fs/ext4/super.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/ext4/super.c
   4  *
   5  * Copyright (C) 1992, 1993, 1994, 1995
   6  * Remy Card (card@masi.ibp.fr)
   7  * Laboratoire MASI - Institut Blaise Pascal
   8  * Universite Pierre et Marie Curie (Paris VI)
   9  *
  10  *  from
  11  *
  12  *  linux/fs/minix/inode.c
  13  *
  14  *  Copyright (C) 1991, 1992  Linus Torvalds
  15  *
  16  *  Big-endian to little-endian byte-swapping/bitmaps by
  17  *        David S. Miller (davem@caip.rutgers.edu), 1995
  18  */
  19
  20 #include <linux/module.h>
  21 #include <linux/string.h>
  22 #include <linux/fs.h>
  23 #include <linux/time.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/backing-dev.h>
  29 #include <linux/parser.h>
  30 #include <linux/buffer_head.h>
  31 #include <linux/exportfs.h>
  32 #include <linux/vfs.h>
  33 #include <linux/random.h>
  34 #include <linux/mount.h>
  35 #include <linux/namei.h>
  36 #include <linux/quotaops.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/dax.h>
  42 #include <linux/uaccess.h>
  43 #include <linux/iversion.h>
  44 #include <linux/unicode.h>
  45 #include <linux/part_stat.h>
  46 #include <linux/kthread.h>
  47 #include <linux/freezer.h>
  48 #include <linux/fsnotify.h>
  49 #include <linux/fs_context.h>
  50 #include <linux/fs_parser.h>
  51
  52 #include "ext4.h"
  53 #include "ext4_extents.h"       /* Needed for trace points definition */
  54 #include "ext4_jbd2.h"
  55 #include "xattr.h"
  56 #include "acl.h"
  57 #include "mballoc.h"
  58 #include "fsmap.h"
  59
  60 #define CREATE_TRACE_POINTS
  61 #include <trace/events/ext4.h>
  62
  63 static struct ext4_lazy_init *ext4_li_info;
  64 static DEFINE_MUTEX(ext4_li_mtx);
  65 static struct ratelimit_state ext4_mount_msg_ratelimit;
  66
  67 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  68                              unsigned long journal_devnum);
  69 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  70 static void ext4_update_super(struct super_block *sb);
  71 static int ext4_commit_super(struct super_block *sb);
  72 static int ext4_mark_recovery_complete(struct super_block *sb,
  73                                         struct ext4_super_block *es);
  74 static int ext4_clear_journal_err(struct super_block *sb,
  75                                   struct ext4_super_block *es);
  76 static int ext4_sync_fs(struct super_block *sb, int wait);
  77 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  78 static int ext4_unfreeze(struct super_block *sb);
  79 static int ext4_freeze(struct super_block *sb);
  80 static inline int ext2_feature_set_ok(struct super_block *sb);
  81 static inline int ext3_feature_set_ok(struct super_block *sb);
  82 static void ext4_destroy_lazyinit_thread(void);
  83 static void ext4_unregister_li_request(struct super_block *sb);
  84 static void ext4_clear_request_list(void);
  85 static struct inode *ext4_get_journal_inode(struct super_block *sb,
  86                                             unsigned int journal_inum);
  87 static int ext4_validate_options(struct fs_context *fc);
  88 static int ext4_check_opt_consistency(struct fs_context *fc,
  89                                       struct super_block *sb);
  90 static void ext4_apply_options(struct fs_context *fc, struct super_block *sb);
  91 static int ext4_parse_param(struct fs_context *fc, struct fs_parameter *param);
  92 static int ext4_get_tree(struct fs_context *fc);
  93 static int ext4_reconfigure(struct fs_context *fc);
  94 static void ext4_fc_free(struct fs_context *fc);
  95 static int ext4_init_fs_context(struct fs_context *fc);
  96 static const struct fs_parameter_spec ext4_param_specs[];
  97
  98 /*
  99  * Lock ordering
 100  *
 101  * page fault path:
 102  * mmap_lock -> sb_start_pagefault -> invalidate_lock (r) -> transaction start
 103  *   -> page lock -> i_data_sem (rw)
 104  *
 105  * buffered write path:
 106  * sb_start_write -> i_mutex -> mmap_lock
 107  * sb_start_write -> i_mutex -> transaction start -> page lock ->
 108  *   i_data_sem (rw)
 109  *
 110  * truncate:
 111  * sb_start_write -> i_mutex -> invalidate_lock (w) -> i_mmap_rwsem (w) ->
 112  *   page lock
 113  * sb_start_write -> i_mutex -> invalidate_lock (w) -> transaction start ->
 114  *   i_data_sem (rw)
 115  *
 116  * direct IO:
 117  * sb_start_write -> i_mutex -> mmap_lock
 118  * sb_start_write -> i_mutex -> transaction start -> i_data_sem (rw)
 119  *
 120  * writepages:
 121  * transaction start -> page lock(s) -> i_data_sem (rw)
 122  */
 123
 124 static const struct fs_context_operations ext4_context_ops = {
 125         .parse_param    = ext4_parse_param,
 126         .get_tree       = ext4_get_tree,
 127         .reconfigure    = ext4_reconfigure,
 128         .free           = ext4_fc_free,
 129 };
 130
 131
 132 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
 133 static struct file_system_type ext2_fs_type = {
 134         .owner                  = THIS_MODULE,
 135         .name                   = "ext2",
 136         .init_fs_context        = ext4_init_fs_context,
 137         .parameters             = ext4_param_specs,
 138         .kill_sb                = kill_block_super,
 139         .fs_flags               = FS_REQUIRES_DEV,
 140 };
 141 MODULE_ALIAS_FS("ext2");
 142 MODULE_ALIAS("ext2");
 143 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
 144 #else
 145 #define IS_EXT2_SB(sb) (0)
 146 #endif
 147
 148
 149 static struct file_system_type ext3_fs_type = {
 150         .owner                  = THIS_MODULE,
 151         .name                   = "ext3",
 152         .init_fs_context        = ext4_init_fs_context,
 153         .parameters             = ext4_param_specs,
 154         .kill_sb                = kill_block_super,
 155         .fs_flags               = FS_REQUIRES_DEV,
 156 };
 157 MODULE_ALIAS_FS("ext3");
 158 MODULE_ALIAS("ext3");
 159 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 160
 161
 162 static inline void __ext4_read_bh(struct buffer_head *bh, blk_opf_t op_flags,
 163                                   bh_end_io_t *end_io)
 164 {
 165         /*
 166          * buffer's verified bit is no longer valid after reading from
 167          * disk again due to write out error, clear it to make sure we
 168          * recheck the buffer contents.
