         if (usr_qf_name) {
2743                 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA);
2744                 usrquota = false;
2745         }
2746         if (grp_qf_name) {
2747                 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA);
2748                 grpquota = false;
2749         }
2750
2751         if (usr_qf_name || grp_qf_name) {
2752                 if (usrquota || grpquota) {
2753                         ext4_msg(NULL, KERN_ERR, "old and new quota "
2754                                  "format mixing");
2755                         return -EINVAL;
2756                 }
2757
2758                 if (!(ctx->spec & EXT4_SPEC_JQFMT || sbi->s_jquota_fmt)) {
2759                         ext4_msg(NULL, KERN_ERR, "journaled quota format "
2760                                  "not specified");
2761                         return -EINVAL;
2762                 }
2763         }
2764
2765         return 0;
2766
2767 err_quota_change:
2768         ext4_msg(NULL, KERN_ERR,
2769                  "Cannot change quota options when quota turned on");
2770         return -EINVAL;
2771 err_jquota_change:
2772         ext4_msg(NULL, KERN_ERR, "Cannot change journaled quota "
2773                  "options when quota turned on");
2774         return -EINVAL;
2775 err_jquota_specified:
2776         ext4_msg(NULL, KERN_ERR, "%s quota file already specified",
2777                  QTYPE2NAME(i));
2778         return -EINVAL;
2779 #else
2780         return 0;
2781 #endif
2782 }
2783
2784 static int ext4_check_opt_consistency(struct fs_context *fc,
2785                                       struct super_block *sb)
2786 {
2787         struct ext4_fs_context *ctx = fc->fs_private;
2788         struct ext4_sb_info *sbi = fc->s_fs_info;
2789         int is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE;
2790
2791         if ((ctx->opt_flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
2792                 ext4_msg(NULL, KERN_ERR,
2793                          "Mount option(s) incompatible with ext2");
2794                 return -EINVAL;
2795         }
2796         if ((ctx->opt_flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
2797                 ext4_msg(NULL, KERN_ERR,
2798                          "Mount option(s) incompatible with ext3");
2799                 return -EINVAL;
2800         }
2801
2802         if (ctx->s_want_extra_isize >
2803             (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE)) {
2804                 ext4_msg(NULL, KERN_ERR,
2805                          "Invalid want_extra_isize %d",
2806                          ctx->s_want_extra_isize);
2807                 return -EINVAL;
2808         }
2809
2810         if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DIOREAD_NOLOCK)) {
2811                 int blocksize =
2812                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
2813                 if (blocksize < PAGE_SIZE)
2814                         ext4_msg(NULL, KERN_WARNING, "Warning: mounting with an "
2815                                  "experimental mount option 'dioread_nolock' "
2816                                  "for blocksize < PAGE_SIZE");
2817         }
2818
2819 #ifdef CONFIG_FS_ENCRYPTION
2820         /*
2821          * This mount option is just for testing, and it's not worthwhile to
2822          * implement the extra complexity (e.g. RCU protection) that would be
2823          * needed to allow it to be set or changed during remount.  We do allow
2824          * it to be specified during remount, but only if there is no change.
2825          */
2826         if ((ctx->spec & EXT4_SPEC_DUMMY_ENCRYPTION) &&
2827             is_remount && !sbi->s_dummy_enc_policy.policy) {
2828                 ext4_msg(NULL, KERN_WARNING,
2829                          "Can't set test_dummy_encryption on remount");
2830                 return -1;
2831         }
2832 #endif
2833
2834         if ((ctx->spec & EXT4_SPEC_DATAJ) && is_remount) {
2835                 if (!sbi->s_journal) {
2836                         ext4_msg(NULL, KERN_WARNING,
2837                                  "Remounting file system with no journal "
2838                                  "so ignoring journalled data option");
2839                         ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS);
2840                 } else if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS) !=
2841                            test_opt(sb, DATA_FLAGS)) {
2842                         ext4_msg(NULL, KERN_ERR, "Cannot change data mode "
2843                                  "on remount");
2844                         return -EINVAL;
2845                 }
2846         }
2847
2848         if (is_remount) {
2849                 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) &&
2850                     (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) {
2851                         ext4_msg(NULL, KERN_ERR, "can't mount with "
2852                                  "both data=journal and dax");
2853                         return -EINVAL;
2854                 }
2855
2856                 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) &&
2857                     (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2858                      (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) {
2859 fail_dax_change_remount:
2860                         ext4_msg(NULL, KERN_ERR, "can't change "
2861                                  "dax mount option while remounting");
2862                         return -EINVAL;
2863                 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER) &&
2864                          (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2865                           (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS))) {
2866                         goto fail_dax_change_remount;
2867                 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE) &&
2868                            ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2869                             (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2870                             !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE))) {
2871                         goto fail_dax_change_remount;
2872                 }
2873         }
2874
2875         return ext4_check_quota_consistency(fc, sb);
2876 }
2877
2878 static int ext4_apply_options(struct fs_context *fc, struct super_block *sb)
2879 {
2880         struct ext4_fs_context *ctx = fc->fs_private;
2881         struct ext4_sb_info *sbi = fc->s_fs_info;
2882         int ret = 0;
2883
2884         sbi->s_mount_opt &= ~ctx->mask_s_mount_opt;
2885         sbi->s_mount_opt |= ctx->vals_s_mount_opt;
2886         sbi->s_mount_opt2 &= ~ctx->mask_s_mount_opt2;
2887         sbi->s_mount_opt2 |= ctx->vals_s_mount_opt2;
2888         sbi->s_mount_flags &= ~ctx->mask_s_mount_flags;
2889         sbi->s_mount_flags |= ctx->vals_s_mount_flags;
2890         sb->s_flags &= ~ctx->mask_s_flags;
2891         sb->s_flags |= ctx->vals_s_flags;
2892
2893         /*
2894          * i_version differs from common mount option iversion so we have
2895          * to let vfs know that it was set, otherwise it would get cleared
2896          * on remount
2897          */
2898         if (ctx->mask_s_flags & SB_I_VERSION)
2899                 fc->sb_flags |= SB_I_VERSION;
2900
2901 #define APPLY(X) ({ if (ctx->spec & EXT4_SPEC_##X) sbi->X = ctx->X; })
2902         APPLY(s_commit_interval);
2903         APPLY(s_stripe);
2904         APPLY(s_max_batch_time);
2905         APPLY(s_min_batch_time);
2906         APPLY(s_want_extra_isize);
2907         APPLY(s_inode_readahead_blks);
2908         APPLY(s_max_dir_size_kb);
2909         APPLY(s_li_wait_mult);
2910         APPLY(s_resgid);
2911         APPLY(s_resuid);
2912
2913 #ifdef CONFIG_EXT4_DEBUG
2914         APPLY(s_fc_debug_max_replay);
2915 #endif
2916
2917         ext4_apply_quota_options(fc, sb);
2918
2919         if (ctx->spec & EXT4_SPEC_DUMMY_ENCRYPTION)
2920                 ret = ext4_set_test_dummy_encryption(sb, ctx->test_dummy_enc_arg);
2921
2922         return ret;
2923 }
2924
2925
2926 static int ext4_validate_options(struct fs_context *fc)
2927 {
2928 #ifdef CONFIG_QUOTA
2929         struct ext4_fs_context *ctx = fc->fs_private;
2930         char *usr_qf_name, *grp_qf_name;
2931
2932         usr_qf_name = ctx->s_qf_names[USRQUOTA];
2933         grp_qf_name = ctx->s_qf_names[GRPQUOTA];
2934
2935         if (usr_qf_name || grp_qf_name) {
2936                 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) && usr_qf_name)
2937                         ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA);
2938
2939                 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) && grp_qf_name)
2940                         ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA);
2941
2942                 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) ||
2943                     ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA)) {
2944                         ext4_msg(NULL, KERN_ERR, "old and new quota "
2945                                  "format mixing");
2946                         return -EINVAL;
2947                 }
2948         }
2949 #endif
2950         return 1;
2951 }
2952
2953 static inline void ext4_show_quota_options(struct seq_file *seq,
2954                                            struct super_block *sb)
2955 {
2956 #if defined(CONFIG_QUOTA)
2957         struct ext4_sb_info *sbi = EXT4_SB(sb);
2958         char *usr_qf_name, *grp_qf_name;
2959
2960         if (sbi->s_jquota_fmt) {
2961                 char *fmtname = "";
2962
2963                 switch (sbi->s_jquota_fmt) {
2964                 case QFMT_VFS_OLD:
2965                         fmtname = "vfsold";
2966                         break;
2967                 case QFMT_VFS_V0:
2968                         fmtname = "vfsv0";
2969                         break;
2970                 case QFMT_VFS_V1:
2971                         fmtname = "vfsv1";
2972                         break;
2973                 }
2974                 seq_printf(seq, ",jqfmt=%s", fmtname);
2975         }
2976
2977         rcu_read_lock();
2978         usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]);
2979         grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]);
2980         if (usr_qf_name)
2981                 seq_show_option(seq, "usrjquota", usr_qf_name);
2982         if (grp_qf_name)
2983                 seq_show_option(seq, "grpjquota", grp_qf_name);
2984         rcu_read_unlock();
2985 #endif
2986 }
2987
2988 static const char *token2str(int token)
2989 {
2990         const struct fs_parameter_spec *spec;
2991
2992         for (spec = ext4_param_specs; spec->name != NULL; spec++)
2993                 if (spec->opt == token && !spec->type)
2994                         break;
2995         return spec->name;
2996 }
2997
2998 /*
2999  * Show an option if
3000  *  - it's set to a non-default value OR
3001  *  - if the per-sb default is different from the global default
3002  */
3003 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
3004                               int nodefs)
3005 {
3006         struct ext4_sb_info *sbi = EXT4_SB(sb);
3007         struct ext4_super_block *es = sbi->s_es;
3008         int def_errors, def_mount_opt = sbi->s_def_mount_opt;
3009         const struct mount_opts *m;
3010         char sep = nodefs ? '\n' : ',';
3011
3012 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
3013 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
3014
3015         if (sbi->s_sb_block != 1)
3016                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
3017
3018         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
3019                 int want_set = m->flags & MOPT_SET;
3020                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
3021                     m->flags & MOPT_SKIP)
3022                         continue;
3023                 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
3024                         continue; /* skip if same as the default */
3025                 if ((want_set &&
3026                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
3027                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
3028                         continue; /* select Opt_noFoo vs Opt_Foo */
3029                 SEQ_OPTS_PRINT("%s", token2str(m->token));
3030         }
3031
3032         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
3033             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
3034                 SEQ_OPTS_PRINT("resuid=%u",
3035                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
3036         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
3037             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
3038                 SEQ_OPTS_PRINT("resgid=%u",
3039                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
3040         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
3041         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
3042                 SEQ_OPTS_PUTS("errors=remount-ro");
3043         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
3044                 SEQ_OPTS_PUTS("errors=continue");
3045         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
3046                 SEQ_OPTS_PUTS("errors=panic");
3047         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
3048                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
3049         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
3050                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
3051         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
3052                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
3053         if (sb->s_flags & SB_I_VERSION)
3054                 SEQ_OPTS_PUTS("i_version");
3055         if (nodefs || sbi->s_stripe)
3056                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
3057         if (nodefs || EXT4_MOUNT_DATA_FLAGS &
3058                         (sbi->s_mount_opt ^ def_mount_opt)) {
3059                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3060                         SEQ_OPTS_PUTS("data=journal");
3061                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3062                         SEQ_OPTS_PUTS("data=ordered");
3063                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
3064                         SEQ_OPTS_PUTS("data=writeback");
3065         }
3066         if (nodefs ||
3067             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
3068                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
3069                                sbi->s_inode_readahead_blks);
3070
3071         if (test_opt(sb, INIT_INODE_TABLE) && (nodefs ||
3072                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
3073                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
3074         if (nodefs || sbi->s_max_dir_size_kb)
3075                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
3076         if (test_opt(sb, DATA_ERR_ABORT))
3077                 SEQ_OPTS_PUTS("data_err=abort");
3078
3079         fscrypt_show_test_dummy_encryption(seq, sep, sb);
3080
3081         if (sb->s_flags & SB_INLINECRYPT)
3082                 SEQ_OPTS_PUTS("inlinecrypt");
3083
3084         if (test_opt(sb, DAX_ALWAYS)) {
3085                 if (IS_EXT2_SB(sb))
3086                         SEQ_OPTS_PUTS("dax");
3087                 else
3088                         SEQ_OPTS_PUTS("dax=always");
3089         } else if (test_opt2(sb, DAX_NEVER)) {
3090                 SEQ_OPTS_PUTS("dax=never");
3091         } else if (test_opt2(sb, DAX_INODE)) {
3092                 SEQ_OPTS_PUTS("dax=inode");
3093         }
3094         ext4_show_quota_options(seq, sb);
3095         return 0;
3096 }
3097
3098 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
3099 {
3100         return _ext4_show_options(seq, root->d_sb, 0);
3101 }
3102
3103 int ext4_seq_options_show(struct seq_file *seq, void *offset)
3104 {
3105         struct super_block *sb = seq->private;
3106         int rc;
3107
3108         seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
3109         rc = _ext4_show_options(seq, sb, 1);
3110         seq_puts(seq, "\n");
3111         return rc;
3112 }
3113
3114 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
3115                             int read_only)
3116 {
3117         struct ext4_sb_info *sbi = EXT4_SB(sb);
3118         int err = 0;
3119
3120         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
3121                 ext4_msg(sb, KERN_ERR, "revision level too high, "
3122                          "forcing read-only mode");
3123                 err = -EROFS;
3124                 goto done;
3125         }
3126         if (read_only)
3127                 goto done;
3128         if (!(sbi->s_mount_state & EXT4_VALID_FS))
3129                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
3130                          "running e2fsck is recommended");
3131         else if (sbi->s_mount_state & EXT4_ERROR_FS)
3132                 ext4_msg(sb, KERN_WARNING,
3133                          "warning: mounting fs with errors, "
3134                          "running e2fsck is recommended");
3135         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
3136                  le16_to_cpu(es->s_mnt_count) >=
3137                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
3138                 ext4_msg(sb, KERN_WARNING,
3139                          "warning: maximal mount count reached, "
3140                          "running e2fsck is recommended");
3141         else if (le32_to_cpu(es->s_checkinterval) &&
3142                  (ext4_get_tstamp(es, s_lastcheck) +
3143                   le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds()))
3144                 ext4_msg(sb, KERN_WARNING,
3145                          "warning: checktime reached, "
3146                          "running e2fsck is recommended");
3147         if (!sbi->s_journal)
3148                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
3149         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
3150                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
3151         le16_add_cpu(&es->s_mnt_count, 1);
3152         ext4_update_tstamp(es, s_mtime);
3153         if (sbi->s_journal) {
3154                 ext4_set_feature_journal_needs_recovery(sb);
3155                 if (ext4_has_feature_orphan_file(sb))
3156                         ext4_set_feature_orphan_present(sb);
3157         }
3158
3159         err = ext4_commit_super(sb);
3160 done:
3161         if (test_opt(sb, DEBUG))
3162                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
3163                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
3164                         sb->s_blocksize,
3165                         sbi->s_groups_count,
3166                         EXT4_BLOCKS_PER_GROUP(sb),
3167                         EXT4_INODES_PER_GROUP(sb),
3168                         sbi->s_mount_opt, sbi->s_mount_opt2);
3169         return err;
3170 }
3171
3172 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
3173 {
3174         struct ext4_sb_info *sbi = EXT4_SB(sb);
3175         struct flex_groups **old_groups, **new_groups;
3176         int size, i, j;
3177
3178         if (!sbi->s_log_groups_per_flex)
3179                 return 0;
3180
3181         size = ext4_flex_group(sbi, ngroup - 1) + 1;
3182         if (size <= sbi->s_flex_groups_allocated)
3183                 return 0;
3184
3185         new_groups = kvzalloc(roundup_pow_of_two(size *
3186                               sizeof(*sbi->s_flex_groups)), GFP_KERNEL);
3187         if (!new_groups) {
3188                 ext4_msg(sb, KERN_ERR,
3189                          "not enough memory for %d flex group pointers", size);
3190                 return -ENOMEM;
3191         }
3192         for (i = sbi->s_flex_groups_allocated; i < size; i++) {
3193                 new_groups[i] = kvzalloc(roundup_pow_of_two(
3194                                          sizeof(struct flex_groups)),
3195                                          GFP_KERNEL);
3196                 if (!new_groups[i]) {
3197                         for (j = sbi->s_flex_groups_allocated; j < i; j++)
3198                                 kvfree(new_groups[j]);
3199                         kvfree(new_groups);
3200                         ext4_msg(sb, KERN_ERR,
3201                                  "not enough memory for %d flex groups", size);
3202                         return -ENOMEM;
3203                 }
3204         }
3205         rcu_read_lock();
3206         old_groups = rcu_dereference(sbi->s_flex_groups);
3207         if (old_groups)
3208                 memcpy(new_groups, old_groups,
3209                        (sbi->s_flex_groups_allocated *
3210                         sizeof(struct flex_groups *)));
3211         rcu_read_unlock();
3212         rcu_assign_pointer(sbi->s_flex_groups, new_groups);
3213         sbi->s_flex_groups_allocated = size;
3214         if (old_groups)
3215                 ext4_kvfree_array_rcu(old_groups);
3216         return 0;
3217 }
3218
3219 static int ext4_fill_flex_info(struct super_block *sb)
3220 {
3221         struct ext4_sb_info *sbi = EXT4_SB(sb);
3222         struct ext4_group_desc *gdp = NULL;
3223         struct flex_groups *fg;
3224         ext4_group_t flex_group;
3225         int i, err;
3226
3227         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
3228         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
3229                 sbi->s_log_groups_per_flex = 0;
3230                 return 1;
3231         }
3232
3233         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
3234         if (err)
3235                 goto failed;
3236
3237         for (i = 0; i < sbi->s_groups_count; i++) {
3238                 gdp = ext4_get_group_desc(sb, i, NULL);
3239
3240                 flex_group = ext4_flex_group(sbi, i);
3241                 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
3242                 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes);
3243                 atomic64_add(ext4_free_group_clusters(sb, gdp),
3244                              &fg->free_clusters);
3245                 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs);
3246         }
3247
3248         return 1;
3249 failed:
3250         return 0;
3251 }
3252
3253 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
3254                                    struct ext4_group_desc *gdp)
3255 {
3256         int offset = offsetof(struct ext4_group_desc, bg_checksum);
3257         __u16 crc = 0;
3258         __le32 le_group = cpu_to_le32(block_group);
3259         struct ext4_sb_info *sbi = EXT4_SB(sb);
3260
3261         if (ext4_has_metadata_csum(sbi->s_sb)) {
3262                 /* Use new metadata_csum algorithm */
3263                 __u32 csum32;
3264                 __u16 dummy_csum = 0;
3265
3266                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
3267                                      sizeof(le_group));
3268                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
3269                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
3270                                      sizeof(dummy_csum));
3271                 offset += sizeof(dummy_csum);
3272                 if (offset < sbi->s_desc_size)
3273                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
3274                                              sbi->s_desc_size - offset);
3275
3276                 crc = csum32 & 0xFFFF;
3277                 goto out;
3278         }
3279
3280         /* old crc16 code */
3281         if (!ext4_has_feature_gdt_csum(sb))
3282                 return 0;
3283
3284         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
3285         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
3286         crc = crc16(crc, (__u8 *)gdp, offset);
3287         offset += sizeof(gdp->bg_checksum); /* skip checksum */
3288         /* for checksum of struct ext4_group_desc do the rest...*/
3289         if (ext4_has_feature_64bit(sb) &&
3290             offset < le16_to_cpu(sbi->s_es->s_desc_size))
3291                 crc = crc16(crc, (__u8 *)gdp + offset,
3292                             le16_to_cpu(sbi->s_es->s_desc_size) -
3293                                 offset);
3294
3295 out:
3296         return cpu_to_le16(crc);
3297 }
3298
3299 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
3300                                 struct ext4_group_desc *gdp)
3301 {
3302         if (ext4_has_group_desc_csum(sb) &&
3303             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
3304                 return 0;
3305
3306         return 1;
3307 }
3308
3309 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
3310                               struct ext4_group_desc *gdp)
3311 {
3312         if (!ext4_has_group_desc_csum(sb))
3313                 return;
3314         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
3315 }
3316
3317 /* Called at mount-time, super-block is locked */
3318 static int ext4_check_descriptors(struct super_block *sb,
3319                                   ext4_fsblk_t sb_block,
3320                                   ext4_group_t *first_not_zeroed)
3321 {
3322         struct ext4_sb_info *sbi = EXT4_SB(sb);
3323         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
3324         ext4_fsblk_t last_block;
3325         ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0);
3326         ext4_fsblk_t block_bitmap;
3327         ext4_fsblk_t inode_bitmap;
3328         ext4_fsblk_t inode_table;
3329         int flexbg_flag = 0;
3330         ext4_group_t i, grp = sbi->s_groups_count;
3331
3332         if (ext4_has_feature_flex_bg(sb))
3333                 flexbg_flag = 1;
3334
3335         ext4_debug("Checking group descriptors");
3336
3337         for (i = 0; i < sbi->s_groups_count; i++) {
3338                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
3339
3340                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
3341                         last_block = ext4_blocks_count(sbi->s_es) - 1;
3342                 else
3343                         last_block = first_block +
3344                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
3345
3346                 if ((grp == sbi->s_groups_count) &&
3347                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3348                         grp = i;
3349
3350                 block_bitmap = ext4_block_bitmap(sb, gdp);
3351                 if (block_bitmap == sb_block) {
3352                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3353                                  "Block bitmap for group %u overlaps "
3354                                  "superblock", i);
3355                         if (!sb_rdonly(sb))
3356                                 return 0;
3357                 }
3358                 if (block_bitmap >= sb_block + 1 &&
3359                     block_bitmap <= last_bg_block) {
3360                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3361                                  "Block bitmap for group %u overlaps "
3362                                  "block group descriptors", i);
3363                         if (!sb_rdonly(sb))
3364                                 return 0;
3365                 }
3366                 if (block_bitmap < first_block || block_bitmap > last_block) {
3367                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3368                                "Block bitmap for group %u not in group "
3369                                "(block %llu)!", i, block_bitmap);
3370                         return 0;
3371                 }
3372                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
3373                 if (inode_bitmap == sb_block) {
3374                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3375                                  "Inode bitmap for group %u overlaps "
3376                                  "superblock", i);
3377                         if (!sb_rdonly(sb))
3378                                 return 0;
3379                 }
3380                 if (inode_bitmap >= sb_block + 1 &&
3381                     inode_bitmap <= last_bg_block) {
3382                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3383                                  "Inode bitmap for group %u overlaps "
3384                                  "block group descriptors", i);
3385                         if (!sb_rdonly(sb))
3386                                 return 0;
3387                 }
3388                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
3389                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3390                                "Inode bitmap for group %u not in group "
3391                                "(block %llu)!", i, inode_bitmap);
3392                         return 0;
3393                 }
3394                 inode_table = ext4_inode_table(sb, gdp);
3395                 if (inode_table == sb_block) {
3396                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3397                                  "Inode table for group %u overlaps "
3398                                  "superblock", i);
3399                         if (!sb_rdonly(sb))
3400                                 return 0;
3401                 }
3402                 if (inode_table >= sb_block + 1 &&
3403                     inode_table <= last_bg_block) {
3404                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3405                                  "Inode table for group %u overlaps "
3406                                  "block group descriptors", i);
3407                         if (!sb_rdonly(sb))
3408                                 return 0;
3409                 }
3410                 if (inode_table < first_block ||
3411                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
3412                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3413                                "Inode table for group %u not in group "
3414                                "(block %llu)!", i, inode_table);
3415                         return 0;
3416                 }
3417                 ext4_lock_group(sb, i);
3418                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
3419                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3420                                  "Checksum for group %u failed (%u!=%u)",
3421                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
3422                                      gdp)), le16_to_cpu(gdp->bg_checksum));
3423                         if (!sb_rdonly(sb)) {
3424                                 ext4_unlock_group(sb, i);
3425                                 return 0;
3426                         }
3427                 }
3428                 ext4_unlock_group(sb, i);
3429                 if (!flexbg_flag)
3430                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
3431         }
3432         if (NULL != first_not_zeroed)
3433                 *first_not_zeroed = grp;
3434         return 1;
3435 }
3436
3437 /*
3438  * Maximal extent format file size.