 169          */
 170         clear_buffer_verified(bh);
 171
 172         bh->b_end_io = end_io ? end_io : end_buffer_read_sync;
 173         get_bh(bh);
 174         submit_bh(REQ_OP_READ | op_flags, bh);
 175 }
 176
 177 void ext4_read_bh_nowait(struct buffer_head *bh, blk_opf_t op_flags,
 178                          bh_end_io_t *end_io)
 179 {
 180         BUG_ON(!buffer_locked(bh));
 181
 182         if (ext4_buffer_uptodate(bh)) {
 183                 unlock_buffer(bh);
 184                 return;
 185         }
 186         __ext4_read_bh(bh, op_flags, end_io);
 187 }
 188
 189 int ext4_read_bh(struct buffer_head *bh, blk_opf_t op_flags, bh_end_io_t *end_io)
 190 {
 191         BUG_ON(!buffer_locked(bh));
 192
 193         if (ext4_buffer_uptodate(bh)) {
 194                 unlock_buffer(bh);
 195                 return 0;
 196         }
 197
 198         __ext4_read_bh(bh, op_flags, end_io);
 199
 200         wait_on_buffer(bh);
 201         if (buffer_uptodate(bh))
 202                 return 0;
 203         return -EIO;
 204 }
 205
 206 int ext4_read_bh_lock(struct buffer_head *bh, blk_opf_t op_flags, bool wait)
 207 {
 208         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,
 231                                                blk_opf_t op_flags, 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                                    blk_opf_t op_flags)
 252 {
 253         return __ext4_sb_bread_gfp(sb, block, op_flags, __GFP_MOVABLE);
 254 }
 255
 256 struct buffer_head *ext4_sb_bread_unmovable(struct super_block *sb,
 257                                             sector_t block)
 258 {
 259         return __ext4_sb_bread_gfp(sb, block, 0, 0);
 260 }
 261
 262 void ext4_sb_breadahead_unmovable(struct super_block *sb, sector_t block)
 263 {
 264         struct buffer_head *bh = sb_getblk_gfp(sb, block, 0);
 265
 266         if (likely(bh)) {
 267                 ext4_read_bh_lock(bh, REQ_RAHEAD, false);
 268                 brelse(bh);
 269         }
 270 }
 271
 272 static int ext4_verify_csum_type(struct super_block *sb,
 273                                  struct ext4_super_block *es)
 274 {
 275         if (!ext4_has_feature_metadata_csum(sb))
 276                 return 1;
 277
 278         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 279 }
 280
 281 __le32 ext4_superblock_csum(struct super_block *sb,
 282                             struct ext4_super_block *es)
 283 {
 284         struct ext4_sb_info *sbi = EXT4_SB(sb);
 285         int offset = offsetof(struct ext4_super_block, s_checksum);
 286         __u32 csum;
 287
 288         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 289
 290         return cpu_to_le32(csum);
 291 }
 292
 293 static int ext4_superblock_csum_verify(struct super_block *sb,
 294                                        struct ext4_super_block *es)
 295 {
 296         if (!ext4_has_metadata_csum(sb))
 297                 return 1;
 298
 299         return es->s_checksum == ext4_superblock_csum(sb, es);
 300 }
 301
 302 void ext4_superblock_csum_set(struct super_block *sb)
 303 {
 304         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 305
 306         if (!ext4_has_metadata_csum(sb))
 307                 return;
 308
 309         es->s_checksum = ext4_superblock_csum(sb, es);
 310 }
 311
 312 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 313                                struct ext4_group_desc *bg)
 314 {
 315         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 316                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 317                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 318 }
 319
 320 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 321                                struct ext4_group_desc *bg)
 322 {
 323         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 324                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 325                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 326 }
 327
 328 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 329                               struct ext4_group_desc *bg)
 330 {
 331         return le32_to_cpu(bg->bg_inode_table_lo) |
 332                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 333                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 334 }
 335
 336 __u32 ext4_free_group_clusters(struct super_block *sb,
 337                                struct ext4_group_desc *bg)
 338 {
 339         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 340                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 341                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 342 }
 343
 344 __u32 ext4_free_inodes_count(struct super_block *sb,
 345                               struct ext4_group_desc *bg)
 346 {
 347         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 348                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 349                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 350 }
 351
 352 __u32 ext4_used_dirs_count(struct super_block *sb,
 353                               struct ext4_group_desc *bg)
 354 {
 355         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 356                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 357                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 358 }
 359
 360 __u32 ext4_itable_unused_count(struct super_block *sb,
 361                               struct ext4_group_desc *bg)
 362 {
 363         return le16_to_cpu(bg->bg_itable_unused_lo) |
 364                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 365                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 366 }
 367
 368 void ext4_block_bitmap_set(struct super_block *sb,
 369                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 370 {
 371         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 372         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 373                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 374 }