3439  * Resulting logical blkno at s_maxbytes must fit in our on-disk
3440  * extent format containers, within a sector_t, and within i_blocks
3441  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
3442  * so that won't be a limiting factor.
3443  *
3444  * However there is other limiting factor. We do store extents in the form
3445  * of starting block and length, hence the resulting length of the extent
3446  * covering maximum file size must fit into on-disk format containers as
3447  * well. Given that length is always by 1 unit bigger than max unit (because
3448  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
3449  *
3450  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
3451  */
3452 static loff_t ext4_max_size(int blkbits, int has_huge_files)
3453 {
3454         loff_t res;
3455         loff_t upper_limit = MAX_LFS_FILESIZE;
3456
3457         BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64));
3458
3459         if (!has_huge_files) {
3460                 upper_limit = (1LL << 32) - 1;
3461
3462                 /* total blocks in file system block size */
3463                 upper_limit >>= (blkbits - 9);
3464                 upper_limit <<= blkbits;
3465         }
3466
3467         /*
3468          * 32-bit extent-start container, ee_block. We lower the maxbytes
3469          * by one fs block, so ee_len can cover the extent of maximum file
3470          * size
3471          */
3472         res = (1LL << 32) - 1;
3473         res <<= blkbits;
3474
3475         /* Sanity check against vm- & vfs- imposed limits */
3476         if (res > upper_limit)
3477                 res = upper_limit;
3478
3479         return res;
3480 }
3481
3482 /*
3483  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
3484  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
3485  * We need to be 1 filesystem block less than the 2^48 sector limit.
3486  */
3487 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
3488 {
3489         loff_t upper_limit, res = EXT4_NDIR_BLOCKS;
3490         int meta_blocks;
3491         unsigned int ppb = 1 << (bits - 2);
3492
3493         /*
3494          * This is calculated to be the largest file size for a dense, block
3495          * mapped file such that the file's total number of 512-byte sectors,
3496          * including data and all indirect blocks, does not exceed (2^48 - 1).
3497          *
3498          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
3499          * number of 512-byte sectors of the file.
3500          */
3501         if (!has_huge_files) {
3502                 /*
3503                  * !has_huge_files or implies that the inode i_block field
3504                  * represents total file blocks in 2^32 512-byte sectors ==
3505                  * size of vfs inode i_blocks * 8
3506                  */
3507                 upper_limit = (1LL << 32) - 1;
3508
3509                 /* total blocks in file system block size */
3510                 upper_limit >>= (bits - 9);
3511
3512         } else {
3513                 /*
3514                  * We use 48 bit ext4_inode i_blocks
3515                  * With EXT4_HUGE_FILE_FL set the i_blocks
3516                  * represent total number of blocks in
3517                  * file system block size
3518                  */
3519                 upper_limit = (1LL << 48) - 1;
3520
3521         }
3522
3523         /* Compute how many blocks we can address by block tree */
3524         res += ppb;
3525         res += ppb * ppb;
3526         res += ((loff_t)ppb) * ppb * ppb;
3527         /* Compute how many metadata blocks are needed */
3528         meta_blocks = 1;
3529         meta_blocks += 1 + ppb;
3530         meta_blocks += 1 + ppb + ppb * ppb;
3531         /* Does block tree limit file size? */
3532         if (res + meta_blocks <= upper_limit)
3533                 goto check_lfs;
3534
3535         res = upper_limit;
3536         /* How many metadata blocks are needed for addressing upper_limit? */
3537         upper_limit -= EXT4_NDIR_BLOCKS;
3538         /* indirect blocks */
3539         meta_blocks = 1;
3540         upper_limit -= ppb;
3541         /* double indirect blocks */
3542         if (upper_limit < ppb * ppb) {
3543                 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb);
3544                 res -= meta_blocks;
3545                 goto check_lfs;
3546         }
3547         meta_blocks += 1 + ppb;
3548         upper_limit -= ppb * ppb;
3549         /* tripple indirect blocks for the rest */
3550         meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb) +
3551                 DIV_ROUND_UP_ULL(upper_limit, ppb*ppb);
3552         res -= meta_blocks;
3553 check_lfs:
3554         res <<= bits;
3555         if (res > MAX_LFS_FILESIZE)
3556                 res = MAX_LFS_FILESIZE;
3557
3558         return res;
3559 }
3560
3561 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
3562                                    ext4_fsblk_t logical_sb_block, int nr)
3563 {
3564         struct ext4_sb_info *sbi = EXT4_SB(sb);
3565         ext4_group_t bg, first_meta_bg;
3566         int has_super = 0;
3567
3568         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
3569
3570         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
3571                 return logical_sb_block + nr + 1;
3572         bg = sbi->s_desc_per_block * nr;
3573         if (ext4_bg_has_super(sb, bg))
3574                 has_super = 1;
3575
3576         /*
3577          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
3578          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
3579          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
3580          * compensate.
3581          */
3582         if (sb->s_blocksize == 1024 && nr == 0 &&
3583             le32_to_cpu(sbi->s_es->s_first_data_block) == 0)
3584                 has_super++;
3585
3586         return (has_super + ext4_group_first_block_no(sb, bg));
3587 }
3588
3589 /**
3590  * ext4_get_stripe_size: Get the stripe size.
3591  * @sbi: In memory super block info
3592  *
3593  * If we have specified it via mount option, then
3594  * use the mount option value. If the value specified at mount time is
3595  * greater than the blocks per group use the super block value.
3596  * If the super block value is greater than blocks per group return 0.
3597  * Allocator needs it be less than blocks per group.
3598  *
3599  */
3600 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
3601 {
3602         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
3603         unsigned long stripe_width =
3604                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
3605         int ret;
3606
3607         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
3608                 ret = sbi->s_stripe;
3609         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
3610                 ret = stripe_width;
3611         else if (stride && stride <= sbi->s_blocks_per_group)
3612                 ret = stride;
3613         else
3614                 ret = 0;
3615
3616         /*
3617          * If the stripe width is 1, this makes no sense and
3618          * we set it to 0 to turn off stripe handling code.
3619          */
3620         if (ret <= 1)
3621                 ret = 0;
3622
3623         return ret;
3624 }
3625
3626 /*
3627  * Check whether this filesystem can be mounted based on
3628  * the features present and the RDONLY/RDWR mount requested.
3629  * Returns 1 if this filesystem can be mounted as requested,
3630  * 0 if it cannot be.
3631  */
3632 int ext4_feature_set_ok(struct super_block *sb, int readonly)
3633 {
3634         if (ext4_has_unknown_ext4_incompat_features(sb)) {
3635                 ext4_msg(sb, KERN_ERR,
3636                         "Couldn't mount because of "
3637                         "unsupported optional features (%x)",
3638                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
3639                         ~EXT4_FEATURE_INCOMPAT_SUPP));
3640                 return 0;
3641         }
3642
3643 #if !IS_ENABLED(CONFIG_UNICODE)
3644         if (ext4_has_feature_casefold(sb)) {
3645                 ext4_msg(sb, KERN_ERR,
3646                          "Filesystem with casefold feature cannot be "
3647                          "mounted without CONFIG_UNICODE");
3648                 return 0;
3649         }
3650 #endif
3651
3652         if (readonly)
3653                 return 1;
3654
3655         if (ext4_has_feature_readonly(sb)) {
3656                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
3657                 sb->s_flags |= SB_RDONLY;
3658                 return 1;
3659         }
3660
3661         /* Check that feature set is OK for a read-write mount */
3662         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
3663                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
3664                          "unsupported optional features (%x)",
3665                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
3666                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
3667                 return 0;
3668         }
3669         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
3670                 ext4_msg(sb, KERN_ERR,
3671                          "Can't support bigalloc feature without "
3672                          "extents feature\n");
3673                 return 0;
3674         }
3675
3676 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2)
3677         if (!readonly && (ext4_has_feature_quota(sb) ||
3678                           ext4_has_feature_project(sb))) {
3679                 ext4_msg(sb, KERN_ERR,
3680                          "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2");
3681                 return 0;
3682         }
3683 #endif  /* CONFIG_QUOTA */
3684         return 1;
3685 }
3686
3687 /*
3688  * This function is called once a day if we have errors logged
3689  * on the file system
3690  */
3691 static void print_daily_error_info(struct timer_list *t)
3692 {
3693         struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report);
3694         struct super_block *sb = sbi->s_sb;
3695         struct ext4_super_block *es = sbi->s_es;
3696
3697         if (es->s_error_count)
3698                 /* fsck newer than v1.41.13 is needed to clean this condition. */
3699                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
3700                          le32_to_cpu(es->s_error_count));
3701         if (es->s_first_error_time) {
3702                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d",
3703                        sb->s_id,
3704                        ext4_get_tstamp(es, s_first_error_time),
3705                        (int) sizeof(es->s_first_error_func),
3706                        es->s_first_error_func,
3707                        le32_to_cpu(es->s_first_error_line));
3708                 if (es->s_first_error_ino)
3709                         printk(KERN_CONT ": inode %u",
3710                                le32_to_cpu(es->s_first_error_ino));
3711                 if (es->s_first_error_block)
3712                         printk(KERN_CONT ": block %llu", (unsigned long long)
3713                                le64_to_cpu(es->s_first_error_block));
3714                 printk(KERN_CONT "\n");
3715         }
3716         if (es->s_last_error_time) {
3717                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d",
3718                        sb->s_id,
3719                        ext4_get_tstamp(es, s_last_error_time),
3720                        (int) sizeof(es->s_last_error_func),
3721                        es->s_last_error_func,
3722                        le32_to_cpu(es->s_last_error_line));
3723                 if (es->s_last_error_ino)
3724                         printk(KERN_CONT ": inode %u",
3725                                le32_to_cpu(es->s_last_error_ino));
3726                 if (es->s_last_error_block)
3727                         printk(KERN_CONT ": block %llu", (unsigned long long)
3728                                le64_to_cpu(es->s_last_error_block));
3729                 printk(KERN_CONT "\n");
3730         }
3731         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
3732 }
3733
3734 /* Find next suitable group and run ext4_init_inode_table */
3735 static int ext4_run_li_request(struct ext4_li_request *elr)
3736 {
3737         struct ext4_group_desc *gdp = NULL;
3738         struct super_block *sb = elr->lr_super;
3739         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3740         ext4_group_t group = elr->lr_next_group;
3741         unsigned int prefetch_ios = 0;
3742         int ret = 0;
3743         u64 start_time;
3744
3745         if (elr->lr_mode == EXT4_LI_MODE_PREFETCH_BBITMAP) {
3746                 elr->lr_next_group = ext4_mb_prefetch(sb, group,
3747                                 EXT4_SB(sb)->s_mb_prefetch, &prefetch_ios);
3748                 if (prefetch_ios)
3749                         ext4_mb_prefetch_fini(sb, elr->lr_next_group,
3750                                               prefetch_ios);
3751                 trace_ext4_prefetch_bitmaps(sb, group, elr->lr_next_group,
3752                                             prefetch_ios);
3753                 if (group >= elr->lr_next_group) {
3754                         ret = 1;
3755                         if (elr->lr_first_not_zeroed != ngroups &&
3756                             !sb_rdonly(sb) && test_opt(sb, INIT_INODE_TABLE)) {
3757                                 elr->lr_next_group = elr->lr_first_not_zeroed;
3758                                 elr->lr_mode = EXT4_LI_MODE_ITABLE;
3759                                 ret = 0;
3760                         }
3761                 }
3762                 return ret;
3763         }
3764
3765         for (; group < ngroups; group++) {
3766                 gdp = ext4_get_group_desc(sb, group, NULL);
3767                 if (!gdp) {
3768                         ret = 1;
3769                         break;
3770                 }
3771
3772                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3773                         break;
3774         }
3775
3776         if (group >= ngroups)
3777                 ret = 1;
3778
3779         if (!ret) {
3780                 start_time = ktime_get_real_ns();
3781                 ret = ext4_init_inode_table(sb, group,
3782                                             elr->lr_timeout ? 0 : 1);
3783                 trace_ext4_lazy_itable_init(sb, group);
3784                 if (elr->lr_timeout == 0) {
3785                         elr->lr_timeout = nsecs_to_jiffies((ktime_get_real_ns() - start_time) *
3786                                 EXT4_SB(elr->lr_super)->s_li_wait_mult);
3787                 }
3788                 elr->lr_next_sched = jiffies + elr->lr_timeout;
3789                 elr->lr_next_group = group + 1;
3790         }
3791         return ret;
3792 }
3793
3794 /*
3795  * Remove lr_request from the list_request and free the
3796  * request structure. Should be called with li_list_mtx held
3797  */
3798 static void ext4_remove_li_request(struct ext4_li_request *elr)
3799 {
3800         if (!elr)
3801                 return;
3802
3803         list_del(&elr->lr_request);
3804         EXT4_SB(elr->lr_super)->s_li_request = NULL;
3805         kfree(elr);
3806 }
3807
3808 static void ext4_unregister_li_request(struct super_block *sb)
3809 {
3810         mutex_lock(&ext4_li_mtx);
3811         if (!ext4_li_info) {
3812                 mutex_unlock(&ext4_li_mtx);
3813                 return;
3814         }
3815
3816         mutex_lock(&ext4_li_info->li_list_mtx);
3817         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
3818         mutex_unlock(&ext4_li_info->li_list_mtx);
3819         mutex_unlock(&ext4_li_mtx);
3820 }
3821
3822 static struct task_struct *ext4_lazyinit_task;
3823
3824 /*
3825  * This is the function where ext4lazyinit thread lives. It walks
3826  * through the request list searching for next scheduled filesystem.
3827  * When such a fs is found, run the lazy initialization request
3828  * (ext4_rn_li_request) and keep track of the time spend in this
3829  * function. Based on that time we compute next schedule time of
3830  * the request. When walking through the list is complete, compute
3831  * next waking time and put itself into sleep.
3832  */
3833 static int ext4_lazyinit_thread(void *arg)
3834 {
3835         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
3836         struct list_head *pos, *n;
3837         struct ext4_li_request *elr;
3838         unsigned long next_wakeup, cur;
3839
3840         BUG_ON(NULL == eli);
3841
3842 cont_thread:
3843         while (true) {
3844                 next_wakeup = MAX_JIFFY_OFFSET;
3845
3846                 mutex_lock(&eli->li_list_mtx);
3847                 if (list_empty(&eli->li_request_list)) {
3848                         mutex_unlock(&eli->li_list_mtx);
3849                         goto exit_thread;
3850                 }
3851                 list_for_each_safe(pos, n, &eli->li_request_list) {
3852                         int err = 0;
3853                         int progress = 0;
3854                         elr = list_entry(pos, struct ext4_li_request,
3855                                          lr_request);
3856
3857                         if (time_before(jiffies, elr->lr_next_sched)) {
3858                                 if (time_before(elr->lr_next_sched, next_wakeup))
3859                                         next_wakeup = elr->lr_next_sched;
3860                                 continue;
3861                         }
3862                         if (down_read_trylock(&elr->lr_super->s_umount)) {
3863                                 if (sb_start_write_trylock(elr->lr_super)) {
3864                                         progress = 1;
3865                                         /*
3866                                          * We hold sb->s_umount, sb can not
3867                                          * be removed from the list, it is
3868                                          * now safe to drop li_list_mtx
3869                                          */
3870                                         mutex_unlock(&eli->li_list_mtx);
3871                                         err = ext4_run_li_request(elr);
3872                                         sb_end_write(elr->lr_super);
3873                                         mutex_lock(&eli->li_list_mtx);
3874                                         n = pos->next;
3875                                 }
3876                                 up_read((&elr->lr_super->s_umount));
3877                         }
3878                         /* error, remove the lazy_init job */
3879                         if (err) {
3880                                 ext4_remove_li_request(elr);
3881                                 continue;
3882                         }
3883                         if (!progress) {
3884                                 elr->lr_next_sched = jiffies +
3885                                         (prandom_u32()
3886                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3887                         }
3888                         if (time_before(elr->lr_next_sched, next_wakeup))
3889                                 next_wakeup = elr->lr_next_sched;
3890                 }
3891                 mutex_unlock(&eli->li_list_mtx);
3892
3893                 try_to_freeze();
3894
3895                 cur = jiffies;
3896                 if ((time_after_eq(cur, next_wakeup)) ||
3897                     (MAX_JIFFY_OFFSET == next_wakeup)) {
3898                         cond_resched();
3899                         continue;
3900                 }
3901
3902                 schedule_timeout_interruptible(next_wakeup - cur);
3903
3904                 if (kthread_should_stop()) {
3905                         ext4_clear_request_list();
3906                         goto exit_thread;
3907                 }
3908         }
3909
3910 exit_thread:
3911         /*
3912          * It looks like the request list is empty, but we need
3913          * to check it under the li_list_mtx lock, to prevent any
3914          * additions into it, and of course we should lock ext4_li_mtx
3915          * to atomically free the list and ext4_li_info, because at
3916          * this point another ext4 filesystem could be registering
3917          * new one.
3918          */
3919         mutex_lock(&ext4_li_mtx);
3920         mutex_lock(&eli->li_list_mtx);
3921         if (!list_empty(&eli->li_request_list)) {
3922                 mutex_unlock(&eli->li_list_mtx);
3923                 mutex_unlock(&ext4_li_mtx);
3924                 goto cont_thread;
3925         }
3926         mutex_unlock(&eli->li_list_mtx);
3927         kfree(ext4_li_info);
3928         ext4_li_info = NULL;
3929         mutex_unlock(&ext4_li_mtx);
3930
3931         return 0;
3932 }
3933
3934 static void ext4_clear_request_list(void)
3935 {
3936         struct list_head *pos, *n;
3937         struct ext4_li_request *elr;
3938
3939         mutex_lock(&ext4_li_info->li_list_mtx);
3940         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3941                 elr = list_entry(pos, struct ext4_li_request,
3942                                  lr_request);
3943                 ext4_remove_li_request(elr);
3944         }
3945         mutex_unlock(&ext4_li_info->li_list_mtx);
3946 }
3947
3948 static int ext4_run_lazyinit_thread(void)
3949 {
3950         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3951                                          ext4_li_info, "ext4lazyinit");
3952         if (IS_ERR(ext4_lazyinit_task)) {
3953                 int err = PTR_ERR(ext4_lazyinit_task);
3954                 ext4_clear_request_list();
3955                 kfree(ext4_li_info);
3956                 ext4_li_info = NULL;
3957                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3958                                  "initialization thread\n",
3959                                  err);
3960                 return err;
3961         }
3962         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3963         return 0;
3964 }
3965
3966 /*
3967  * Check whether it make sense to run itable init. thread or not.
3968  * If there is at least one uninitialized inode table, return
3969  * corresponding group number, else the loop goes through all
3970  * groups and return total number of groups.
3971  */
3972 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3973 {
3974         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3975         struct ext4_group_desc *gdp = NULL;
3976
3977         if (!ext4_has_group_desc_csum(sb))
3978                 return ngroups;
3979
3980         for (group = 0; group < ngroups; group++) {
3981                 gdp = ext4_get_group_desc(sb, group, NULL);
3982                 if (!gdp)
3983                         continue;
3984
3985                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3986                         break;
3987         }
3988
3989         return group;
3990 }
3991
3992 static int ext4_li_info_new(void)
3993 {
3994         struct ext4_lazy_init *eli = NULL;
3995
3996         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3997         if (!eli)
3998                 return -ENOMEM;
3999
4000         INIT_LIST_HEAD(&eli->li_request_list);
4001         mutex_init(&eli->li_list_mtx);
4002
4003         eli->li_state |= EXT4_LAZYINIT_QUIT;
4004
4005         ext4_li_info = eli;
4006
4007         return 0;
4008 }
4009
4010 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
4011                                             ext4_group_t start)
4012 {
4013         struct ext4_li_request *elr;
4014
4015         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
4016         if (!elr)
4017                 return NULL;
4018
4019         elr->lr_super = sb;
4020         elr->lr_first_not_zeroed = start;
4021         if (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS)) {
4022                 elr->lr_mode = EXT4_LI_MODE_ITABLE;
4023                 elr->lr_next_group = start;
4024         } else {
4025                 elr->lr_mode = EXT4_LI_MODE_PREFETCH_BBITMAP;
4026         }
4027
4028         /*
4029          * Randomize first schedule time of the request to
4030          * spread the inode table initialization requests
4031          * better.
4032          */
4033         elr->lr_next_sched = jiffies + (prandom_u32() %
4034                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
4035         return elr;
4036 }
4037
4038 int ext4_register_li_request(struct super_block *sb,
4039                              ext4_group_t first_not_zeroed)
4040 {
4041         struct ext4_sb_info *sbi = EXT4_SB(sb);
4042         struct ext4_li_request *elr = NULL;
4043         ext4_group_t ngroups = sbi->s_groups_count;
4044         int ret = 0;
4045
4046         mutex_lock(&ext4_li_mtx);
4047         if (sbi->s_li_request != NULL) {
4048                 /*
4049                  * Reset timeout so it can be computed again, because
4050                  * s_li_wait_mult might have changed.
4051                  */
4052                 sbi->s_li_request->lr_timeout = 0;
4053                 goto out;
4054         }
4055
4056         if (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS) &&
4057             (first_not_zeroed == ngroups || sb_rdonly(sb) ||
4058              !test_opt(sb, INIT_INODE_TABLE)))
4059                 goto out;
4060
4061         elr = ext4_li_request_new(sb, first_not_zeroed);
4062         if (!elr) {
4063                 ret = -ENOMEM;
4064                 goto out;
4065         }
4066
4067         if (NULL == ext4_li_info) {
4068                 ret = ext4_li_info_new();
4069                 if (ret)
4070                         goto out;
4071         }
4072
4073         mutex_lock(&ext4_li_info->li_list_mtx);
4074         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
4075         mutex_unlock(&ext4_li_info->li_list_mtx);
4076
4077         sbi->s_li_request = elr;
4078         /*
4079          * set elr to NULL here since it has been inserted to
4080          * the request_list and the removal and free of it is
4081          * handled by ext4_clear_request_list from now on.
4082          */
4083         elr = NULL;
4084
4085         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
4086                 ret = ext4_run_lazyinit_thread();
4087                 if (ret)
4088                         goto out;
4089         }
4090 out:
4091         mutex_unlock(&ext4_li_mtx);
4092         if (ret)
4093                 kfree(elr);
4094         return ret;
4095 }
4096
4097 /*
4098  * We do not need to lock anything since this is called on
4099  * module unload.
4100  */
4101 static void ext4_destroy_lazyinit_thread(void)
4102 {
4103         /*
4104          * If thread exited earlier
4105          * there's nothing to be done.
4106          */
4107         if (!ext4_li_info || !ext4_lazyinit_task)
4108                 return;
4109
4110         kthread_stop(ext4_lazyinit_task);
4111 }
4112
4113 static int set_journal_csum_feature_set(struct super_block *sb)
4114 {
4115         int ret = 1;
4116         int compat, incompat;
4117         struct ext4_sb_info *sbi = EXT4_SB(sb);
4118
4119         if (ext4_has_metadata_csum(sb)) {
4120                 /* journal checksum v3 */
4121                 compat = 0;
4122                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
4123         } else {
4124                 /* journal checksum v1 */
4125                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
4126                 incompat = 0;
4127         }
4128
4129         jbd2_journal_clear_features(sbi->s_journal,
4130                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
4131                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
4132                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
4133         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4134                 ret = jbd2_journal_set_features(sbi->s_journal,
4135                                 compat, 0,
4136                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
4137                                 incompat);
4138         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
4139                 ret = jbd2_journal_set_features(sbi->s_journal,
4140                                 compat, 0,
4141                                 incompat);
4142                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
4143                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
4144         } else {
4145                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
4146                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
4147         }
4148
4149         return ret;
4150 }
4151
4152 /*
4153  * Note: calculating the overhead so we can be compatible with
4154  * historical BSD practice is quite difficult in the face of
4155  * clusters/bigalloc.  This is because multiple metadata blocks from
4156  * different block group can end up in the same allocation cluster.
4157  * Calculating the exact overhead in the face of clustered allocation
4158  * requires either O(all block bitmaps) in memory or O(number of block
4159  * groups**2) in time.  We will still calculate the superblock for
4160  * older file systems --- and if we come across with a bigalloc file
4161  * system with zero in s_overhead_clusters the estimate will be close to
4162  * correct especially for very large cluster sizes --- but for newer
4163  * file systems, it's better to calculate this figure once at mkfs
4164  * time, and store it in the superblock.  If the superblock value is
4165  * present (even for non-bigalloc file systems), we will use it.
4166  */
4167 static int count_overhead(struct super_block *sb, ext4_group_t grp,
4168                           char *buf)
4169 {
4170         struct ext4_sb_info     *sbi = EXT4_SB(sb);
4171         struct ext4_group_desc  *gdp;
4172         ext4_fsblk_t            first_block, last_block, b;
4173         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
4174         int                     s, j, count = 0;
4175
4176         if (!ext4_has_feature_bigalloc(sb))
4177                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
4178                         sbi->s_itb_per_group + 2);
4179
4180         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
4181                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
4182         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
4183         for (i = 0; i < ngroups; i++) {
4184                 gdp = ext4_get_group_desc(sb, i, NULL);
4185                 b = ext4_block_bitmap(sb, gdp);
4186                 if (b >= first_block && b <= last_block) {
4187                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
4188                         count++;
4189                 }
4190                 b = ext4_inode_bitmap(sb, gdp);
4191                 if (b >= first_block && b <= last_block) {
4192                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
4193                         count++;
4194                 }
4195                 b = ext4_inode_table(sb, gdp);
4196                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
4197                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
4198                                 int c = EXT4_B2C(sbi, b - first_block);
4199                                 ext4_set_bit(c, buf);
4200                                 count++;
4201                         }
4202                 if (i != grp)
4203                         continue;
4204                 s = 0;
4205                 if (ext4_bg_has_super(sb, grp)) {
4206                         ext4_set_bit(s++, buf);
4207                         count++;
4208                 }
4209                 j = ext4_bg_num_gdb(sb, grp);
4210                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
4211                         ext4_error(sb, "Invalid number of block group "
4212                                    "descriptor blocks: %d", j);
4213                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
4214                 }
4215                 count += j;
4216                 for (; j > 0; j--)
4217                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
4218         }
4219         if (!count)
4220                 return 0;
4221         return EXT4_CLUSTERS_PER_GROUP(sb) -
4222                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
4223 }
4224
4225 /*
4226  * Compute the overhead and stash it in sbi->s_overhead
4227  */
4228 int ext4_calculate_overhead(struct super_block *sb)
4229 {
4230         struct ext4_sb_info *sbi = EXT4_SB(sb);
4231         struct ext4_super_block *es = sbi->s_es;
4232         struct inode *j_inode;
4233         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
4234         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4235         ext4_fsblk_t overhead = 0;
4236         char *buf = (char *) get_zeroed_page(GFP_NOFS);
4237
4238         if (!buf)
4239                 return -ENOMEM;
4240
4241         /*
4242          * Compute the overhead (FS structures).  This is constant
4243          * for a given filesystem unless the number of block groups
4244          * changes so we cache the previous value until it does.
4245          */
4246
4247         /*
4248          * All of the blocks before first_data_block are overhead
4249          */
4250         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
4251
4252         /*
4253          * Add the overhead found in each block group
4254          */
4255         for (i = 0; i < ngroups; i++) {
4256                 int blks;
4257
4258                 blks = count_overhead(sb, i, buf);
4259                 overhead += blks;
4260                 if (blks)
4261                         memset(buf, 0, PAGE_SIZE);
4262                 cond_resched();
4263         }
4264
4265         /*
4266          * Add the internal journal blocks whether the journal has been
4267          * loaded or not
4268          */
4269         if (sbi->s_journal && !sbi->s_journal_bdev)
4270                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_total_len);
4271         else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) {
4272                 /* j_inum for internal journal is non-zero */
4273                 j_inode = ext4_get_journal_inode(sb, j_inum);
4274                 if (j_inode) {
4275                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
4276                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
4277                         iput(j_inode);
4278                 } else {
4279                         ext4_msg(sb, KERN_ERR, "can't get journal size");
4280                 }
4281         }
4282         sbi->s_overhead = overhead;
4283         smp_wmb();
4284         free_page((unsigned long) buf);
4285         return 0;
4286 }
4287
4288 static void ext4_set_resv_clusters(struct super_block *sb)
4289 {
4290         ext4_fsblk_t resv_clusters;
4291         struct ext4_sb_info *sbi = EXT4_SB(sb);
4292
4293         /*
4294          * There's no need to reserve anything when we aren't using extents.
4295          * The space estimates are exact, there are no unwritten extents,
4296          * hole punching doesn't need new metadata... This is needed especially
4297          * to keep ext2/3 backward compatibility.
4298          */
4299         if (!ext4_has_feature_extents(sb))
4300                 return;
4301         /*
4302          * By default we reserve 2% or 4096 clusters, whichever is smaller.
4303          * This should cover the situations where we can not afford to run
4304          * out of space like for example punch hole, or converting
4305          * unwritten extents in delalloc path. In most cases such
4306          * allocation would require 1, or 2 blocks, higher numbers are
4307          * very rare.
4308          */
4309         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
4310                          sbi->s_cluster_bits);
4311
4312         do_div(resv_clusters, 50);
4313         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
4314
4315         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
4316 }
4317
4318 static const char *ext4_quota_mode(struct super_block *sb)
4319 {
4320 #ifdef CONFIG_QUOTA
4321         if (!ext4_quota_capable(sb))
4322                 return "none";
4323
4324         if (EXT4_SB(sb)->s_journal && ext4_is_quota_journalled(sb))
4325                 return "journalled";
4326         else
4327                 return "writeback";
4328 #else
4329         return "disabled";
4330 #endif
4331 }
4332
4333 static void ext4_setup_csum_trigger(struct super_block *sb,
4334                                     enum ext4_journal_trigger_type type,
4335                                     void (*trigger)(
4336                                         struct jbd2_buffer_trigger_type *type,
4337                                         struct buffer_head *bh,
4338                                         void *mapped_data,
4339                                         size_t size))
4340 {
4341         struct ext4_sb_info *sbi = EXT4_SB(sb);
4342
4343         sbi->s_journal_triggers[type].sb = sb;
4344         sbi->s_journal_triggers[type].tr_triggers.t_frozen = trigger;
4345 }
4346
4347 static void ext4_free_sbi(struct ext4_sb_info *sbi)
4348 {
4349         if (!sbi)
4350                 return;
4351
4352         kfree(sbi->s_blockgroup_lock);
4353         fs_put_dax(sbi->s_daxdev);
4354         kfree(sbi);
4355 }
4356
4357 static struct ext4_sb_info *ext4_alloc_sbi(struct super_block *sb)
4358 {
4359         struct ext4_sb_info *sbi;
4360
4361         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
4362         if (!sbi)
4363                 return NULL;
4364
4365         sbi->s_daxdev = fs_dax_get_by_bdev(sb->s_bdev, &sbi->s_dax_part_off);
4366
4367         sbi->s_blockgroup_lock =
4368                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
4369
4370         if (!sbi->s_blockgroup_lock)
4371                 goto err_out;
4372
4373         sb->s_fs_info = sbi;
4374         sbi->s_sb = sb;
4375         return sbi;
4376 err_out:
4377         fs_put_dax(sbi->s_daxdev);
4378         kfree(sbi);
4379         return NULL;
4380 }
4381
4382 static int __ext4_fill_super(struct fs_context *fc, struct super_block *sb)
4383 {
4384         struct buffer_head *bh, **group_desc;
4385         struct ext4_super_block *es = NULL;
4386         struct ext4_sb_info *sbi = EXT4_SB(sb);
4387         struct flex_groups **flex_groups;
4388         ext4_fsblk_t block;
4389         ext4_fsblk_t logical_sb_block;
4390         unsigned long offset = 0;
4391         unsigned long def_mount_opts;
4392         struct inode *root;
4393         int ret = -ENOMEM;
4394         int blocksize, clustersize;
4395         unsigned int db_count;
4396         unsigned int i;
4397         int needs_recovery, has_huge_files;
4398         __u64 blocks_count;
4399         int err = 0;
4400         ext4_group_t first_not_zeroed;
4401         struct ext4_fs_context *ctx = fc->fs_private;
4402         int silent = fc->sb_flags & SB_SILENT;
4403
4404         /* Set defaults for the variables that will be set during parsing */
4405         ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4406
4407         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
4408         sbi->s_sectors_written_start =
4409                 part_stat_read(sb->s_bdev, sectors[STAT_WRITE]);
4410
4411         /* -EINVAL is default */
4412         ret = -EINVAL;
4413         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
4414         if (!blocksize) {
4415                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
4416                 goto out_fail;
4417         }
4418
4419         /*
4420          * The ext4 superblock will not be buffer aligned for other than 1kB
4421          * block sizes.  We need to calculate the offset from buffer start.