 375
 376 void ext4_inode_bitmap_set(struct super_block *sb,
 377                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 378 {
 379         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 380         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 381                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 382 }
 383
 384 void ext4_inode_table_set(struct super_block *sb,
 385                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 386 {
 387         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 388         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 389                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 390 }
 391
 392 void ext4_free_group_clusters_set(struct super_block *sb,
 393                                   struct ext4_group_desc *bg, __u32 count)
 394 {
 395         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 396         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 397                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 398 }
 399
 400 void ext4_free_inodes_set(struct super_block *sb,
 401                           struct ext4_group_desc *bg, __u32 count)
 402 {
 403         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 404         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 405                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 406 }
 407
 408 void ext4_used_dirs_set(struct super_block *sb,
 409                           struct ext4_group_desc *bg, __u32 count)
 410 {
 411         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 412         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 413                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 414 }
 415
 416 void ext4_itable_unused_set(struct super_block *sb,
 417                           struct ext4_group_desc *bg, __u32 count)
 418 {
 419         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 420         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 421                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 422 }
 423
 424 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi, time64_t now)
 425 {
 426         now = clamp_val(now, 0, (1ull << 40) - 1);
 427
 428         *lo = cpu_to_le32(lower_32_bits(now));
 429         *hi = upper_32_bits(now);
 430 }
 431
 432 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi)
 433 {
 434         return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo);
 435 }
 436 #define ext4_update_tstamp(es, tstamp) \
 437         __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi, \
 438                              ktime_get_real_seconds())
 439 #define ext4_get_tstamp(es, tstamp) \
 440         __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
 441
 442 /*
 443  * The del_gendisk() function uninitializes the disk-specific data
 444  * structures, including the bdi structure, without telling anyone
 445  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 446  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 447  * This is a kludge to prevent these oops until we can put in a proper
 448  * hook in del_gendisk() to inform the VFS and file system layers.
 449  */
 450 static int block_device_ejected(struct super_block *sb)
 451 {
 452         struct inode *bd_inode = sb->s_bdev->bd_inode;
 453         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 454
 455         return bdi->dev == NULL;
 456 }
 457
 458 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 459 {
 460         struct super_block              *sb = journal->j_private;
 461         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 462         int                             error = is_journal_aborted(journal);
 463         struct ext4_journal_cb_entry    *jce;
 464
 465         BUG_ON(txn->t_state == T_FINISHED);
 466
 467         ext4_process_freed_data(sb, txn->t_tid);
 468
 469         spin_lock(&sbi->s_md_lock);
 470         while (!list_empty(&txn->t_private_list)) {
 471                 jce = list_entry(txn->t_private_list.next,
 472                                  struct ext4_journal_cb_entry, jce_list);
 473                 list_del_init(&jce->jce_list);
 474                 spin_unlock(&sbi->s_md_lock);
 475                 jce->jce_func(sb, jce, error);
 476                 spin_lock(&sbi->s_md_lock);
 477         }
 478         spin_unlock(&sbi->s_md_lock);
 479 }
 480
 481 /*
 482  * This writepage callback for write_cache_pages()
 483  * takes care of a few cases after page cleaning.
 484  *
 485  * write_cache_pages() already checks for dirty pages
 486  * and calls clear_page_dirty_for_io(), which we want,
 487  * to write protect the pages.
 488  *
 489  * However, we may have to redirty a page (see below.)
 490  */
 491 static int ext4_journalled_writepage_callback(struct page *page,
 492                                               struct writeback_control *wbc,
 493                                               void *data)
 494 {
 495         transaction_t *transaction = (transaction_t *) data;
 496         struct buffer_head *bh, *head;
 497         struct journal_head *jh;
 498
 499         bh = head = page_buffers(page);
 500         do {
 501                 /*
 502                  * We have to redirty a page in these cases:
 503                  * 1) If buffer is dirty, it means the page was dirty because it
 504                  * contains a buffer that needs checkpointing. So the dirty bit
 505                  * needs to be preserved so that checkpointing writes the buffer
 506                  * properly.
 507                  * 2) If buffer is not part of the committing transaction
 508                  * (we may have just accidentally come across this buffer because
 509                  * inode range tracking is not exact) or if the currently running
 510                  * transaction already contains this buffer as well, dirty bit
 511                  * needs to be preserved so that the buffer gets writeprotected
 512                  * properly on running transaction's commit.