4422          */
4423         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
4424                 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE;
4425                 offset = do_div(logical_sb_block, blocksize);
4426         } else {
4427                 logical_sb_block = sbi->s_sb_block;
4428         }
4429
4430         bh = ext4_sb_bread_unmovable(sb, logical_sb_block);
4431         if (IS_ERR(bh)) {
4432                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
4433                 ret = PTR_ERR(bh);
4434                 goto out_fail;
4435         }
4436         /*
4437          * Note: s_es must be initialized as soon as possible because
4438          *       some ext4 macro-instructions depend on its value
4439          */
4440         es = (struct ext4_super_block *) (bh->b_data + offset);
4441         sbi->s_es = es;
4442         sb->s_magic = le16_to_cpu(es->s_magic);
4443         if (sb->s_magic != EXT4_SUPER_MAGIC)
4444                 goto cantfind_ext4;
4445         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
4446
4447         /* Warn if metadata_csum and gdt_csum are both set. */
4448         if (ext4_has_feature_metadata_csum(sb) &&
4449             ext4_has_feature_gdt_csum(sb))
4450                 ext4_warning(sb, "metadata_csum and uninit_bg are "
4451                              "redundant flags; please run fsck.");
4452
4453         /* Check for a known checksum algorithm */
4454         if (!ext4_verify_csum_type(sb, es)) {
4455                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
4456                          "unknown checksum algorithm.");
4457                 silent = 1;
4458                 goto cantfind_ext4;
4459         }
4460         ext4_setup_csum_trigger(sb, EXT4_JTR_ORPHAN_FILE,
4461                                 ext4_orphan_file_block_trigger);
4462
4463         /* Load the checksum driver */
4464         sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
4465         if (IS_ERR(sbi->s_chksum_driver)) {
4466                 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
4467                 ret = PTR_ERR(sbi->s_chksum_driver);
4468                 sbi->s_chksum_driver = NULL;
4469                 goto failed_mount;
4470         }
4471
4472         /* Check superblock checksum */
4473         if (!ext4_superblock_csum_verify(sb, es)) {
4474                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
4475                          "invalid superblock checksum.  Run e2fsck?");
4476                 silent = 1;
4477                 ret = -EFSBADCRC;
4478                 goto cantfind_ext4;
4479         }
4480
4481         /* Precompute checksum seed for all metadata */
4482         if (ext4_has_feature_csum_seed(sb))
4483                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
4484         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
4485                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
4486                                                sizeof(es->s_uuid));
4487
4488         /* Set defaults before we parse the mount options */
4489         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
4490         set_opt(sb, INIT_INODE_TABLE);
4491         if (def_mount_opts & EXT4_DEFM_DEBUG)
4492                 set_opt(sb, DEBUG);
4493         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
4494                 set_opt(sb, GRPID);
4495         if (def_mount_opts & EXT4_DEFM_UID16)
4496                 set_opt(sb, NO_UID32);
4497         /* xattr user namespace & acls are now defaulted on */
4498         set_opt(sb, XATTR_USER);
4499 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4500         set_opt(sb, POSIX_ACL);
4501 #endif
4502         if (ext4_has_feature_fast_commit(sb))
4503                 set_opt2(sb, JOURNAL_FAST_COMMIT);
4504         /* don't forget to enable journal_csum when metadata_csum is enabled. */
4505         if (ext4_has_metadata_csum(sb))
4506                 set_opt(sb, JOURNAL_CHECKSUM);
4507
4508         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
4509                 set_opt(sb, JOURNAL_DATA);
4510         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
4511                 set_opt(sb, ORDERED_DATA);
4512         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
4513                 set_opt(sb, WRITEBACK_DATA);
4514
4515         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
4516                 set_opt(sb, ERRORS_PANIC);
4517         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
4518                 set_opt(sb, ERRORS_CONT);
4519         else
4520                 set_opt(sb, ERRORS_RO);
4521         /* block_validity enabled by default; disable with noblock_validity */
4522         set_opt(sb, BLOCK_VALIDITY);
4523         if (def_mount_opts & EXT4_DEFM_DISCARD)
4524                 set_opt(sb, DISCARD);
4525
4526         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
4527         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
4528         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
4529         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
4530         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
4531
4532         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
4533                 set_opt(sb, BARRIER);
4534
4535         /*
4536          * enable delayed allocation by default
4537          * Use -o nodelalloc to turn it off
4538          */
4539         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
4540             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
4541                 set_opt(sb, DELALLOC);
4542
4543         /*
4544          * set default s_li_wait_mult for lazyinit, for the case there is
4545          * no mount option specified.
4546          */
4547         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
4548
4549         if (le32_to_cpu(es->s_log_block_size) >
4550             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4551                 ext4_msg(sb, KERN_ERR,
4552                          "Invalid log block size: %u",
4553                          le32_to_cpu(es->s_log_block_size));
4554                 goto failed_mount;
4555         }
4556         if (le32_to_cpu(es->s_log_cluster_size) >
4557             (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4558                 ext4_msg(sb, KERN_ERR,
4559                          "Invalid log cluster size: %u",
4560                          le32_to_cpu(es->s_log_cluster_size));
4561                 goto failed_mount;
4562         }
4563
4564         blocksize = EXT4_MIN_BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
4565
4566         if (blocksize == PAGE_SIZE)
4567                 set_opt(sb, DIOREAD_NOLOCK);
4568
4569         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
4570                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
4571                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
4572         } else {
4573                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
4574                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
4575                 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) {
4576                         ext4_msg(sb, KERN_ERR, "invalid first ino: %u",
4577                                  sbi->s_first_ino);
4578                         goto failed_mount;
4579                 }
4580                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
4581                     (!is_power_of_2(sbi->s_inode_size)) ||
4582                     (sbi->s_inode_size > blocksize)) {
4583                         ext4_msg(sb, KERN_ERR,
4584                                "unsupported inode size: %d",
4585                                sbi->s_inode_size);
4586                         ext4_msg(sb, KERN_ERR, "blocksize: %d", blocksize);
4587                         goto failed_mount;
4588                 }
4589                 /*
4590                  * i_atime_extra is the last extra field available for
4591                  * [acm]times in struct ext4_inode. Checking for that
4592                  * field should suffice to ensure we have extra space
4593                  * for all three.
4594                  */
4595                 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) +
4596                         sizeof(((struct ext4_inode *)0)->i_atime_extra)) {
4597                         sb->s_time_gran = 1;
4598                         sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX;
4599                 } else {
4600                         sb->s_time_gran = NSEC_PER_SEC;
4601                         sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX;
4602                 }
4603                 sb->s_time_min = EXT4_TIMESTAMP_MIN;
4604         }
4605         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
4606                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4607                         EXT4_GOOD_OLD_INODE_SIZE;
4608                 if (ext4_has_feature_extra_isize(sb)) {
4609                         unsigned v, max = (sbi->s_inode_size -
4610                                            EXT4_GOOD_OLD_INODE_SIZE);
4611
4612                         v = le16_to_cpu(es->s_want_extra_isize);
4613                         if (v > max) {
4614                                 ext4_msg(sb, KERN_ERR,
4615                                          "bad s_want_extra_isize: %d", v);
4616                                 goto failed_mount;
4617                         }
4618                         if (sbi->s_want_extra_isize < v)
4619                                 sbi->s_want_extra_isize = v;
4620
4621                         v = le16_to_cpu(es->s_min_extra_isize);
4622                         if (v > max) {
4623                                 ext4_msg(sb, KERN_ERR,
4624                                          "bad s_min_extra_isize: %d", v);
4625                                 goto failed_mount;
4626                         }
4627                         if (sbi->s_want_extra_isize < v)
4628                                 sbi->s_want_extra_isize = v;
4629                 }
4630         }
4631
4632         err = parse_apply_sb_mount_options(sb, ctx);
4633         if (err < 0)
4634                 goto failed_mount;
4635
4636         sbi->s_def_mount_opt = sbi->s_mount_opt;
4637
4638         err = ext4_check_opt_consistency(fc, sb);
4639         if (err < 0)
4640                 goto failed_mount;
4641
4642         err = ext4_apply_options(fc, sb);
4643         if (err < 0)
4644                 goto failed_mount;
4645
4646 #if IS_ENABLED(CONFIG_UNICODE)
4647         if (ext4_has_feature_casefold(sb) && !sb->s_encoding) {
4648                 const struct ext4_sb_encodings *encoding_info;
4649                 struct unicode_map *encoding;
4650                 __u16 encoding_flags = le16_to_cpu(es->s_encoding_flags);
4651
4652                 encoding_info = ext4_sb_read_encoding(es);
4653                 if (!encoding_info) {
4654                         ext4_msg(sb, KERN_ERR,
4655                                  "Encoding requested by superblock is unknown");
4656                         goto failed_mount;
4657                 }
4658
4659                 encoding = utf8_load(encoding_info->version);
4660                 if (IS_ERR(encoding)) {
4661                         ext4_msg(sb, KERN_ERR,
4662                                  "can't mount with superblock charset: %s-%u.%u.%u "
4663                                  "not supported by the kernel. flags: 0x%x.",
4664                                  encoding_info->name,
4665                                  unicode_major(encoding_info->version),
4666                                  unicode_minor(encoding_info->version),
4667                                  unicode_rev(encoding_info->version),
4668                                  encoding_flags);
4669                         goto failed_mount;
4670                 }
4671                 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: "
4672                          "%s-%u.%u.%u with flags 0x%hx", encoding_info->name,
4673                          unicode_major(encoding_info->version),
4674                          unicode_minor(encoding_info->version),
4675                          unicode_rev(encoding_info->version),
4676                          encoding_flags);
4677
4678                 sb->s_encoding = encoding;
4679                 sb->s_encoding_flags = encoding_flags;
4680         }
4681 #endif
4682
4683         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4684                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with data=journal disables delayed allocation, dioread_nolock, O_DIRECT and fast_commit support!\n");
4685                 /* can't mount with both data=journal and dioread_nolock. */
4686                 clear_opt(sb, DIOREAD_NOLOCK);
4687                 clear_opt2(sb, JOURNAL_FAST_COMMIT);
4688                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4689                         ext4_msg(sb, KERN_ERR, "can't mount with "
4690                                  "both data=journal and delalloc");
4691                         goto failed_mount;
4692                 }
4693                 if (test_opt(sb, DAX_ALWAYS)) {
4694                         ext4_msg(sb, KERN_ERR, "can't mount with "
4695                                  "both data=journal and dax");
4696                         goto failed_mount;
4697                 }
4698                 if (ext4_has_feature_encrypt(sb)) {
4699                         ext4_msg(sb, KERN_WARNING,
4700                                  "encrypted files will use data=ordered "
4701                                  "instead of data journaling mode");
4702                 }
4703                 if (test_opt(sb, DELALLOC))
4704                         clear_opt(sb, DELALLOC);
4705         } else {
4706                 sb->s_iflags |= SB_I_CGROUPWB;
4707         }
4708
4709         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
4710                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
4711
4712         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
4713             (ext4_has_compat_features(sb) ||
4714              ext4_has_ro_compat_features(sb) ||
4715              ext4_has_incompat_features(sb)))
4716                 ext4_msg(sb, KERN_WARNING,
4717                        "feature flags set on rev 0 fs, "
4718                        "running e2fsck is recommended");
4719
4720         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
4721                 set_opt2(sb, HURD_COMPAT);
4722                 if (ext4_has_feature_64bit(sb)) {
4723                         ext4_msg(sb, KERN_ERR,
4724                                  "The Hurd can't support 64-bit file systems");
4725                         goto failed_mount;
4726                 }
4727
4728                 /*
4729                  * ea_inode feature uses l_i_version field which is not
4730                  * available in HURD_COMPAT mode.
4731                  */
4732                 if (ext4_has_feature_ea_inode(sb)) {
4733                         ext4_msg(sb, KERN_ERR,
4734                                  "ea_inode feature is not supported for Hurd");
4735                         goto failed_mount;
4736                 }
4737         }
4738
4739         if (IS_EXT2_SB(sb)) {
4740                 if (ext2_feature_set_ok(sb))
4741                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
4742                                  "using the ext4 subsystem");
4743                 else {
4744                         /*
4745                          * If we're probing be silent, if this looks like
4746                          * it's actually an ext[34] filesystem.
4747                          */
4748                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4749                                 goto failed_mount;
4750                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
4751                                  "to feature incompatibilities");
4752                         goto failed_mount;
4753                 }
4754         }
4755
4756         if (IS_EXT3_SB(sb)) {
4757                 if (ext3_feature_set_ok(sb))
4758                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
4759                                  "using the ext4 subsystem");
4760                 else {
4761                         /*
4762                          * If we're probing be silent, if this looks like
4763                          * it's actually an ext4 filesystem.
4764                          */
4765                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4766                                 goto failed_mount;
4767                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
4768                                  "to feature incompatibilities");
4769                         goto failed_mount;
4770                 }
4771         }
4772
4773         /*
4774          * Check feature flags regardless of the revision level, since we
4775          * previously didn't change the revision level when setting the flags,
4776          * so there is a chance incompat flags are set on a rev 0 filesystem.
4777          */
4778         if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
4779                 goto failed_mount;
4780
4781         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
4782                 ext4_msg(sb, KERN_ERR,
4783                          "Number of reserved GDT blocks insanely large: %d",
4784                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
4785                 goto failed_mount;
4786         }
4787
4788         if (sbi->s_daxdev) {
4789                 if (blocksize == PAGE_SIZE)
4790                         set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags);
4791                 else
4792                         ext4_msg(sb, KERN_ERR, "unsupported blocksize for DAX\n");
4793         }
4794
4795         if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) {
4796                 if (ext4_has_feature_inline_data(sb)) {
4797                         ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
4798                                         " that may contain inline data");
4799                         goto failed_mount;
4800                 }
4801                 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) {
4802                         ext4_msg(sb, KERN_ERR,
4803                                 "DAX unsupported by block device.");
4804                         goto failed_mount;
4805                 }
4806         }
4807
4808         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
4809                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
4810                          es->s_encryption_level);
4811                 goto failed_mount;
4812         }
4813
4814         if (sb->s_blocksize != blocksize) {
4815                 /*
4816                  * bh must be released before kill_bdev(), otherwise
4817                  * it won't be freed and its page also. kill_bdev()
4818                  * is called by sb_set_blocksize().
4819                  */
4820                 brelse(bh);
4821                 /* Validate the filesystem blocksize */
4822                 if (!sb_set_blocksize(sb, blocksize)) {
4823                         ext4_msg(sb, KERN_ERR, "bad block size %d",
4824                                         blocksize);
4825                         bh = NULL;
4826                         goto failed_mount;
4827                 }
4828
4829                 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE;
4830                 offset = do_div(logical_sb_block, blocksize);
4831                 bh = ext4_sb_bread_unmovable(sb, logical_sb_block);
4832                 if (IS_ERR(bh)) {
4833                         ext4_msg(sb, KERN_ERR,
4834                                "Can't read superblock on 2nd try");
4835                         ret = PTR_ERR(bh);
4836                         bh = NULL;
4837                         goto failed_mount;
4838                 }
4839                 es = (struct ext4_super_block *)(bh->b_data + offset);
4840                 sbi->s_es = es;
4841                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
4842                         ext4_msg(sb, KERN_ERR,
4843                                "Magic mismatch, very weird!");
4844                         goto failed_mount;
4845                 }
4846         }
4847
4848         has_huge_files = ext4_has_feature_huge_file(sb);
4849         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
4850                                                       has_huge_files);
4851         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
4852
4853         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
4854         if (ext4_has_feature_64bit(sb)) {
4855                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
4856                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
4857                     !is_power_of_2(sbi->s_desc_size)) {
4858                         ext4_msg(sb, KERN_ERR,
4859                                "unsupported descriptor size %lu",
4860                                sbi->s_desc_size);
4861                         goto failed_mount;
4862                 }
4863         } else
4864                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
4865
4866         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
4867         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
4868
4869         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
4870         if (sbi->s_inodes_per_block == 0)
4871                 goto cantfind_ext4;
4872         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
4873             sbi->s_inodes_per_group > blocksize * 8) {
4874                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
4875                          sbi->s_inodes_per_group);
4876                 goto failed_mount;
4877         }
4878         sbi->s_itb_per_group = sbi->s_inodes_per_group /
4879                                         sbi->s_inodes_per_block;
4880         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
4881         sbi->s_sbh = bh;
4882         sbi->s_mount_state = le16_to_cpu(es->s_state);
4883         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
4884         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
4885
4886         for (i = 0; i < 4; i++)
4887                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
4888         sbi->s_def_hash_version = es->s_def_hash_version;
4889         if (ext4_has_feature_dir_index(sb)) {
4890                 i = le32_to_cpu(es->s_flags);
4891                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
4892                         sbi->s_hash_unsigned = 3;
4893                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
4894 #ifdef __CHAR_UNSIGNED__
4895                         if (!sb_rdonly(sb))
4896                                 es->s_flags |=
4897                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
4898                         sbi->s_hash_unsigned = 3;
4899 #else
4900                         if (!sb_rdonly(sb))
4901                                 es->s_flags |=
4902                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
4903 #endif
4904                 }
4905         }
4906
4907         /* Handle clustersize */
4908         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
4909         if (ext4_has_feature_bigalloc(sb)) {
4910                 if (clustersize < blocksize) {
4911                         ext4_msg(sb, KERN_ERR,
4912                                  "cluster size (%d) smaller than "
4913                                  "block size (%d)", clustersize, blocksize);
4914                         goto failed_mount;
4915                 }
4916                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
4917                         le32_to_cpu(es->s_log_block_size);
4918                 sbi->s_clusters_per_group =
4919                         le32_to_cpu(es->s_clusters_per_group);
4920                 if (sbi->s_clusters_per_group > blocksize * 8) {
4921                         ext4_msg(sb, KERN_ERR,
4922                                  "#clusters per group too big: %lu",
4923                                  sbi->s_clusters_per_group);
4924                         goto failed_mount;
4925                 }
4926                 if (sbi->s_blocks_per_group !=
4927                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
4928                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
4929                                  "clusters per group (%lu) inconsistent",
4930                                  sbi->s_blocks_per_group,
4931                                  sbi->s_clusters_per_group);
4932                         goto failed_mount;
4933                 }
4934         } else {
4935                 if (clustersize != blocksize) {
4936                         ext4_msg(sb, KERN_ERR,
4937                                  "fragment/cluster size (%d) != "
4938                                  "block size (%d)", clustersize, blocksize);
4939                         goto failed_mount;
4940                 }
4941                 if (sbi->s_blocks_per_group > blocksize * 8) {
4942                         ext4_msg(sb, KERN_ERR,
4943                                  "#blocks per group too big: %lu",
4944                                  sbi->s_blocks_per_group);
4945                         goto failed_mount;
4946                 }
4947                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
4948                 sbi->s_cluster_bits = 0;
4949         }
4950         sbi->s_cluster_ratio = clustersize / blocksize;
4951
4952         /* Do we have standard group size of clustersize * 8 blocks ? */
4953         if (sbi->s_blocks_per_group == clustersize << 3)
4954                 set_opt2(sb, STD_GROUP_SIZE);
4955
4956         /*
4957          * Test whether we have more sectors than will fit in sector_t,
4958          * and whether the max offset is addressable by the page cache.
4959          */
4960         err = generic_check_addressable(sb->s_blocksize_bits,
4961                                         ext4_blocks_count(es));
4962         if (err) {
4963                 ext4_msg(sb, KERN_ERR, "filesystem"
4964                          " too large to mount safely on this system");
4965                 goto failed_mount;
4966         }
4967
4968         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
4969                 goto cantfind_ext4;
4970
4971         /* check blocks count against device size */
4972         blocks_count = sb_bdev_nr_blocks(sb);
4973         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
4974                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
4975                        "exceeds size of device (%llu blocks)",
4976                        ext4_blocks_count(es), blocks_count);
4977                 goto failed_mount;
4978         }
4979
4980         /*
4981          * It makes no sense for the first data block to be beyond the end
4982          * of the filesystem.
4983          */
4984         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
4985                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4986                          "block %u is beyond end of filesystem (%llu)",
4987                          le32_to_cpu(es->s_first_data_block),
4988                          ext4_blocks_count(es));
4989                 goto failed_mount;
4990         }
4991         if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) &&
4992             (sbi->s_cluster_ratio == 1)) {
4993                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4994                          "block is 0 with a 1k block and cluster size");
4995                 goto failed_mount;
4996         }
4997
4998         blocks_count = (ext4_blocks_count(es) -
4999                         le32_to_cpu(es->s_first_data_block) +
5000                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
5001         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
5002         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
5003                 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu "
5004                        "(block count %llu, first data block %u, "
5005                        "blocks per group %lu)", blocks_count,
5006                        ext4_blocks_count(es),
5007                        le32_to_cpu(es->s_first_data_block),
5008                        EXT4_BLOCKS_PER_GROUP(sb));
5009                 goto failed_mount;
5010         }
5011         sbi->s_groups_count = blocks_count;
5012         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
5013                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
5014         if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) !=
5015             le32_to_cpu(es->s_inodes_count)) {
5016                 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu",
5017                          le32_to_cpu(es->s_inodes_count),
5018                          ((u64)sbi->s_groups_count * sbi->s_inodes_per_group));
5019                 ret = -EINVAL;
5020                 goto failed_mount;
5021         }
5022         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
5023                    EXT4_DESC_PER_BLOCK(sb);
5024         if (ext4_has_feature_meta_bg(sb)) {
5025                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
5026                         ext4_msg(sb, KERN_WARNING,
5027                                  "first meta block group too large: %u "
5028                                  "(group descriptor block count %u)",
5029                                  le32_to_cpu(es->s_first_meta_bg), db_count);
5030                         goto failed_mount;
5031                 }
5032         }
5033         rcu_assign_pointer(sbi->s_group_desc,
5034                            kvmalloc_array(db_count,
5035                                           sizeof(struct buffer_head *),
5036                                           GFP_KERNEL));
5037         if (sbi->s_group_desc == NULL) {
5038                 ext4_msg(sb, KERN_ERR, "not enough memory");
5039                 ret = -ENOMEM;
5040                 goto failed_mount;
5041         }
5042
5043         bgl_lock_init(sbi->s_blockgroup_lock);
5044
5045         /* Pre-read the descriptors into the buffer cache */
5046         for (i = 0; i < db_count; i++) {
5047                 block = descriptor_loc(sb, logical_sb_block, i);
5048                 ext4_sb_breadahead_unmovable(sb, block);
5049         }
5050
5051         for (i = 0; i < db_count; i++) {
5052                 struct buffer_head *bh;
5053
5054                 block = descriptor_loc(sb, logical_sb_block, i);
5055                 bh = ext4_sb_bread_unmovable(sb, block);
5056                 if (IS_ERR(bh)) {
5057                         ext4_msg(sb, KERN_ERR,
5058                                "can't read group descriptor %d", i);
5059                         db_count = i;
5060                         ret = PTR_ERR(bh);
5061                         goto failed_mount2;
5062                 }
5063                 rcu_read_lock();
5064                 rcu_dereference(sbi->s_group_desc)[i] = bh;
5065                 rcu_read_unlock();
5066         }
5067         sbi->s_gdb_count = db_count;
5068         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
5069                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
5070                 ret = -EFSCORRUPTED;
5071                 goto failed_mount2;
5072         }
5073
5074         timer_setup(&sbi->s_err_report, print_daily_error_info, 0);
5075         spin_lock_init(&sbi->s_error_lock);
5076         INIT_WORK(&sbi->s_error_work, flush_stashed_error_work);
5077
5078         /* Register extent status tree shrinker */
5079         if (ext4_es_register_shrinker(sbi))
5080                 goto failed_mount3;
5081
5082         sbi->s_stripe = ext4_get_stripe_size(sbi);
5083         sbi->s_extent_max_zeroout_kb = 32;
5084
5085         /*
5086          * set up enough so that it can read an inode
5087          */
5088         sb->s_op = &ext4_sops;
5089         sb->s_export_op = &ext4_export_ops;
5090         sb->s_xattr = ext4_xattr_handlers;
5091 #ifdef CONFIG_FS_ENCRYPTION
5092         sb->s_cop = &ext4_cryptops;
5093 #endif
5094 #ifdef CONFIG_FS_VERITY
5095         sb->s_vop = &ext4_verityops;
5096 #endif
5097 #ifdef CONFIG_QUOTA
5098         sb->dq_op = &ext4_quota_operations;
5099         if (ext4_has_feature_quota(sb))
5100                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
5101         else
5102                 sb->s_qcop = &ext4_qctl_operations;
5103         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
5104 #endif
5105         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
5106
5107         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
5108         mutex_init(&sbi->s_orphan_lock);
5109
5110         /* Initialize fast commit stuff */
5111         atomic_set(&sbi->s_fc_subtid, 0);
5112         INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_MAIN]);
5113         INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_STAGING]);
5114         INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_MAIN]);
5115         INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_STAGING]);
5116         sbi->s_fc_bytes = 0;
5117         ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE);
5118         sbi->s_fc_ineligible_tid = 0;
5119         spin_lock_init(&sbi->s_fc_lock);
5120         memset(&sbi->s_fc_stats, 0, sizeof(sbi->s_fc_stats));
5121         sbi->s_fc_replay_state.fc_regions = NULL;
5122         sbi->s_fc_replay_state.fc_regions_size = 0;
5123         sbi->s_fc_replay_state.fc_regions_used = 0;
5124         sbi->s_fc_replay_state.fc_regions_valid = 0;
5125         sbi->s_fc_replay_state.fc_modified_inodes = NULL;
5126         sbi->s_fc_replay_state.fc_modified_inodes_size = 0;
5127         sbi->s_fc_replay_state.fc_modified_inodes_used = 0;
5128
5129         sb->s_root = NULL;
5130
5131         needs_recovery = (es->s_last_orphan != 0 ||
5132                           ext4_has_feature_orphan_present(sb) ||
5133                           ext4_has_feature_journal_needs_recovery(sb));
5134
5135         if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb))
5136                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
5137                         goto failed_mount3a;
5138
5139         /*
5140          * The first inode we look at is the journal inode.  Don't try
5141          * root first: it may be modified in the journal!
5142          */
5143         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
5144                 err = ext4_load_journal(sb, es, ctx->journal_devnum);
5145                 if (err)
5146                         goto failed_mount3a;
5147         } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
5148                    ext4_has_feature_journal_needs_recovery(sb)) {
5149                 ext4_msg(sb, KERN_ERR, "required journal recovery "
5150                        "suppressed and not mounted read-only");
5151                 goto failed_mount_wq;
5152         } else {
5153                 /* Nojournal mode, all journal mount options are illegal */
5154                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
5155                         ext4_msg(sb, KERN_ERR, "can't mount with "
5156                                  "journal_checksum, fs mounted w/o journal");
5157                         goto failed_mount_wq;
5158                 }
5159                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5160                         ext4_msg(sb, KERN_ERR, "can't mount with "
5161                                  "journal_async_commit, fs mounted w/o journal");
5162                         goto failed_mount_wq;
5163                 }
5164                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
5165                         ext4_msg(sb, KERN_ERR, "can't mount with "
5166                                  "commit=%lu, fs mounted w/o journal",
5167                                  sbi->s_commit_interval / HZ);
5168                         goto failed_mount_wq;
5169                 }
5170                 if (EXT4_MOUNT_DATA_FLAGS &
5171                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
5172                         ext4_msg(sb, KERN_ERR, "can't mount with "
5173                                  "data=, fs mounted w/o journal");
5174                         goto failed_mount_wq;
5175                 }
5176                 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM;
5177                 clear_opt(sb, JOURNAL_CHECKSUM);
5178                 clear_opt(sb, DATA_FLAGS);
5179                 clear_opt2(sb, JOURNAL_FAST_COMMIT);
5180                 sbi->s_journal = NULL;
5181                 needs_recovery = 0;
5182                 goto no_journal;
5183         }
5184
5185         if (ext4_has_feature_64bit(sb) &&
5186             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
5187                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
5188                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
5189                 goto failed_mount_wq;
5190         }
5191
5192         if (!set_journal_csum_feature_set(sb)) {
5193                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
5194                          "feature set");
5195                 goto failed_mount_wq;
5196         }
5197
5198         if (test_opt2(sb, JOURNAL_FAST_COMMIT) &&
5199                 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
5200                                           JBD2_FEATURE_INCOMPAT_FAST_COMMIT)) {
5201                 ext4_msg(sb, KERN_ERR,
5202                         "Failed to set fast commit journal feature");
5203                 goto failed_mount_wq;
5204         }
5205
5206         /* We have now updated the journal if required, so we can
5207          * validate the data journaling mode. */
5208         switch (test_opt(sb, DATA_FLAGS)) {
5209         case 0:
5210                 /* No mode set, assume a default based on the journal
5211                  * capabilities: ORDERED_DATA if the journal can
5212                  * cope, else JOURNAL_DATA
5213                  */
5214                 if (jbd2_journal_check_available_features
5215                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
5216                         set_opt(sb, ORDERED_DATA);
5217                         sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
5218                 } else {
5219                         set_opt(sb, JOURNAL_DATA);
5220                         sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
5221                 }
5222                 break;
5223
5224         case EXT4_MOUNT_ORDERED_DATA:
5225         case EXT4_MOUNT_WRITEBACK_DATA:
5226                 if (!jbd2_journal_check_available_features
5227                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
5228                         ext4_msg(sb, KERN_ERR, "Journal does not support "
5229                                "requested data journaling mode");
5230                         goto failed_mount_wq;
5231                 }
5232                 break;
5233         default:
5234                 break;
5235         }
5236
5237         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
5238             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5239                 ext4_msg(sb, KERN_ERR, "can't mount with "
5240                         "journal_async_commit in data=ordered mode");
5241                 goto failed_mount_wq;
5242         }
5243
5244         set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio);
5245
5246         sbi->s_journal->j_submit_inode_data_buffers =
5247                 ext4_journal_submit_inode_data_buffers;
5248         sbi->s_journal->j_finish_inode_data_buffers =
5249                 ext4_journal_finish_inode_data_buffers;
5250
5251 no_journal:
5252         if (!test_opt(sb, NO_MBCACHE)) {
5253                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
5254                 if (!sbi->s_ea_block_cache) {
5255                         ext4_msg(sb, KERN_ERR,
5256                                  "Failed to create ea_block_cache");
5257                         goto failed_mount_wq;
5258                 }
5259
5260                 if (ext4_has_feature_ea_inode(sb)) {
5261                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
5262                         if (!sbi->s_ea_inode_cache) {
5263                                 ext4_msg(sb, KERN_ERR,
5264                                          "Failed to create ea_inode_cache");
5265                                 goto failed_mount_wq;
5266                         }
5267                 }
5268         }
5269
5270         if (ext4_has_feature_verity(sb) && blocksize != PAGE_SIZE) {
5271                 ext4_msg(sb, KERN_ERR, "Unsupported blocksize for fs-verity");
5272                 goto failed_mount_wq;
5273         }
5274
5275         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) &&
5276             !ext4_has_feature_encrypt(sb)) {
5277                 ext4_set_feature_encrypt(sb);
5278                 ext4_commit_super(sb);
5279         }
5280
5281         /*
5282          * Get the # of file system overhead blocks from the
5283          * superblock if present.
5284          */
5285         if (es->s_overhead_clusters)
5286                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
5287         else {
5288                 err = ext4_calculate_overhead(sb);
5289                 if (err)
5290                         goto failed_mount_wq;
5291         }
5292
5293         /*
5294          * The maximum number of concurrent works can be high and
5295          * concurrency isn't really necessary.  Limit it to 1.
5296          */
5297         EXT4_SB(sb)->rsv_conversion_wq =
5298                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
5299         if (!EXT4_SB(sb)->rsv_conversion_wq) {
5300                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
5301                 ret = -ENOMEM;
5302                 goto failed_mount4;
5303         }
5304
5305         /*
5306          * The jbd2_journal_load will have done any necessary log recovery,
5307          * so we can safely mount the rest of the filesystem now.
5308          */
5309
5310         root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL);
5311         if (IS_ERR(root)) {
5312                 ext4_msg(sb, KERN_ERR, "get root inode failed");
5313                 ret = PTR_ERR(root);
5314                 root = NULL;
5315                 goto failed_mount4;
5316         }
5317         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
5318                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
5319                 iput(root);
5320                 goto failed_mount4;
5321         }
5322
5323         sb->s_root = d_make_root(root);
5324         if (!sb->s_root) {
5325                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
5326                 ret = -ENOMEM;
5327                 goto failed_mount4;
5328         }
5329
5330         ret = ext4_setup_super(sb, es, sb_rdonly(sb));
5331         if (ret == -EROFS) {
5332                 sb->s_flags |= SB_RDONLY;
5333                 ret = 0;
5334         } else if (ret)
5335                 goto failed_mount4a;
5336
5337         ext4_set_resv_clusters(sb);
5338
5339         if (test_opt(sb, BLOCK_VALIDITY)) {
5340                 err = ext4_setup_system_zone(sb);
5341                 if (err) {
5342                         ext4_msg(sb, KERN_ERR, "failed to initialize system "
5343                                  "zone (%d)", err);
5344                         goto failed_mount4a;
5345                 }
5346         }
5347         ext4_fc_replay_cleanup(sb);
5348
5349         ext4_ext_init(sb);
5350
5351         /*
5352          * Enable optimize_scan if number of groups is > threshold. This can be
5353          * turned off by passing "mb_optimize_scan=0". This can also be
5354          * turned on forcefully by passing "mb_optimize_scan=1".
5355          */
5356         if (!(ctx->spec & EXT4_SPEC_mb_optimize_scan)) {
5357                 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD)
5358                         set_opt2(sb, MB_OPTIMIZE_SCAN);
5359                 else
5360                         clear_opt2(sb, MB_OPTIMIZE_SCAN);
5361         }
5362
5363         err = ext4_mb_init(sb);
5364         if (err) {
5365                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
5366                          err);
5367                 goto failed_mount5;
5368         }
5369
5370         /*
5371          * We can only set up the journal commit callback once
5372          * mballoc is initialized
5373          */
5374         if (sbi->s_journal)
5375                 sbi->s_journal->j_commit_callback =
5376                         ext4_journal_commit_callback;
5377
5378         block = ext4_count_free_clusters(sb);
5379         ext4_free_blocks_count_set(sbi->s_es,
5380                                    EXT4_C2B(sbi, block));
5381         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
5382                                   GFP_KERNEL);
5383         if (!err) {
5384                 unsigned long freei = ext4_count_free_inodes(sb);
5385                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
5386                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
5387                                           GFP_KERNEL);
5388         }
5389         /*
5390          * Update the checksum after updating free space/inode
5391          * counters.  Otherwise the superblock can have an incorrect
5392          * checksum in the buffer cache until it is written out and
5393          * e2fsprogs programs trying to open a file system immediately
5394          * after it is mounted can fail.
5395          */
5396         ext4_superblock_csum_set(sb);
5397         if (!err)
5398                 err = percpu_counter_init(&sbi->s_dirs_counter,
5399                                           ext4_count_dirs(sb), GFP_KERNEL);
5400         if (!err)
5401                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
5402                                           GFP_KERNEL);
5403         if (!err)
5404                 err = percpu_counter_init(&sbi->s_sra_exceeded_retry_limit, 0,
5405                                           GFP_KERNEL);
5406         if (!err)
5407                 err = percpu_init_rwsem(&sbi->s_writepages_rwsem);
5408
5409         if (err) {
5410                 ext4_msg(sb, KERN_ERR, "insufficient memory");
5411                 goto failed_mount6;
5412         }
5413
5414         if (ext4_has_feature_flex_bg(sb))
5415                 if (!ext4_fill_flex_info(sb)) {
5416                         ext4_msg(sb, KERN_ERR,
5417                                "unable to initialize "
5418                                "flex_bg meta info!");
5419                         ret = -ENOMEM;
5420                         goto failed_mount6;
5421                 }
5422
5423         err = ext4_register_li_request(sb, first_not_zeroed);
5424         if (err)
5425                 goto failed_mount6;
5426
5427         err = ext4_register_sysfs(sb);
5428         if (err)
5429                 goto failed_mount7;
5430
5431         err = ext4_init_orphan_info(sb);
5432         if (err)
5433                 goto failed_mount8;
5434 #ifdef CONFIG_QUOTA
5435         /* Enable quota usage during mount. */
5436         if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
5437                 err = ext4_enable_quotas(sb);
5438                 if (err)
5439                         goto failed_mount9;
5440         }
5441 #endif  /* CONFIG_QUOTA */
5442
5443         /*
5444          * Save the original bdev mapping's wb_err value which could be
5445          * used to detect the metadata async write error.