 513                  */
 514                 jh = bh2jh(bh);
 515                 if (buffer_dirty(bh) ||
 516                     (jh && (jh->b_transaction != transaction ||
 517                             jh->b_next_transaction))) {
 518                         redirty_page_for_writepage(wbc, page);
 519                         goto out;
 520                 }
 521         } while ((bh = bh->b_this_page) != head);
 522
 523 out:
 524         return AOP_WRITEPAGE_ACTIVATE;
 525 }
 526
 527 static int ext4_journalled_submit_inode_data_buffers(struct jbd2_inode *jinode)
 528 {
 529         struct address_space *mapping = jinode->i_vfs_inode->i_mapping;
 530         struct writeback_control wbc = {
 531                 .sync_mode =  WB_SYNC_ALL,
 532                 .nr_to_write = LONG_MAX,
 533                 .range_start = jinode->i_dirty_start,
 534                 .range_end = jinode->i_dirty_end,
 535         };
 536
 537         return write_cache_pages(mapping, &wbc,
 538                                  ext4_journalled_writepage_callback,
 539                                  jinode->i_transaction);
 540 }
 541
 542 static int ext4_journal_submit_inode_data_buffers(struct jbd2_inode *jinode)
 543 {
 544         int ret;
 545
 546         if (ext4_should_journal_data(jinode->i_vfs_inode))
 547                 ret = ext4_journalled_submit_inode_data_buffers(jinode);
 548         else
 549                 ret = jbd2_journal_submit_inode_data_buffers(jinode);
 550
 551         return ret;
 552 }
 553
 554 static int ext4_journal_finish_inode_data_buffers(struct jbd2_inode *jinode)
 555 {
 556         int ret = 0;
 557
 558         if (!ext4_should_journal_data(jinode->i_vfs_inode))
 559                 ret = jbd2_journal_finish_inode_data_buffers(jinode);
 560
 561         return ret;
 562 }
 563
 564 static bool system_going_down(void)
 565 {
 566         return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF
 567                 || system_state == SYSTEM_RESTART;
 568 }
 569
 570 struct ext4_err_translation {
 571         int code;
 572         int errno;
 573 };
 574
 575 #define EXT4_ERR_TRANSLATE(err) { .code = EXT4_ERR_##err, .errno = err }
 576
 577 static struct ext4_err_translation err_translation[] = {
 578         EXT4_ERR_TRANSLATE(EIO),
 579         EXT4_ERR_TRANSLATE(ENOMEM),
 580         EXT4_ERR_TRANSLATE(EFSBADCRC),
 581         EXT4_ERR_TRANSLATE(EFSCORRUPTED),
 582         EXT4_ERR_TRANSLATE(ENOSPC),
 583         EXT4_ERR_TRANSLATE(ENOKEY),
 584         EXT4_ERR_TRANSLATE(EROFS),
 585         EXT4_ERR_TRANSLATE(EFBIG),
 586         EXT4_ERR_TRANSLATE(EEXIST),
 587         EXT4_ERR_TRANSLATE(ERANGE),
 588         EXT4_ERR_TRANSLATE(EOVERFLOW),
 589         EXT4_ERR_TRANSLATE(EBUSY),
 590         EXT4_ERR_TRANSLATE(ENOTDIR),
 591         EXT4_ERR_TRANSLATE(ENOTEMPTY),
 592         EXT4_ERR_TRANSLATE(ESHUTDOWN),
 593         EXT4_ERR_TRANSLATE(EFAULT),
 594 };
 595
 596 static int ext4_errno_to_code(int errno)
 597 {
 598         int i;
 599
 600         for (i = 0; i < ARRAY_SIZE(err_translation); i++)
 601                 if (err_translation[i].errno == errno)
 602                         return err_translation[i].code;
 603         return EXT4_ERR_UNKNOWN;
 604 }
 605
 606 static void save_error_info(struct super_block *sb, int error,
 607                             __u32 ino, __u64 block,
 608                             const char *func, unsigned int line)
 609 {
 610         struct ext4_sb_info *sbi = EXT4_SB(sb);
 611
 612         /* We default to EFSCORRUPTED error... */
 613         if (error == 0)
 614                 error = EFSCORRUPTED;
 615
 616         spin_lock(&sbi->s_error_lock);
 617         sbi->s_add_error_count++;
 618         sbi->s_last_error_code = error;
 619         sbi->s_last_error_line = line;
 620         sbi->s_last_error_ino = ino;
 621         sbi->s_last_error_block = block;
 622         sbi->s_last_error_func = func;
 623         sbi->s_last_error_time = ktime_get_real_seconds();
 624         if (!sbi->s_first_error_time) {
 625                 sbi->s_first_error_code = error;
 626                 sbi->s_first_error_line = line;
 627                 sbi->s_first_error_ino = ino;
 628                 sbi->s_first_error_block = block;
 629                 sbi->s_first_error_func = func;
 630                 sbi->s_first_error_time = sbi->s_last_error_time;
 631         }
 632         spin_unlock(&sbi->s_error_lock);
 633 }
 634
 635 /* Deal with the reporting of failure conditions on a filesystem such as
 636  * inconsistencies detected or read IO failures.
 637  *
 638  * On ext2, we can store the error state of the filesystem in the
 639  * superblock.  That is not possible on ext4, because we may have other
 640  * write ordering constraints on the superblock which prevent us from
 641  * writing it out straight away; and given that the journal is about to
 642  * be aborted, we can't rely on the current, or future, transactions to
 643  * write out the superblock safely.
 644  *
 645  * We'll just use the jbd2_journal_abort() error code to record an error in
 646  * the journal instead.  On recovery, the journal will complain about
 647  * that error until we've noted it down and cleared it.
 648  *
 649  * If force_ro is set, we unconditionally force the filesystem into an
 650  * ABORT|READONLY state, unless the error response on the fs has been set to
 651  * panic in which case we take the easy way out and panic immediately. This is
 652  * used to deal with unrecoverable failures such as journal IO errors or ENOMEM
 653  * at a critical moment in log management.