5446          */
5447         spin_lock_init(&sbi->s_bdev_wb_lock);
5448         errseq_check_and_advance(&sb->s_bdev->bd_inode->i_mapping->wb_err,
5449                                  &sbi->s_bdev_wb_err);
5450         sb->s_bdev->bd_super = sb;
5451         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
5452         ext4_orphan_cleanup(sb, es);
5453         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
5454         if (needs_recovery) {
5455                 ext4_msg(sb, KERN_INFO, "recovery complete");
5456                 err = ext4_mark_recovery_complete(sb, es);
5457                 if (err)
5458                         goto failed_mount9;
5459         }
5460
5461         if (test_opt(sb, DISCARD)) {
5462                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
5463                 if (!blk_queue_discard(q))
5464                         ext4_msg(sb, KERN_WARNING,
5465                                  "mounting with \"discard\" option, but "
5466                                  "the device does not support discard");
5467         }
5468
5469         if (es->s_error_count)
5470                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
5471
5472         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
5473         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
5474         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
5475         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
5476         atomic_set(&sbi->s_warning_count, 0);
5477         atomic_set(&sbi->s_msg_count, 0);
5478
5479         return 0;
5480
5481 cantfind_ext4:
5482         if (!silent)
5483                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
5484         goto failed_mount;
5485
5486 failed_mount9:
5487         ext4_release_orphan_info(sb);
5488 failed_mount8:
5489         ext4_unregister_sysfs(sb);
5490         kobject_put(&sbi->s_kobj);
5491 failed_mount7:
5492         ext4_unregister_li_request(sb);
5493 failed_mount6:
5494         ext4_mb_release(sb);
5495         rcu_read_lock();
5496         flex_groups = rcu_dereference(sbi->s_flex_groups);
5497         if (flex_groups) {
5498                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
5499                         kvfree(flex_groups[i]);
5500                 kvfree(flex_groups);
5501         }
5502         rcu_read_unlock();
5503         percpu_counter_destroy(&sbi->s_freeclusters_counter);
5504         percpu_counter_destroy(&sbi->s_freeinodes_counter);
5505         percpu_counter_destroy(&sbi->s_dirs_counter);
5506         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
5507         percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit);
5508         percpu_free_rwsem(&sbi->s_writepages_rwsem);
5509 failed_mount5:
5510         ext4_ext_release(sb);
5511         ext4_release_system_zone(sb);
5512 failed_mount4a:
5513         dput(sb->s_root);
5514         sb->s_root = NULL;
5515 failed_mount4:
5516         ext4_msg(sb, KERN_ERR, "mount failed");
5517         if (EXT4_SB(sb)->rsv_conversion_wq)
5518                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
5519 failed_mount_wq:
5520         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
5521         sbi->s_ea_inode_cache = NULL;
5522
5523         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
5524         sbi->s_ea_block_cache = NULL;
5525
5526         if (sbi->s_journal) {
5527                 /* flush s_error_work before journal destroy. */
5528                 flush_work(&sbi->s_error_work);
5529                 jbd2_journal_destroy(sbi->s_journal);
5530                 sbi->s_journal = NULL;
5531         }
5532 failed_mount3a:
5533         ext4_es_unregister_shrinker(sbi);
5534 failed_mount3:
5535         /* flush s_error_work before sbi destroy */
5536         flush_work(&sbi->s_error_work);
5537         del_timer_sync(&sbi->s_err_report);
5538         ext4_stop_mmpd(sbi);
5539 failed_mount2:
5540         rcu_read_lock();
5541         group_desc = rcu_dereference(sbi->s_group_desc);
5542         for (i = 0; i < db_count; i++)
5543                 brelse(group_desc[i]);
5544         kvfree(group_desc);
5545         rcu_read_unlock();
5546 failed_mount:
5547         if (sbi->s_chksum_driver)
5548                 crypto_free_shash(sbi->s_chksum_driver);
5549
5550 #if IS_ENABLED(CONFIG_UNICODE)
5551         utf8_unload(sb->s_encoding);
5552 #endif
5553
5554 #ifdef CONFIG_QUOTA
5555         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5556                 kfree(get_qf_name(sb, sbi, i));
5557 #endif
5558         fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy);
5559         /* ext4_blkdev_remove() calls kill_bdev(), release bh before it. */
5560         brelse(bh);
5561         ext4_blkdev_remove(sbi);
5562 out_fail:
5563         sb->s_fs_info = NULL;
5564         return err ? err : ret;
5565 }
5566
5567 static int ext4_fill_super(struct super_block *sb, struct fs_context *fc)
5568 {
5569         struct ext4_fs_context *ctx = fc->fs_private;
5570         struct ext4_sb_info *sbi;
5571         const char *descr;
5572         int ret;
5573
5574         sbi = ext4_alloc_sbi(sb);
5575         if (!sbi)
5576                 return -ENOMEM;
5577
5578         fc->s_fs_info = sbi;
5579
5580         /* Cleanup superblock name */
5581         strreplace(sb->s_id, '/', '!');
5582
5583         sbi->s_sb_block = 1;    /* Default super block location */
5584         if (ctx->spec & EXT4_SPEC_s_sb_block)
5585                 sbi->s_sb_block = ctx->s_sb_block;
5586
5587         ret = __ext4_fill_super(fc, sb);
5588         if (ret < 0)
5589                 goto free_sbi;
5590
5591         if (sbi->s_journal) {
5592                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
5593                         descr = " journalled data mode";
5594                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
5595                         descr = " ordered data mode";
5596                 else
5597                         descr = " writeback data mode";
5598         } else
5599                 descr = "out journal";
5600
5601         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
5602                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
5603                          "Quota mode: %s.", descr, ext4_quota_mode(sb));
5604
5605         return 0;
5606
5607 free_sbi:
5608         ext4_free_sbi(sbi);
5609         fc->s_fs_info = NULL;
5610         return ret;
5611 }
5612
5613 static int ext4_get_tree(struct fs_context *fc)
5614 {
5615         return get_tree_bdev(fc, ext4_fill_super);
5616 }
5617
5618 /*
5619  * Setup any per-fs journal parameters now.  We'll do this both on
5620  * initial mount, once the journal has been initialised but before we've
5621  * done any recovery; and again on any subsequent remount.
5622  */
5623 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
5624 {
5625         struct ext4_sb_info *sbi = EXT4_SB(sb);
5626
5627         journal->j_commit_interval = sbi->s_commit_interval;
5628         journal->j_min_batch_time = sbi->s_min_batch_time;
5629         journal->j_max_batch_time = sbi->s_max_batch_time;
5630         ext4_fc_init(sb, journal);
5631
5632         write_lock(&journal->j_state_lock);
5633         if (test_opt(sb, BARRIER))
5634                 journal->j_flags |= JBD2_BARRIER;
5635         else
5636                 journal->j_flags &= ~JBD2_BARRIER;
5637         if (test_opt(sb, DATA_ERR_ABORT))
5638                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
5639         else
5640                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
5641         write_unlock(&journal->j_state_lock);
5642 }
5643
5644 static struct inode *ext4_get_journal_inode(struct super_block *sb,
5645                                              unsigned int journal_inum)
5646 {
5647         struct inode *journal_inode;
5648
5649         /*
5650          * Test for the existence of a valid inode on disk.  Bad things
5651          * happen if we iget() an unused inode, as the subsequent iput()
5652          * will try to delete it.
5653          */
5654         journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL);
5655         if (IS_ERR(journal_inode)) {
5656                 ext4_msg(sb, KERN_ERR, "no journal found");
5657                 return NULL;
5658         }
5659         if (!journal_inode->i_nlink) {
5660                 make_bad_inode(journal_inode);
5661                 iput(journal_inode);
5662                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
5663                 return NULL;
5664         }
5665
5666         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
5667                   journal_inode, journal_inode->i_size);
5668         if (!S_ISREG(journal_inode->i_mode)) {
5669                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
5670                 iput(journal_inode);
5671                 return NULL;
5672         }
5673         return journal_inode;
5674 }
5675
5676 static journal_t *ext4_get_journal(struct super_block *sb,
5677                                    unsigned int journal_inum)
5678 {
5679         struct inode *journal_inode;
5680         journal_t *journal;
5681
5682         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5683                 return NULL;
5684
5685         journal_inode = ext4_get_journal_inode(sb, journal_inum);
5686         if (!journal_inode)
5687                 return NULL;
5688
5689         journal = jbd2_journal_init_inode(journal_inode);
5690         if (!journal) {
5691                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
5692                 iput(journal_inode);
5693                 return NULL;
5694         }
5695         journal->j_private = sb;
5696         ext4_init_journal_params(sb, journal);
5697         return journal;
5698 }
5699
5700 static journal_t *ext4_get_dev_journal(struct super_block *sb,
5701                                        dev_t j_dev)
5702 {
5703         struct buffer_head *bh;
5704         journal_t *journal;
5705         ext4_fsblk_t start;
5706         ext4_fsblk_t len;
5707         int hblock, blocksize;
5708         ext4_fsblk_t sb_block;
5709         unsigned long offset;
5710         struct ext4_super_block *es;
5711         struct block_device *bdev;
5712
5713         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5714                 return NULL;
5715
5716         bdev = ext4_blkdev_get(j_dev, sb);
5717         if (bdev == NULL)
5718                 return NULL;
5719
5720         blocksize = sb->s_blocksize;
5721         hblock = bdev_logical_block_size(bdev);
5722         if (blocksize < hblock) {
5723                 ext4_msg(sb, KERN_ERR,
5724                         "blocksize too small for journal device");
5725                 goto out_bdev;
5726         }
5727
5728         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
5729         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
5730         set_blocksize(bdev, blocksize);
5731         if (!(bh = __bread(bdev, sb_block, blocksize))) {
5732                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
5733                        "external journal");
5734                 goto out_bdev;
5735         }
5736
5737         es = (struct ext4_super_block *) (bh->b_data + offset);
5738         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
5739             !(le32_to_cpu(es->s_feature_incompat) &
5740               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
5741                 ext4_msg(sb, KERN_ERR, "external journal has "
5742                                         "bad superblock");
5743                 brelse(bh);
5744                 goto out_bdev;
5745         }
5746
5747         if ((le32_to_cpu(es->s_feature_ro_compat) &
5748              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
5749             es->s_checksum != ext4_superblock_csum(sb, es)) {
5750                 ext4_msg(sb, KERN_ERR, "external journal has "
5751                                        "corrupt superblock");
5752                 brelse(bh);
5753                 goto out_bdev;
5754         }
5755
5756         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
5757                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
5758                 brelse(bh);
5759                 goto out_bdev;
5760         }
5761
5762         len = ext4_blocks_count(es);
5763         start = sb_block + 1;
5764         brelse(bh);     /* we're done with the superblock */
5765
5766         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
5767                                         start, len, blocksize);
5768         if (!journal) {
5769                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
5770                 goto out_bdev;
5771         }
5772         journal->j_private = sb;
5773         if (ext4_read_bh_lock(journal->j_sb_buffer, REQ_META | REQ_PRIO, true)) {
5774                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
5775                 goto out_journal;
5776         }
5777         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
5778                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
5779                                         "user (unsupported) - %d",
5780                         be32_to_cpu(journal->j_superblock->s_nr_users));
5781                 goto out_journal;
5782         }
5783         EXT4_SB(sb)->s_journal_bdev = bdev;
5784         ext4_init_journal_params(sb, journal);
5785         return journal;
5786
5787 out_journal:
5788         jbd2_journal_destroy(journal);
5789 out_bdev:
5790         ext4_blkdev_put(bdev);
5791         return NULL;
5792 }
5793
5794 static int ext4_load_journal(struct super_block *sb,
5795                              struct ext4_super_block *es,
5796                              unsigned long journal_devnum)
5797 {
5798         journal_t *journal;
5799         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
5800         dev_t journal_dev;
5801         int err = 0;
5802         int really_read_only;
5803         int journal_dev_ro;
5804
5805         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5806                 return -EFSCORRUPTED;
5807
5808         if (journal_devnum &&
5809             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5810                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
5811                         "numbers have changed");
5812                 journal_dev = new_decode_dev(journal_devnum);
5813         } else
5814                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
5815
5816         if (journal_inum && journal_dev) {
5817                 ext4_msg(sb, KERN_ERR,
5818                          "filesystem has both journal inode and journal device!");
5819                 return -EINVAL;
5820         }
5821
5822         if (journal_inum) {
5823                 journal = ext4_get_journal(sb, journal_inum);
5824                 if (!journal)
5825                         return -EINVAL;
5826         } else {
5827                 journal = ext4_get_dev_journal(sb, journal_dev);
5828                 if (!journal)
5829                         return -EINVAL;
5830         }
5831
5832         journal_dev_ro = bdev_read_only(journal->j_dev);
5833         really_read_only = bdev_read_only(sb->s_bdev) | journal_dev_ro;
5834
5835         if (journal_dev_ro && !sb_rdonly(sb)) {
5836                 ext4_msg(sb, KERN_ERR,
5837                          "journal device read-only, try mounting with '-o ro'");
5838                 err = -EROFS;
5839                 goto err_out;
5840         }
5841
5842         /*
5843          * Are we loading a blank journal or performing recovery after a
5844          * crash?  For recovery, we need to check in advance whether we
5845          * can get read-write access to the device.
5846          */
5847         if (ext4_has_feature_journal_needs_recovery(sb)) {
5848                 if (sb_rdonly(sb)) {
5849                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
5850                                         "required on readonly filesystem");
5851                         if (really_read_only) {
5852                                 ext4_msg(sb, KERN_ERR, "write access "
5853                                         "unavailable, cannot proceed "
5854                                         "(try mounting with noload)");
5855                                 err = -EROFS;
5856                                 goto err_out;
5857                         }
5858                         ext4_msg(sb, KERN_INFO, "write access will "
5859                                "be enabled during recovery");
5860                 }
5861         }
5862
5863         if (!(journal->j_flags & JBD2_BARRIER))
5864                 ext4_msg(sb, KERN_INFO, "barriers disabled");
5865
5866         if (!ext4_has_feature_journal_needs_recovery(sb))
5867                 err = jbd2_journal_wipe(journal, !really_read_only);
5868         if (!err) {
5869                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
5870                 if (save)
5871                         memcpy(save, ((char *) es) +
5872                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
5873                 err = jbd2_journal_load(journal);
5874                 if (save)
5875                         memcpy(((char *) es) + EXT4_S_ERR_START,
5876                                save, EXT4_S_ERR_LEN);
5877                 kfree(save);
5878         }
5879
5880         if (err) {
5881                 ext4_msg(sb, KERN_ERR, "error loading journal");
5882                 goto err_out;
5883         }
5884
5885         EXT4_SB(sb)->s_journal = journal;
5886         err = ext4_clear_journal_err(sb, es);
5887         if (err) {
5888                 EXT4_SB(sb)->s_journal = NULL;
5889                 jbd2_journal_destroy(journal);
5890                 return err;
5891         }
5892
5893         if (!really_read_only && journal_devnum &&
5894             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5895                 es->s_journal_dev = cpu_to_le32(journal_devnum);
5896
5897                 /* Make sure we flush the recovery flag to disk. */
5898                 ext4_commit_super(sb);
5899         }
5900
5901         return 0;
5902
5903 err_out:
5904         jbd2_journal_destroy(journal);
5905         return err;
5906 }
5907
5908 /* Copy state of EXT4_SB(sb) into buffer for on-disk superblock */
5909 static void ext4_update_super(struct super_block *sb)
5910 {
5911         struct ext4_sb_info *sbi = EXT4_SB(sb);
5912         struct ext4_super_block *es = sbi->s_es;
5913         struct buffer_head *sbh = sbi->s_sbh;
5914
5915         lock_buffer(sbh);
5916         /*
5917          * If the file system is mounted read-only, don't update the
5918          * superblock write time.  This avoids updating the superblock
5919          * write time when we are mounting the root file system
5920          * read/only but we need to replay the journal; at that point,
5921          * for people who are east of GMT and who make their clock
5922          * tick in localtime for Windows bug-for-bug compatibility,
5923          * the clock is set in the future, and this will cause e2fsck
5924          * to complain and force a full file system check.
5925          */
5926         if (!(sb->s_flags & SB_RDONLY))
5927                 ext4_update_tstamp(es, s_wtime);
5928         es->s_kbytes_written =
5929                 cpu_to_le64(sbi->s_kbytes_written +
5930                     ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) -
5931                       sbi->s_sectors_written_start) >> 1));
5932         if (percpu_counter_initialized(&sbi->s_freeclusters_counter))
5933                 ext4_free_blocks_count_set(es,
5934                         EXT4_C2B(sbi, percpu_counter_sum_positive(
5935                                 &sbi->s_freeclusters_counter)));
5936         if (percpu_counter_initialized(&sbi->s_freeinodes_counter))
5937                 es->s_free_inodes_count =
5938                         cpu_to_le32(percpu_counter_sum_positive(
5939                                 &sbi->s_freeinodes_counter));
5940         /* Copy error information to the on-disk superblock */
5941         spin_lock(&sbi->s_error_lock);
5942         if (sbi->s_add_error_count > 0) {
5943                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
5944                 if (!es->s_first_error_time && !es->s_first_error_time_hi) {
5945                         __ext4_update_tstamp(&es->s_first_error_time,
5946                                              &es->s_first_error_time_hi,
5947                                              sbi->s_first_error_time);
5948                         strncpy(es->s_first_error_func, sbi->s_first_error_func,
5949                                 sizeof(es->s_first_error_func));
5950                         es->s_first_error_line =
5951                                 cpu_to_le32(sbi->s_first_error_line);
5952                         es->s_first_error_ino =
5953                                 cpu_to_le32(sbi->s_first_error_ino);
5954                         es->s_first_error_block =
5955                                 cpu_to_le64(sbi->s_first_error_block);
5956                         es->s_first_error_errcode =
5957                                 ext4_errno_to_code(sbi->s_first_error_code);
5958                 }
5959                 __ext4_update_tstamp(&es->s_last_error_time,
5960                                      &es->s_last_error_time_hi,
5961                                      sbi->s_last_error_time);
5962                 strncpy(es->s_last_error_func, sbi->s_last_error_func,
5963                         sizeof(es->s_last_error_func));
5964                 es->s_last_error_line = cpu_to_le32(sbi->s_last_error_line);
5965                 es->s_last_error_ino = cpu_to_le32(sbi->s_last_error_ino);
5966                 es->s_last_error_block = cpu_to_le64(sbi->s_last_error_block);
5967                 es->s_last_error_errcode =
5968                                 ext4_errno_to_code(sbi->s_last_error_code);
5969                 /*
5970                  * Start the daily error reporting function if it hasn't been
5971                  * started already
5972                  */
5973                 if (!es->s_error_count)
5974                         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);
5975                 le32_add_cpu(&es->s_error_count, sbi->s_add_error_count);
5976                 sbi->s_add_error_count = 0;
5977         }
5978         spin_unlock(&sbi->s_error_lock);
5979
5980         ext4_superblock_csum_set(sb);
5981         unlock_buffer(sbh);
5982 }
5983
5984 static int ext4_commit_super(struct super_block *sb)
5985 {
5986         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
5987         int error = 0;
5988
5989         if (!sbh)
5990                 return -EINVAL;
5991         if (block_device_ejected(sb))
5992                 return -ENODEV;
5993
5994         ext4_update_super(sb);
5995
5996         if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
5997                 /*
5998                  * Oh, dear.  A previous attempt to write the
5999                  * superblock failed.  This could happen because the
6000                  * USB device was yanked out.  Or it could happen to
6001                  * be a transient write error and maybe the block will
6002                  * be remapped.  Nothing we can do but to retry the
6003                  * write and hope for the best.
6004                  */
6005                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
6006                        "superblock detected");
6007                 clear_buffer_write_io_error(sbh);
6008                 set_buffer_uptodate(sbh);
6009         }
6010         BUFFER_TRACE(sbh, "marking dirty");
6011         mark_buffer_dirty(sbh);
6012         error = __sync_dirty_buffer(sbh,
6013                 REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
6014         if (buffer_write_io_error(sbh)) {
6015                 ext4_msg(sb, KERN_ERR, "I/O error while writing "
6016                        "superblock");
6017                 clear_buffer_write_io_error(sbh);
6018                 set_buffer_uptodate(sbh);
6019         }
6020         return error;
6021 }
6022
6023 /*
6024  * Have we just finished recovery?  If so, and if we are mounting (or
6025  * remounting) the filesystem readonly, then we will end up with a
6026  * consistent fs on disk.  Record that fact.