 654  */
 655 static void ext4_handle_error(struct super_block *sb, bool force_ro, int error,
 656                               __u32 ino, __u64 block,
 657                               const char *func, unsigned int line)
 658 {
 659         journal_t *journal = EXT4_SB(sb)->s_journal;
 660         bool continue_fs = !force_ro && test_opt(sb, ERRORS_CONT);
 661
 662         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 663         if (test_opt(sb, WARN_ON_ERROR))
 664                 WARN_ON_ONCE(1);
 665
 666         if (!continue_fs && !sb_rdonly(sb)) {
 667                 ext4_set_mount_flag(sb, EXT4_MF_FS_ABORTED);
 668                 if (journal)
 669                         jbd2_journal_abort(journal, -EIO);
 670         }
 671
 672         if (!bdev_read_only(sb->s_bdev)) {
 673                 save_error_info(sb, error, ino, block, func, line);
 674                 /*
 675                  * In case the fs should keep running, we need to writeout
 676                  * superblock through the journal. Due to lock ordering
 677                  * constraints, it may not be safe to do it right here so we
 678                  * defer superblock flushing to a workqueue.
 679                  */
 680                 if (continue_fs && journal)
 681                         schedule_work(&EXT4_SB(sb)->s_error_work);
 682                 else
 683                         ext4_commit_super(sb);
 684         }
 685
 686         /*
 687          * We force ERRORS_RO behavior when system is rebooting. Otherwise we
 688          * could panic during 'reboot -f' as the underlying device got already
 689          * disabled.
 690          */
 691         if (test_opt(sb, ERRORS_PANIC) && !system_going_down()) {
 692                 panic("EXT4-fs (device %s): panic forced after error\n",
 693                         sb->s_id);
 694         }
 695
 696         if (sb_rdonly(sb) || continue_fs)
 697                 return;
 698
 699         ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 700         /*
 701          * Make sure updated value of ->s_mount_flags will be visible before
 702          * ->s_flags update
 703          */
 704         smp_wmb();
 705         sb->s_flags |= SB_RDONLY;
 706 }
 707
 708 static void flush_stashed_error_work(struct work_struct *work)
 709 {
 710         struct ext4_sb_info *sbi = container_of(work, struct ext4_sb_info,
 711                                                 s_error_work);
 712         journal_t *journal = sbi->s_journal;
 713         handle_t *handle;
 714
 715         /*
 716          * If the journal is still running, we have to write out superblock
 717          * through the journal to avoid collisions of other journalled sb
 718          * updates.
 719          *
 720          * We use directly jbd2 functions here to avoid recursing back into
 721          * ext4 error handling code during handling of previous errors.
 722          */
 723         if (!sb_rdonly(sbi->s_sb) && journal) {
 724                 struct buffer_head *sbh = sbi->s_sbh;
 725                 handle = jbd2_journal_start(journal, 1);
 726                 if (IS_ERR(handle))
 727                         goto write_directly;
 728                 if (jbd2_journal_get_write_access(handle, sbh)) {
 729                         jbd2_journal_stop(handle);
 730                         goto write_directly;
 731                 }
 732                 ext4_update_super(sbi->s_sb);
 733                 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
 734                         ext4_msg(sbi->s_sb, KERN_ERR, "previous I/O error to "
 735                                  "superblock detected");
 736                         clear_buffer_write_io_error(sbh);
 737                         set_buffer_uptodate(sbh);
 738                 }
 739
 740                 if (jbd2_journal_dirty_metadata(handle, sbh)) {
 741                         jbd2_journal_stop(handle);
 742                         goto write_directly;
 743                 }
 744                 jbd2_journal_stop(handle);
 745                 ext4_notify_error_sysfs(sbi);
 746                 return;
 747         }
 748 write_directly:
 749         /*
 750          * Write through journal failed. Write sb directly to get error info
 751          * out and hope for the best.
 752          */
 753         ext4_commit_super(sbi->s_sb);
 754         ext4_notify_error_sysfs(sbi);
 755 }
 756
 757 #define ext4_error_ratelimit(sb)                                        \
 758                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 759                              "EXT4-fs error")
 760
 761 void __ext4_error(struct super_block *sb, const char *function,
 762                   unsigned int line, bool force_ro, int error, __u64 block,
 763                   const char *fmt, ...)
 764 {
 765         struct va_format vaf;
 766         va_list args;
 767
 768         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 769                 return;
 770
 771         trace_ext4_error(sb, function, line);
 772         if (ext4_error_ratelimit(sb)) {
 773                 va_start(args, fmt);
 774                 vaf.fmt = fmt;
 775                 vaf.va = &args;
 776                 printk(KERN_CRIT
 777                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 778                        sb->s_id, function, line, current->comm, &vaf);
 779                 va_end(args);
 780         }
 781         fsnotify_sb_error(sb, NULL, error ? error : EFSCORRUPTED);
 782
 783         ext4_handle_error(sb, force_ro, error, 0, block, function, line);
 784 }
 785
 786 void __ext4_error_inode(struct inode *inode, const char *function,
 787                         unsigned int line, ext4_fsblk_t block, int error,
 788                         const char *fmt, ...)