6027  */
6028 static int ext4_mark_recovery_complete(struct super_block *sb,
6029                                        struct ext4_super_block *es)
6030 {
6031         int err;
6032         journal_t *journal = EXT4_SB(sb)->s_journal;
6033
6034         if (!ext4_has_feature_journal(sb)) {
6035                 if (journal != NULL) {
6036                         ext4_error(sb, "Journal got removed while the fs was "
6037                                    "mounted!");
6038                         return -EFSCORRUPTED;
6039                 }
6040                 return 0;
6041         }
6042         jbd2_journal_lock_updates(journal);
6043         err = jbd2_journal_flush(journal, 0);
6044         if (err < 0)
6045                 goto out;
6046
6047         if (sb_rdonly(sb) && (ext4_has_feature_journal_needs_recovery(sb) ||
6048             ext4_has_feature_orphan_present(sb))) {
6049                 if (!ext4_orphan_file_empty(sb)) {
6050                         ext4_error(sb, "Orphan file not empty on read-only fs.");
6051                         err = -EFSCORRUPTED;
6052                         goto out;
6053                 }
6054                 ext4_clear_feature_journal_needs_recovery(sb);
6055                 ext4_clear_feature_orphan_present(sb);
6056                 ext4_commit_super(sb);
6057         }
6058 out:
6059         jbd2_journal_unlock_updates(journal);
6060         return err;
6061 }
6062
6063 /*
6064  * If we are mounting (or read-write remounting) a filesystem whose journal
6065  * has recorded an error from a previous lifetime, move that error to the
6066  * main filesystem now.
6067  */
6068 static int ext4_clear_journal_err(struct super_block *sb,
6069                                    struct ext4_super_block *es)
6070 {
6071         journal_t *journal;
6072         int j_errno;
6073         const char *errstr;
6074
6075         if (!ext4_has_feature_journal(sb)) {
6076                 ext4_error(sb, "Journal got removed while the fs was mounted!");
6077                 return -EFSCORRUPTED;
6078         }
6079
6080         journal = EXT4_SB(sb)->s_journal;
6081
6082         /*
6083          * Now check for any error status which may have been recorded in the
6084          * journal by a prior ext4_error() or ext4_abort()
6085          */
6086
6087         j_errno = jbd2_journal_errno(journal);
6088         if (j_errno) {
6089                 char nbuf[16];
6090
6091                 errstr = ext4_decode_error(sb, j_errno, nbuf);
6092                 ext4_warning(sb, "Filesystem error recorded "
6093                              "from previous mount: %s", errstr);
6094                 ext4_warning(sb, "Marking fs in need of filesystem check.");
6095
6096                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
6097                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
6098                 ext4_commit_super(sb);
6099
6100                 jbd2_journal_clear_err(journal);
6101                 jbd2_journal_update_sb_errno(journal);
6102         }
6103         return 0;
6104 }
6105
6106 /*
6107  * Force the running and committing transactions to commit,
6108  * and wait on the commit.
6109  */
6110 int ext4_force_commit(struct super_block *sb)
6111 {
6112         journal_t *journal;
6113
6114         if (sb_rdonly(sb))
6115                 return 0;
6116
6117         journal = EXT4_SB(sb)->s_journal;
6118         return ext4_journal_force_commit(journal);
6119 }
6120
6121 static int ext4_sync_fs(struct super_block *sb, int wait)
6122 {
6123         int ret = 0;
6124         tid_t target;
6125         bool needs_barrier = false;
6126         struct ext4_sb_info *sbi = EXT4_SB(sb);
6127
6128         if (unlikely(ext4_forced_shutdown(sbi)))
6129                 return 0;
6130
6131         trace_ext4_sync_fs(sb, wait);
6132         flush_workqueue(sbi->rsv_conversion_wq);
6133         /*
6134          * Writeback quota in non-journalled quota case - journalled quota has
6135          * no dirty dquots
6136          */
6137         dquot_writeback_dquots(sb, -1);
6138         /*
6139          * Data writeback is possible w/o journal transaction, so barrier must
6140          * being sent at the end of the function. But we can skip it if
6141          * transaction_commit will do it for us.
6142          */
6143         if (sbi->s_journal) {
6144                 target = jbd2_get_latest_transaction(sbi->s_journal);
6145                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
6146                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
6147                         needs_barrier = true;
6148
6149                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
6150                         if (wait)
6151                                 ret = jbd2_log_wait_commit(sbi->s_journal,
6152                                                            target);
6153                 }
6154         } else if (wait && test_opt(sb, BARRIER))
6155                 needs_barrier = true;
6156         if (needs_barrier) {
6157                 int err;
6158                 err = blkdev_issue_flush(sb->s_bdev);
6159                 if (!ret)
6160                         ret = err;
6161         }
6162
6163         return ret;
6164 }
6165
6166 /*
6167  * LVM calls this function before a (read-only) snapshot is created.  This
6168  * gives us a chance to flush the journal completely and mark the fs clean.
6169  *
6170  * Note that only this function cannot bring a filesystem to be in a clean
6171  * state independently. It relies on upper layer to stop all data & metadata
6172  * modifications.
6173  */
6174 static int ext4_freeze(struct super_block *sb)
6175 {
6176         int error = 0;
6177         journal_t *journal;
6178
6179         if (sb_rdonly(sb))
6180                 return 0;
6181
6182         journal = EXT4_SB(sb)->s_journal;
6183
6184         if (journal) {
6185                 /* Now we set up the journal barrier. */
6186                 jbd2_journal_lock_updates(journal);
6187
6188                 /*
6189                  * Don't clear the needs_recovery flag if we failed to
6190                  * flush the journal.
6191                  */
6192                 error = jbd2_journal_flush(journal, 0);
6193                 if (error < 0)
6194                         goto out;
6195
6196                 /* Journal blocked and flushed, clear needs_recovery flag. */
6197                 ext4_clear_feature_journal_needs_recovery(sb);
6198                 if (ext4_orphan_file_empty(sb))
6199                         ext4_clear_feature_orphan_present(sb);
6200         }
6201
6202         error = ext4_commit_super(sb);
6203 out:
6204         if (journal)
6205                 /* we rely on upper layer to stop further updates */
6206                 jbd2_journal_unlock_updates(journal);
6207         return error;
6208 }
6209
6210 /*
6211  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
6212  * flag here, even though the filesystem is not technically dirty yet.
6213  */
6214 static int ext4_unfreeze(struct super_block *sb)
6215 {
6216         if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
6217                 return 0;
6218
6219         if (EXT4_SB(sb)->s_journal) {
6220                 /* Reset the needs_recovery flag before the fs is unlocked. */
6221                 ext4_set_feature_journal_needs_recovery(sb);
6222                 if (ext4_has_feature_orphan_file(sb))
6223                         ext4_set_feature_orphan_present(sb);
6224         }
6225
6226         ext4_commit_super(sb);
6227         return 0;
6228 }
6229
6230 /*
6231  * Structure to save mount options for ext4_remount's benefit
6232  */
6233 struct ext4_mount_options {
6234         unsigned long s_mount_opt;
6235         unsigned long s_mount_opt2;
6236         kuid_t s_resuid;
6237         kgid_t s_resgid;
6238         unsigned long s_commit_interval;
6239         u32 s_min_batch_time, s_max_batch_time;
6240 #ifdef CONFIG_QUOTA
6241         int s_jquota_fmt;
6242         char *s_qf_names[EXT4_MAXQUOTAS];
6243 #endif
6244 };
6245
6246 static int __ext4_remount(struct fs_context *fc, struct super_block *sb)
6247 {
6248         struct ext4_fs_context *ctx = fc->fs_private;
6249         struct ext4_super_block *es;
6250         struct ext4_sb_info *sbi = EXT4_SB(sb);
6251         unsigned long old_sb_flags;
6252         struct ext4_mount_options old_opts;
6253         ext4_group_t g;
6254         int err = 0;
6255 #ifdef CONFIG_QUOTA
6256         int enable_quota = 0;
6257         int i, j;
6258         char *to_free[EXT4_MAXQUOTAS];
6259 #endif
6260
6261         ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
6262
6263         /* Store the original options */
6264         old_sb_flags = sb->s_flags;
6265         old_opts.s_mount_opt = sbi->s_mount_opt;
6266         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
6267         old_opts.s_resuid = sbi->s_resuid;
6268         old_opts.s_resgid = sbi->s_resgid;
6269         old_opts.s_commit_interval = sbi->s_commit_interval;
6270         old_opts.s_min_batch_time = sbi->s_min_batch_time;
6271         old_opts.s_max_batch_time = sbi->s_max_batch_time;
6272 #ifdef CONFIG_QUOTA
6273         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
6274         for (i = 0; i < EXT4_MAXQUOTAS; i++)
6275                 if (sbi->s_qf_names[i]) {
6276                         char *qf_name = get_qf_name(sb, sbi, i);
6277
6278                         old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL);
6279                         if (!old_opts.s_qf_names[i]) {
6280                                 for (j = 0; j < i; j++)
6281                                         kfree(old_opts.s_qf_names[j]);
6282                                 return -ENOMEM;
6283                         }
6284                 } else
6285                         old_opts.s_qf_names[i] = NULL;
6286 #endif
6287         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
6288                 ctx->journal_ioprio =
6289                         sbi->s_journal->j_task->io_context->ioprio;
6290
6291         ext4_apply_options(fc, sb);
6292
6293         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
6294             test_opt(sb, JOURNAL_CHECKSUM)) {
6295                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
6296                          "during remount not supported; ignoring");
6297                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
6298         }
6299
6300         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
6301                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
6302                         ext4_msg(sb, KERN_ERR, "can't mount with "
6303                                  "both data=journal and delalloc");
6304                         err = -EINVAL;
6305                         goto restore_opts;
6306                 }
6307                 if (test_opt(sb, DIOREAD_NOLOCK)) {
6308                         ext4_msg(sb, KERN_ERR, "can't mount with "
6309                                  "both data=journal and dioread_nolock");
6310                         err = -EINVAL;
6311                         goto restore_opts;
6312                 }
6313         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
6314                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
6315                         ext4_msg(sb, KERN_ERR, "can't mount with "
6316                                 "journal_async_commit in data=ordered mode");
6317                         err = -EINVAL;
6318                         goto restore_opts;
6319                 }
6320         }
6321
6322         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
6323                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
6324                 err = -EINVAL;
6325                 goto restore_opts;
6326         }
6327
6328         if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED))
6329                 ext4_abort(sb, ESHUTDOWN, "Abort forced by user");
6330
6331         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
6332                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
6333
6334         es = sbi->s_es;
6335
6336         if (sbi->s_journal) {
6337                 ext4_init_journal_params(sb, sbi->s_journal);
6338                 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio);
6339         }
6340
6341         /* Flush outstanding errors before changing fs state */
6342         flush_work(&sbi->s_error_work);
6343
6344         if ((bool)(fc->sb_flags & SB_RDONLY) != sb_rdonly(sb)) {
6345                 if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED)) {
6346                         err = -EROFS;
6347                         goto restore_opts;
6348                 }
6349
6350                 if (fc->sb_flags & SB_RDONLY) {
6351                         err = sync_filesystem(sb);
6352                         if (err < 0)
6353                                 goto restore_opts;
6354                         err = dquot_suspend(sb, -1);
6355                         if (err < 0)
6356                                 goto restore_opts;
6357
6358                         /*
6359                          * First of all, the unconditional stuff we have to do
6360                          * to disable replay of the journal when we next remount
6361                          */
6362                         sb->s_flags |= SB_RDONLY;
6363
6364                         /*
6365                          * OK, test if we are remounting a valid rw partition
6366                          * readonly, and if so set the rdonly flag and then
6367                          * mark the partition as valid again.
6368                          */
6369                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
6370                             (sbi->s_mount_state & EXT4_VALID_FS))
6371                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
6372
6373                         if (sbi->s_journal) {
6374                                 /*
6375                                  * We let remount-ro finish even if marking fs
6376                                  * as clean failed...
6377                                  */
6378                                 ext4_mark_recovery_complete(sb, es);
6379                         }
6380                 } else {
6381                         /* Make sure we can mount this feature set readwrite */
6382                         if (ext4_has_feature_readonly(sb) ||
6383                             !ext4_feature_set_ok(sb, 0)) {
6384                                 err = -EROFS;
6385                                 goto restore_opts;
6386                         }
6387                         /*
6388                          * Make sure the group descriptor checksums
6389                          * are sane.  If they aren't, refuse to remount r/w.
6390                          */
6391                         for (g = 0; g < sbi->s_groups_count; g++) {
6392                                 struct ext4_group_desc *gdp =
6393                                         ext4_get_group_desc(sb, g, NULL);
6394
6395                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
6396                                         ext4_msg(sb, KERN_ERR,
6397                "ext4_remount: Checksum for group %u failed (%u!=%u)",
6398                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
6399                                                le16_to_cpu(gdp->bg_checksum));
6400                                         err = -EFSBADCRC;
6401                                         goto restore_opts;
6402                                 }
6403                         }
6404
6405                         /*
6406                          * If we have an unprocessed orphan list hanging
6407                          * around from a previously readonly bdev mount,
6408                          * require a full umount/remount for now.
6409                          */
6410                         if (es->s_last_orphan || !ext4_orphan_file_empty(sb)) {
6411                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
6412                                        "remount RDWR because of unprocessed "
6413                                        "orphan inode list.  Please "
6414                                        "umount/remount instead");
6415                                 err = -EINVAL;
6416                                 goto restore_opts;
6417                         }
6418
6419                         /*
6420                          * Mounting a RDONLY partition read-write, so reread
6421                          * and store the current valid flag.  (It may have
6422                          * been changed by e2fsck since we originally mounted
6423                          * the partition.)
6424                          */
6425                         if (sbi->s_journal) {
6426                                 err = ext4_clear_journal_err(sb, es);
6427                                 if (err)
6428                                         goto restore_opts;
6429                         }
6430                         sbi->s_mount_state = le16_to_cpu(es->s_state);
6431
6432                         err = ext4_setup_super(sb, es, 0);
6433                         if (err)
6434                                 goto restore_opts;
6435
6436                         sb->s_flags &= ~SB_RDONLY;
6437                         if (ext4_has_feature_mmp(sb))
6438                                 if (ext4_multi_mount_protect(sb,
6439                                                 le64_to_cpu(es->s_mmp_block))) {
6440                                         err = -EROFS;
6441                                         goto restore_opts;
6442                                 }
6443 #ifdef CONFIG_QUOTA
6444                         enable_quota = 1;
6445 #endif
6446                 }
6447         }
6448
6449         /*
6450          * Reinitialize lazy itable initialization thread based on
6451          * current settings
6452          */
6453         if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
6454                 ext4_unregister_li_request(sb);
6455         else {
6456                 ext4_group_t first_not_zeroed;
6457                 first_not_zeroed = ext4_has_uninit_itable(sb);
6458                 ext4_register_li_request(sb, first_not_zeroed);
6459         }
6460
6461         /*
6462          * Handle creation of system zone data early because it can fail.
6463          * Releasing of existing data is done when we are sure remount will
6464          * succeed.
6465          */
6466         if (test_opt(sb, BLOCK_VALIDITY) && !sbi->s_system_blks) {
6467                 err = ext4_setup_system_zone(sb);
6468                 if (err)
6469                         goto restore_opts;
6470         }
6471
6472         if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) {
6473                 err = ext4_commit_super(sb);
6474                 if (err)
6475                         goto restore_opts;
6476         }
6477
6478 #ifdef CONFIG_QUOTA
6479         /* Release old quota file names */
6480         for (i = 0; i < EXT4_MAXQUOTAS; i++)
6481                 kfree(old_opts.s_qf_names[i]);
6482         if (enable_quota) {
6483                 if (sb_any_quota_suspended(sb))
6484                         dquot_resume(sb, -1);
6485                 else if (ext4_has_feature_quota(sb)) {
6486                         err = ext4_enable_quotas(sb);
6487                         if (err)
6488                                 goto restore_opts;
6489                 }
6490         }
6491 #endif
6492         if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks)
6493                 ext4_release_system_zone(sb);
6494
6495         if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb))
6496                 ext4_stop_mmpd(sbi);
6497
6498         return 0;
6499
6500 restore_opts:
6501         sb->s_flags = old_sb_flags;
6502         sbi->s_mount_opt = old_opts.s_mount_opt;
6503         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
6504         sbi->s_resuid = old_opts.s_resuid;
6505         sbi->s_resgid = old_opts.s_resgid;
6506         sbi->s_commit_interval = old_opts.s_commit_interval;
6507         sbi->s_min_batch_time = old_opts.s_min_batch_time;
6508         sbi->s_max_batch_time = old_opts.s_max_batch_time;
6509         if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks)
6510                 ext4_release_system_zone(sb);
6511 #ifdef CONFIG_QUOTA
6512         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
6513         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
6514                 to_free[i] = get_qf_name(sb, sbi, i);
6515                 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]);
6516         }
6517         synchronize_rcu();
6518         for (i = 0; i < EXT4_MAXQUOTAS; i++)
6519                 kfree(to_free[i]);
6520 #endif
6521         if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb))
6522                 ext4_stop_mmpd(sbi);
6523         return err;
6524 }
6525
6526 static int ext4_reconfigure(struct fs_context *fc)
6527 {
6528         struct super_block *sb = fc->root->d_sb;
6529         int ret;
6530
6531         fc->s_fs_info = EXT4_SB(sb);
6532
6533         ret = ext4_check_opt_consistency(fc, sb);
6534         if (ret < 0)
6535                 return ret;
6536
6537         ret = __ext4_remount(fc, sb);
6538         if (ret < 0)
6539                 return ret;
6540
6541         ext4_msg(sb, KERN_INFO, "re-mounted. Quota mode: %s.",
6542                  ext4_quota_mode(sb));
6543
6544         return 0;
6545 }
6546
6547 #ifdef CONFIG_QUOTA
6548 static int ext4_statfs_project(struct super_block *sb,
6549                                kprojid_t projid, struct kstatfs *buf)
6550 {
6551         struct kqid qid;
6552         struct dquot *dquot;
6553         u64 limit;
6554         u64 curblock;
6555
6556         qid = make_kqid_projid(projid);
6557         dquot = dqget(sb, qid);
6558         if (IS_ERR(dquot))
6559                 return PTR_ERR(dquot);
6560         spin_lock(&dquot->dq_dqb_lock);
6561
6562         limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit,
6563                              dquot->dq_dqb.dqb_bhardlimit);
6564         limit >>= sb->s_blocksize_bits;
6565
6566         if (limit && buf->f_blocks > limit) {
6567                 curblock = (dquot->dq_dqb.dqb_curspace +
6568                             dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits;
6569                 buf->f_blocks = limit;
6570                 buf->f_bfree = buf->f_bavail =
6571                         (buf->f_blocks > curblock) ?