 789 {
 790         va_list args;
 791         struct va_format vaf;
 792
 793         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 794                 return;
 795
 796         trace_ext4_error(inode->i_sb, function, line);
 797         if (ext4_error_ratelimit(inode->i_sb)) {
 798                 va_start(args, fmt);
 799                 vaf.fmt = fmt;
 800                 vaf.va = &args;
 801                 if (block)
 802                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 803                                "inode #%lu: block %llu: comm %s: %pV\n",
 804                                inode->i_sb->s_id, function, line, inode->i_ino,
 805                                block, current->comm, &vaf);
 806                 else
 807                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 808                                "inode #%lu: comm %s: %pV\n",
 809                                inode->i_sb->s_id, function, line, inode->i_ino,
 810                                current->comm, &vaf);
 811                 va_end(args);
 812         }
 813         fsnotify_sb_error(inode->i_sb, inode, error ? error : EFSCORRUPTED);
 814
 815         ext4_handle_error(inode->i_sb, false, error, inode->i_ino, block,
 816                           function, line);
 817 }
 818
 819 void __ext4_error_file(struct file *file, const char *function,
 820                        unsigned int line, ext4_fsblk_t block,
 821                        const char *fmt, ...)
 822 {
 823         va_list args;
 824         struct va_format vaf;
 825         struct inode *inode = file_inode(file);
 826         char pathname[80], *path;
 827
 828         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 829                 return;
 830
 831         trace_ext4_error(inode->i_sb, function, line);
 832         if (ext4_error_ratelimit(inode->i_sb)) {
 833                 path = file_path(file, pathname, sizeof(pathname));
 834                 if (IS_ERR(path))
 835                         path = "(unknown)";
 836                 va_start(args, fmt);
 837                 vaf.fmt = fmt;
 838                 vaf.va = &args;
 839                 if (block)
 840                         printk(KERN_CRIT
 841                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 842                                "block %llu: comm %s: path %s: %pV\n",
 843                                inode->i_sb->s_id, function, line, inode->i_ino,
 844                                block, current->comm, path, &vaf);
 845                 else
 846                         printk(KERN_CRIT
 847                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 848                                "comm %s: path %s: %pV\n",
 849                                inode->i_sb->s_id, function, line, inode->i_ino,
 850                                current->comm, path, &vaf);
 851                 va_end(args);
 852         }
 853         fsnotify_sb_error(inode->i_sb, inode, EFSCORRUPTED);
 854
 855         ext4_handle_error(inode->i_sb, false, EFSCORRUPTED, inode->i_ino, block,
 856                           function, line);
 857 }
 858
 859 const char *ext4_decode_error(struct super_block *sb, int errno,
 860                               char nbuf[16])
 861 {
 862         char *errstr = NULL;
 863
 864         switch (errno) {
 865         case -EFSCORRUPTED:
 866                 errstr = "Corrupt filesystem";
 867                 break;
 868         case -EFSBADCRC:
 869                 errstr = "Filesystem failed CRC";
 870                 break;
 871         case -EIO:
 872                 errstr = "IO failure";
 873                 break;
 874         case -ENOMEM:
 875                 errstr = "Out of memory";
 876                 break;
 877         case -EROFS:
 878                 if (!sb || (EXT4_SB(sb)->s_journal &&
 879                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 880                         errstr = "Journal has aborted";
 881                 else
 882                         errstr = "Readonly filesystem";
 883                 break;
 884         default:
 885                 /* If the caller passed in an extra buffer for unknown
 886                  * errors, textualise them now.  Else we just return
 887                  * NULL. */
 888                 if (nbuf) {
 889                         /* Check for truncated error codes... */
 890                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 891                                 errstr = nbuf;
 892                 }
 893                 break;
 894         }
 895
 896         return errstr;
 897 }
 898
 899 /* __ext4_std_error decodes expected errors from journaling functions
 900  * automatically and invokes the appropriate error response.  */
 901
 902 void __ext4_std_error(struct super_block *sb, const char *function,
 903                       unsigned int line, int errno)
 904 {
 905         char nbuf[16];
 906         const char *errstr;
 907
 908         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 909                 return;
 910
 911         /* Special case: if the error is EROFS, and we're not already
 912          * inside a transaction, then there's really no point in logging
 913          * an error. */
 914         if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb))
 915                 return;
 916
 917         if (ext4_error_ratelimit(sb)) {
 918                 errstr = ext4_decode_error(sb, errno, nbuf);
 919                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 920                        sb->s_id, function, line, errstr);
 921         }
 922         fsnotify_sb_error(sb, NULL, errno ? errno : EFSCORRUPTED);
 923
 924         ext4_handle_error(sb, false, -errno, 0, 0, function, line);
 925 }
 926
 927 void __ext4_msg(struct super_block *sb,
 928                 const char *prefix, const char *fmt, ...)
 929 {
 930         struct va_format vaf;
 931         va_list args;
 932
 933         if (sb) {
 934                 atomic_inc(&EXT4_SB(sb)->s_msg_count);
 935                 if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state),
 936                                   "EXT4-fs"))
 937                         return;
 938         }
 939
 940         va_start(args, fmt);
 941         vaf.fmt = fmt;
 942         vaf.va = &args;
 943         if (sb)
 944                 printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 945         else
 946                 printk("%sEXT4-fs: %pV\n", prefix, &vaf);
 947         va_end(args);
 948 }
 949
 950 static int ext4_warning_ratelimit(struct super_block *sb)
 951 {
 952         atomic_inc(&EXT4_SB(sb)->s_warning_count);
 953         return ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state),
 954                             "EXT4-fs warning");
 955 }
 956
 957 void __ext4_warning(struct super_block *sb, const char *function,
 958                     unsigned int line, const char *fmt, ...)
 959 {
 960         struct va_format vaf;
 961         va_list args;
 962
 963         if (!ext4_warning_ratelimit(sb))
 964                 return;
 965
 966         va_start(args, fmt);
 967         vaf.fmt = fmt;
 968         vaf.va = &args;
 969         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 970                sb->s_id, function, line, &vaf);
 971         va_end(args);
 972 }
 973
 974 void __ext4_warning_inode(const struct inode *inode, const char *function,
 975                           unsigned int line, const char *fmt, ...)
 976 {
 977         struct va_format vaf;
 978         va_list args;
 979
 980         if (!ext4_warning_ratelimit(inode->i_sb))
 981                 return;
 982
 983         va_start(args, fmt);
 984         vaf.fmt = fmt;
 985         vaf.va = &args;
 986         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 987                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 988                function, line, inode->i_ino, current->comm, &vaf);
 989         va_end(args);
 990 }
 991
 992 void __ext4_grp_locked_error(const char *function, unsigned int line,
 993                              struct super_block *sb, ext4_group_t grp,
 994                              unsigned long ino, ext4_fsblk_t block,
 995                              const char *fmt, ...)
 996 __releases(bitlock)
 997 __acquires(bitlock)
 998 {
 999         struct va_format vaf;
1000         va_list args;
1001
1002         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
1003                 return;
1004
1005         trace_ext4_error(sb, function, line);
1006         if (ext4_error_ratelimit(sb)) {
1007                 va_start(args, fmt);
1008                 vaf.fmt = fmt;
1009                 vaf.va = &args;
1010                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
1011                        sb->s_id, function, line, grp);
1012                 if (ino)
1013                         printk(KERN_CONT "inode %lu: ", ino);
1014                 if (block)
1015                         printk(KERN_CONT "block %llu:",
1016                                (unsigned long long) block);
1017                 printk(KERN_CONT "%pV\n", &vaf);
1018                 va_end(args);
1019         }
1020
1021         if (test_opt(sb, ERRORS_CONT)) {
1022                 if (test_opt(sb, WARN_ON_ERROR))
1023                         WARN_ON_ONCE(1);
1024                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
1025                 if (!bdev_read_only(sb->s_bdev)) {
1026                         save_error_info(sb, EFSCORRUPTED, ino, block, function,
1027                                         line);
1028                         schedule_work(&EXT4_SB(sb)->s_error_work);
1029                 }
1030                 return;
1031         }
1032         ext4_unlock_group(sb, grp);
1033         ext4_handle_error(sb, false, EFSCORRUPTED, ino, block, function, line);
1034         /*
1035          * We only get here in the ERRORS_RO case; relocking the group
1036          * may be dangerous, but nothing bad will happen since the
1037          * filesystem will have already been marked read/only and the
1038          * journal has been aborted.  We return 1 as a hint to callers
1039          * who might what to use the return value from
1040          * ext4_grp_locked_error() to distinguish between the
1041          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
1042          * aggressively from the ext4 function in question, with a
1043          * more appropriate error code.
1044          */
1045         ext4_lock_group(sb, grp);
1046         return;
1047 }
1048
1049 void ext4_mark_group_bitmap_corrupted(struct super_block *sb,
1050                                      ext4_group_t group,
1051                                      unsigned int flags)
1052 {
1053         struct ext4_sb_info *sbi = EXT4_SB(sb);
1054         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1055         struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
1056         int ret;
1057
1058         if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) {
1059                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT,
1060                                             &grp->bb_state);
1061                 if (!ret)
1062                         percpu_counter_sub(&sbi->s_freeclusters_counter,
1063                                            grp->bb_free);
1064         }
1065
1066         if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) {
1067                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT,
1068                                             &grp->bb_state);
1069                 if (!ret && gdp) {
1070                         int count;
1071
1072                         count = ext4_free_inodes_count(sb, gdp);
1073                         percpu_counter_sub(&sbi->s_freeinodes_counter,
1074                                            count);
1075                 }
1076         }
1077 }
1078
1079 void ext4_update_dynamic_rev(struct super_block *sb)
1080 {
1081         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1082
1083         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
1084                 return;
1085
1086         ext4_warning(sb,
1087                      "updating to rev %d because of new feature flag, "
1088                      "running e2fsck is recommended",
1089                      EXT4_DYNAMIC_REV);
1090
1091         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
1092         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
1093         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
1094         /* leave es->s_feature_*compat flags alone */
1095         /* es->s_uuid will be set by e2fsck if empty */
1096
1097         /*
1098          * The rest of the superblock fields should be zero, and if not it
1099          * means they are likely already in use, so leave them alone.  We
1100          * can leave it up to e2fsck to clean up any inconsistencies there.