6572                          (buf->f_blocks - curblock) : 0;
6573         }
6574
6575         limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit,
6576                              dquot->dq_dqb.dqb_ihardlimit);
6577         if (limit && buf->f_files > limit) {
6578                 buf->f_files = limit;
6579                 buf->f_ffree =
6580                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
6581                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
6582         }
6583
6584         spin_unlock(&dquot->dq_dqb_lock);
6585         dqput(dquot);
6586         return 0;
6587 }
6588 #endif
6589
6590 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
6591 {
6592         struct super_block *sb = dentry->d_sb;
6593         struct ext4_sb_info *sbi = EXT4_SB(sb);
6594         struct ext4_super_block *es = sbi->s_es;
6595         ext4_fsblk_t overhead = 0, resv_blocks;
6596         s64 bfree;
6597         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
6598
6599         if (!test_opt(sb, MINIX_DF))
6600                 overhead = sbi->s_overhead;
6601
6602         buf->f_type = EXT4_SUPER_MAGIC;
6603         buf->f_bsize = sb->s_blocksize;
6604         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
6605         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
6606                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
6607         /* prevent underflow in case that few free space is available */
6608         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
6609         buf->f_bavail = buf->f_bfree -
6610                         (ext4_r_blocks_count(es) + resv_blocks);
6611         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
6612                 buf->f_bavail = 0;
6613         buf->f_files = le32_to_cpu(es->s_inodes_count);
6614         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
6615         buf->f_namelen = EXT4_NAME_LEN;
6616         buf->f_fsid = uuid_to_fsid(es->s_uuid);
6617
6618 #ifdef CONFIG_QUOTA
6619         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
6620             sb_has_quota_limits_enabled(sb, PRJQUOTA))
6621                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
6622 #endif
6623         return 0;
6624 }
6625
6626
6627 #ifdef CONFIG_QUOTA
6628
6629 /*
6630  * Helper functions so that transaction is started before we acquire dqio_sem
6631  * to keep correct lock ordering of transaction > dqio_sem
6632  */
6633 static inline struct inode *dquot_to_inode(struct dquot *dquot)
6634 {
6635         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
6636 }
6637
6638 static int ext4_write_dquot(struct dquot *dquot)
6639 {
6640         int ret, err;
6641         handle_t *handle;
6642         struct inode *inode;
6643
6644         inode = dquot_to_inode(dquot);
6645         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
6646                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
6647         if (IS_ERR(handle))
6648                 return PTR_ERR(handle);
6649         ret = dquot_commit(dquot);
6650         err = ext4_journal_stop(handle);
6651         if (!ret)
6652                 ret = err;
6653         return ret;
6654 }
6655
6656 static int ext4_acquire_dquot(struct dquot *dquot)
6657 {
6658         int ret, err;
6659         handle_t *handle;
6660
6661         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
6662                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
6663         if (IS_ERR(handle))
6664                 return PTR_ERR(handle);
6665         ret = dquot_acquire(dquot);
6666         err = ext4_journal_stop(handle);
6667         if (!ret)
6668                 ret = err;
6669         return ret;
6670 }
6671
6672 static int ext4_release_dquot(struct dquot *dquot)
6673 {
6674         int ret, err;
6675         handle_t *handle;
6676
6677         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
6678                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
6679         if (IS_ERR(handle)) {
6680                 /* Release dquot anyway to avoid endless cycle in dqput() */
6681                 dquot_release(dquot);
6682                 return PTR_ERR(handle);
6683         }
6684         ret = dquot_release(dquot);
6685         err = ext4_journal_stop(handle);
6686         if (!ret)
6687                 ret = err;
6688         return ret;
6689 }
6690
6691 static int ext4_mark_dquot_dirty(struct dquot *dquot)
6692 {
6693         struct super_block *sb = dquot->dq_sb;
6694
6695         if (ext4_is_quota_journalled(sb)) {
6696                 dquot_mark_dquot_dirty(dquot);
6697                 return ext4_write_dquot(dquot);
6698         } else {
6699                 return dquot_mark_dquot_dirty(dquot);
6700         }
6701 }
6702
6703 static int ext4_write_info(struct super_block *sb, int type)
6704 {
6705         int ret, err;
6706         handle_t *handle;
6707
6708         /* Data block + inode block */
6709         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
6710         if (IS_ERR(handle))
6711                 return PTR_ERR(handle);
6712         ret = dquot_commit_info(sb, type);
6713         err = ext4_journal_stop(handle);
6714         if (!ret)
6715                 ret = err;
6716         return ret;
6717 }
6718
6719 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
6720 {
6721         struct ext4_inode_info *ei = EXT4_I(inode);
6722
6723         /* The first argument of lockdep_set_subclass has to be
6724          * *exactly* the same as the argument to init_rwsem() --- in
6725          * this case, in init_once() --- or lockdep gets unhappy
6726          * because the name of the lock is set using the
6727          * stringification of the argument to init_rwsem().
6728          */
6729         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
6730         lockdep_set_subclass(&ei->i_data_sem, subclass);
6731 }
6732
6733 /*
6734  * Standard function to be called on quota_on
6735  */
6736 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
6737                          const struct path *path)
6738 {
6739         int err;
6740
6741         if (!test_opt(sb, QUOTA))
6742                 return -EINVAL;
6743
6744         /* Quotafile not on the same filesystem? */
6745         if (path->dentry->d_sb != sb)
6746                 return -EXDEV;
6747
6748         /* Quota already enabled for this file? */
6749         if (IS_NOQUOTA(d_inode(path->dentry)))
6750                 return -EBUSY;
6751
6752         /* Journaling quota? */
6753         if (EXT4_SB(sb)->s_qf_names[type]) {
6754                 /* Quotafile not in fs root? */
6755                 if (path->dentry->d_parent != sb->s_root)
6756                         ext4_msg(sb, KERN_WARNING,
6757                                 "Quota file not on filesystem root. "
6758                                 "Journaled quota will not work");
6759                 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
6760         } else {
6761                 /*
6762                  * Clear the flag just in case mount options changed since
6763                  * last time.
6764                  */
6765                 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
6766         }
6767
6768         /*
6769          * When we journal data on quota file, we have to flush journal to see
6770          * all updates to the file when we bypass pagecache...
6771          */
6772         if (EXT4_SB(sb)->s_journal &&
6773             ext4_should_journal_data(d_inode(path->dentry))) {
6774                 /*
6775                  * We don't need to lock updates but journal_flush() could
6776                  * otherwise be livelocked...
6777                  */
6778                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
6779                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal, 0);
6780                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
6781                 if (err)
6782                         return err;
6783         }
6784
6785         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
6786         err = dquot_quota_on(sb, type, format_id, path);
6787         if (!err) {
6788                 struct inode *inode = d_inode(path->dentry);
6789                 handle_t *handle;
6790
6791                 /*
6792                  * Set inode flags to prevent userspace from messing with quota
6793                  * files. If this fails, we return success anyway since quotas
6794                  * are already enabled and this is not a hard failure.
6795                  */
6796                 inode_lock(inode);
6797                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
6798                 if (IS_ERR(handle))
6799                         goto unlock_inode;
6800                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
6801                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
6802                                 S_NOATIME | S_IMMUTABLE);
6803                 err = ext4_mark_inode_dirty(handle, inode);
6804                 ext4_journal_stop(handle);
6805         unlock_inode:
6806                 inode_unlock(inode);
6807                 if (err)
6808                         dquot_quota_off(sb, type);
6809         }
6810         if (err)
6811                 lockdep_set_quota_inode(path->dentry->d_inode,
6812                                              I_DATA_SEM_NORMAL);
6813         return err;
6814 }
6815
6816 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
6817                              unsigned int flags)
6818 {
6819         int err;
6820         struct inode *qf_inode;
6821         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
6822                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
6823                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
6824                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
6825         };
6826
6827         BUG_ON(!ext4_has_feature_quota(sb));
6828
6829         if (!qf_inums[type])
6830                 return -EPERM;
6831
6832         qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL);
6833         if (IS_ERR(qf_inode)) {
6834                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
6835                 return PTR_ERR(qf_inode);
6836         }
6837
6838         /* Don't account quota for quota files to avoid recursion */
6839         qf_inode->i_flags |= S_NOQUOTA;
6840         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
6841         err = dquot_load_quota_inode(qf_inode, type, format_id, flags);
6842         if (err)
6843                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
6844         iput(qf_inode);
6845
6846         return err;
6847 }
6848
6849 /* Enable usage tracking for all quota types. */
6850 int ext4_enable_quotas(struct super_block *sb)
6851 {
6852         int type, err = 0;
6853         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
6854                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
6855                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
6856                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
6857         };
6858         bool quota_mopt[EXT4_MAXQUOTAS] = {
6859                 test_opt(sb, USRQUOTA),
6860                 test_opt(sb, GRPQUOTA),
6861                 test_opt(sb, PRJQUOTA),
6862         };
6863
6864         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
6865         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
6866                 if (qf_inums[type]) {
6867                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
6868                                 DQUOT_USAGE_ENABLED |
6869                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
6870                         if (err) {
6871                                 ext4_warning(sb,
6872                                         "Failed to enable quota tracking "
6873                                         "(type=%d, err=%d). Please run "
6874                                         "e2fsck to fix.", type, err);
6875                                 for (type--; type >= 0; type--) {
6876                                         struct inode *inode;
6877
6878                                         inode = sb_dqopt(sb)->files[type];
6879                                         if (inode)
6880                                                 inode = igrab(inode);
6881                                         dquot_quota_off(sb, type);
6882                                         if (inode) {
6883                                                 lockdep_set_quota_inode(inode,
6884                                                         I_DATA_SEM_NORMAL);
6885                                                 iput(inode);
6886                                         }
6887                                 }
6888
6889                                 return err;
6890                         }
6891                 }
6892         }
6893         return 0;
6894 }
6895
6896 static int ext4_quota_off(struct super_block *sb, int type)
6897 {
6898         struct inode *inode = sb_dqopt(sb)->files[type];
6899         handle_t *handle;
6900         int err;
6901
6902         /* Force all delayed allocation blocks to be allocated.
6903          * Caller already holds s_umount sem */
6904         if (test_opt(sb, DELALLOC))
6905                 sync_filesystem(sb);
6906
6907         if (!inode || !igrab(inode))
6908                 goto out;
6909
6910         err = dquot_quota_off(sb, type);
6911         if (err || ext4_has_feature_quota(sb))
6912                 goto out_put;
6913
6914         inode_lock(inode);
6915         /*
6916          * Update modification times of quota files when userspace can
6917          * start looking at them. If we fail, we return success anyway since
6918          * this is not a hard failure and quotas are already disabled.
6919          */
6920         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
6921         if (IS_ERR(handle)) {
6922                 err = PTR_ERR(handle);
6923                 goto out_unlock;
6924         }
6925         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
6926         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
6927         inode->i_mtime = inode->i_ctime = current_time(inode);
6928         err = ext4_mark_inode_dirty(handle, inode);
6929         ext4_journal_stop(handle);
6930 out_unlock:
6931         inode_unlock(inode);
6932 out_put:
6933         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
6934         iput(inode);
6935         return err;
6936 out:
6937         return dquot_quota_off(sb, type);
6938 }
6939
6940 /* Read data from quotafile - avoid pagecache and such because we cannot afford
6941  * acquiring the locks... As quota files are never truncated and quota code
6942  * itself serializes the operations (and no one else should touch the files)
6943  * we don't have to be afraid of races */
6944 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
6945                                size_t len, loff_t off)
6946 {
6947         struct inode *inode = sb_dqopt(sb)->files[type];
6948         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
6949         int offset = off & (sb->s_blocksize - 1);
6950         int tocopy;
6951         size_t toread;
6952         struct buffer_head *bh;
6953         loff_t i_size = i_size_read(inode);
6954
6955         if (off > i_size)
6956                 return 0;
6957         if (off+len > i_size)
6958                 len = i_size-off;
6959         toread = len;
6960         while (toread > 0) {
6961                 tocopy = sb->s_blocksize - offset < toread ?
6962                                 sb->s_blocksize - offset : toread;
6963                 bh = ext4_bread(NULL, inode, blk, 0);
6964                 if (IS_ERR(bh))
6965                         return PTR_ERR(bh);
6966                 if (!bh)        /* A hole? */
6967                         memset(data, 0, tocopy);
6968                 else
6969                         memcpy(data, bh->b_data+offset, tocopy);
6970                 brelse(bh);
6971                 offset = 0;
6972                 toread -= tocopy;
6973                 data += tocopy;
6974                 blk++;
6975         }
6976         return len;
6977 }
6978
6979 /* Write to quotafile (we know the transaction is already started and has
6980  * enough credits) */
6981 static ssize_t ext4_quota_write(struct super_block *sb, int type,
6982                                 const char *data, size_t len, loff_t off)
6983 {
6984         struct inode *inode = sb_dqopt(sb)->files[type];
6985         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
6986         int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1);
6987         int retries = 0;
6988         struct buffer_head *bh;
6989         handle_t *handle = journal_current_handle();
6990
6991         if (!handle) {
6992                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
6993                         " cancelled because transaction is not started",
6994                         (unsigned long long)off, (unsigned long long)len);
6995                 return -EIO;
6996         }
6997         /*
6998          * Since we account only one data block in transaction credits,
6999          * then it is impossible to cross a block boundary.
7000          */
7001         if (sb->s_blocksize - offset < len) {
7002                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
7003                         " cancelled because not block aligned",
7004                         (unsigned long long)off, (unsigned long long)len);
7005                 return -EIO;
7006         }
7007
7008         do {
7009                 bh = ext4_bread(handle, inode, blk,
7010                                 EXT4_GET_BLOCKS_CREATE |
7011                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
7012         } while (PTR_ERR(bh) == -ENOSPC &&
7013                  ext4_should_retry_alloc(inode->i_sb, &retries));
7014         if (IS_ERR(bh))
7015                 return PTR_ERR(bh);
7016         if (!bh)
7017                 goto out;
7018         BUFFER_TRACE(bh, "get write access");
7019         err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
7020         if (err) {
7021                 brelse(bh);
7022                 return err;
7023         }
7024         lock_buffer(bh);
7025         memcpy(bh->b_data+offset, data, len);
7026         flush_dcache_page(bh->b_page);
7027         unlock_buffer(bh);
7028         err = ext4_handle_dirty_metadata(handle, NULL, bh);
7029         brelse(bh);
7030 out:
7031         if (inode->i_size < off + len) {
7032                 i_size_write(inode, off + len);
7033                 EXT4_I(inode)->i_disksize = inode->i_size;
7034                 err2 = ext4_mark_inode_dirty(handle, inode);
7035                 if (unlikely(err2 && !err))
7036                         err = err2;
7037         }
7038         return err ? err : len;
7039 }
7040 #endif
7041
7042 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
7043 static inline void register_as_ext2(void)
7044 {
7045         int err = register_filesystem(&ext2_fs_type);
7046         if (err)
7047                 printk(KERN_WARNING
7048                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
7049 }
7050
7051 static inline void unregister_as_ext2(void)
7052 {
7053         unregister_filesystem(&ext2_fs_type);
7054 }
7055
7056 static inline int ext2_feature_set_ok(struct super_block *sb)
7057 {
7058         if (ext4_has_unknown_ext2_incompat_features(sb))
7059                 return 0;
7060         if (sb_rdonly(sb))
7061                 return 1;
7062         if (ext4_has_unknown_ext2_ro_compat_features(sb))
7063                 return 0;
7064         return 1;
7065 }
7066 #else
7067 static inline void register_as_ext2(void) { }
7068 static inline void unregister_as_ext2(void) { }
7069 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
7070 #endif
7071
7072 static inline void register_as_ext3(void)
7073 {
7074         int err = register_filesystem(&ext3_fs_type);
7075         if (err)
7076                 printk(KERN_WARNING
7077                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
7078 }
7079
7080 static inline void unregister_as_ext3(void)
7081 {
7082         unregister_filesystem(&ext3_fs_type);
7083 }
7084
7085 static inline int ext3_feature_set_ok(struct super_block *sb)
7086 {
7087         if (ext4_has_unknown_ext3_incompat_features(sb))
7088                 return 0;
7089         if (!ext4_has_feature_journal(sb))
7090                 return 0;
7091         if (sb_rdonly(sb))
7092                 return 1;
7093         if (ext4_has_unknown_ext3_ro_compat_features(sb))
7094                 return 0;
7095         return 1;
7096 }
7097
7098 static struct file_system_type ext4_fs_type = {
7099         .owner                  = THIS_MODULE,
7100         .name                   = "ext4",
7101         .init_fs_context        = ext4_init_fs_context,
7102         .parameters             = ext4_param_specs,
7103         .kill_sb                = kill_block_super,
7104         .fs_flags               = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
7105 };
7106 MODULE_ALIAS_FS("ext4");
7107
7108 /* Shared across all ext4 file systems */
7109 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
7110
7111 static int __init ext4_init_fs(void)
7112 {
7113         int i, err;
7114
7115         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
7116         ext4_li_info = NULL;
7117
7118         /* Build-time check for flags consistency */
7119         ext4_check_flag_values();
7120
7121         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
7122                 init_waitqueue_head(&ext4__ioend_wq[i]);
7123
7124         err = ext4_init_es();
7125         if (err)
7126                 return err;
7127
7128         err = ext4_init_pending();
7129         if (err)
7130                 goto out7;
7131
7132         err = ext4_init_post_read_processing();
7133         if (err)
7134                 goto out6;
7135
7136         err = ext4_init_pageio();
7137         if (err)
7138                 goto out5;
7139
7140         err = ext4_init_system_zone();
7141         if (err)
7142                 goto out4;
7143
7144         err = ext4_init_sysfs();
7145         if (err)
7146                 goto out3;
7147
7148         err = ext4_init_mballoc();
7149         if (err)
7150                 goto out2;
7151         err = init_inodecache();
7152         if (err)
7153                 goto out1;
7154
7155         err = ext4_fc_init_dentry_cache();
7156         if (err)
7157                 goto out05;
7158
7159         register_as_ext3();
7160         register_as_ext2();
7161         err = register_filesystem(&ext4_fs_type);
7162         if (err)
7163                 goto out;
7164
7165         return 0;
7166 out:
7167         unregister_as_ext2();
7168         unregister_as_ext3();
7169         ext4_fc_destroy_dentry_cache();
7170 out05:
7171         destroy_inodecache();
7172 out1:
7173         ext4_exit_mballoc();
7174 out2:
7175         ext4_exit_sysfs();
7176 out3:
7177         ext4_exit_system_zone();
7178 out4:
7179         ext4_exit_pageio();
7180 out5:
7181         ext4_exit_post_read_processing();
7182 out6:
7183         ext4_exit_pending();
7184 out7:
7185         ext4_exit_es();
7186
7187         return err;
7188 }
7189
7190 static void __exit ext4_exit_fs(void)
7191 {
7192         ext4_destroy_lazyinit_thread();
7193         unregister_as_ext2();
7194         unregister_as_ext3();
7195         unregister_filesystem(&ext4_fs_type);
7196         ext4_fc_destroy_dentry_cache();
7197         destroy_inodecache();
7198         ext4_exit_mballoc();
7199         ext4_exit_sysfs();
7200         ext4_exit_system_zone();
7201         ext4_exit_pageio();
7202         ext4_exit_post_read_processing();
7203         ext4_exit_es();
7204         ext4_exit_pending();
7205 }
7206
7207 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
7208 MODULE_DESCRIPTION("Fourth Extended Filesystem");
7209 MODULE_LICENSE("GPL");
7210 MODULE_SOFTDEP("pre: crc32c");
7211 module_init(ext4_init_fs)
7212 module_exit(ext4_exit_fs)