1101          */
1102 }
1103
1104 /*
1105  * Open the external journal device
1106  */
1107 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
1108 {
1109         struct block_device *bdev;
1110
1111         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
1112         if (IS_ERR(bdev))
1113                 goto fail;
1114         return bdev;
1115
1116 fail:
1117         ext4_msg(sb, KERN_ERR,
1118                  "failed to open journal device unknown-block(%u,%u) %ld",
1119                  MAJOR(dev), MINOR(dev), PTR_ERR(bdev));
1120         return NULL;
1121 }
1122
1123 /*
1124  * Release the journal device
1125  */
1126 static void ext4_blkdev_put(struct block_device *bdev)
1127 {
1128         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1129 }
1130
1131 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
1132 {
1133         struct block_device *bdev;
1134         bdev = sbi->s_journal_bdev;
1135         if (bdev) {
1136                 ext4_blkdev_put(bdev);
1137                 sbi->s_journal_bdev = NULL;
1138         }
1139 }
1140
1141 static inline struct inode *orphan_list_entry(struct list_head *l)
1142 {
1143         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
1144 }
1145
1146 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
1147 {
1148         struct list_head *l;
1149
1150         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
1151                  le32_to_cpu(sbi->s_es->s_last_orphan));
1152
1153         printk(KERN_ERR "sb_info orphan list:\n");
1154         list_for_each(l, &sbi->s_orphan) {
1155                 struct inode *inode = orphan_list_entry(l);
1156                 printk(KERN_ERR "  "
1157                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
1158                        inode->i_sb->s_id, inode->i_ino, inode,
1159                        inode->i_mode, inode->i_nlink,
1160                        NEXT_ORPHAN(inode));
1161         }
1162 }
1163
1164 #ifdef CONFIG_QUOTA
1165 static int ext4_quota_off(struct super_block *sb, int type);
1166
1167 static inline void ext4_quota_off_umount(struct super_block *sb)
1168 {
1169         int type;
1170
1171         /* Use our quota_off function to clear inode flags etc. */
1172         for (type = 0; type < EXT4_MAXQUOTAS; type++)
1173                 ext4_quota_off(sb, type);
1174 }
1175
1176 /*
1177  * This is a helper function which is used in the mount/remount
1178  * codepaths (which holds s_umount) to fetch the quota file name.
1179  */
1180 static inline char *get_qf_name(struct super_block *sb,
1181                                 struct ext4_sb_info *sbi,
1182                                 int type)
1183 {
1184         return rcu_dereference_protected(sbi->s_qf_names[type],
1185                                          lockdep_is_held(&sb->s_umount));
1186 }
1187 #else
1188 static inline void ext4_quota_off_umount(struct super_block *sb)
1189 {
1190 }
1191 #endif
1192
1193 static void ext4_put_super(struct super_block *sb)
1194 {
1195         struct ext4_sb_info *sbi = EXT4_SB(sb);
1196         struct ext4_super_block *es = sbi->s_es;
1197         struct buffer_head **group_desc;
1198         struct flex_groups **flex_groups;
1199         int aborted = 0;
1200         int i, err;
1201
1202         /*
1203          * Unregister sysfs before destroying jbd2 journal.
1204          * Since we could still access attr_journal_task attribute via sysfs
1205          * path which could have sbi->s_journal->j_task as NULL
1206          * Unregister sysfs before flush sbi->s_error_work.
1207          * Since user may read /proc/fs/ext4/xx/mb_groups during umount, If
1208          * read metadata verify failed then will queue error work.
1209          * flush_stashed_error_work will call start_this_handle may trigger
1210          * BUG_ON.
1211          */
1212         ext4_unregister_sysfs(sb);
1213
1214         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs unmount"))
1215                 ext4_msg(sb, KERN_INFO, "unmounting filesystem.");
1216
1217         ext4_unregister_li_request(sb);
1218         ext4_quota_off_umount(sb);
1219
1220         flush_work(&sbi->s_error_work);
1221         destroy_workqueue(sbi->rsv_conversion_wq);
1222         ext4_release_orphan_info(sb);
1223
1224         if (sbi->s_journal) {
1225                 aborted = is_journal_aborted(sbi->s_journal);
1226                 err = jbd2_journal_destroy(sbi->s_journal);
1227                 sbi->s_journal = NULL;
1228                 if ((err < 0) && !aborted) {
1229                         ext4_abort(sb, -err, "Couldn't clean up the journal");
1230                 }
1231         }
1232
1233         ext4_es_unregister_shrinker(sbi);
1234         del_timer_sync(&sbi->s_err_report);
1235         ext4_release_system_zone(sb);
1236         ext4_mb_release(sb);
1237         ext4_ext_release(sb);
1238
1239         if (!sb_rdonly(sb) && !aborted) {
1240                 ext4_clear_feature_journal_needs_recovery(sb);
1241                 ext4_clear_feature_orphan_present(sb);
1242                 es->s_state = cpu_to_le16(sbi->s_mount_state);
1243         }
1244         if (!sb_rdonly(sb))
1245                 ext4_commit_super(sb);
1246
1247         rcu_read_lock();
1248         group_desc = rcu_dereference(sbi->s_group_desc);
1249         for (i = 0; i < sbi->s_gdb_count; i++)
1250                 brelse(group_desc[i]);
1251         kvfree(group_desc);
1252         flex_groups = rcu_dereference(sbi->s_flex_groups);
1253         if (flex_groups) {
1254                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
1255                         kvfree(flex_groups[i]);
1256                 kvfree(flex_groups);
1257         }
1258         rcu_read_unlock();
1259         percpu_counter_destroy(&sbi->s_freeclusters_counter);
1260         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1261         percpu_counter_destroy(&sbi->s_dirs_counter);
1262         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
1263         percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit);
1264         percpu_free_rwsem(&sbi->s_writepages_rwsem);
1265 #ifdef CONFIG_QUOTA
1266         for (i = 0; i < EXT4_MAXQUOTAS; i++)
1267                 kfree(get_qf_name(sb, sbi, i));
1268 #endif
1269
1270         /* Debugging code just in case the in-memory inode orphan list
1271          * isn't empty.  The on-disk one can be non-empty if we've
1272          * detected an error and taken the fs readonly, but the
1273          * in-memory list had better be clean by this point. */
1274         if (!list_empty(&sbi->s_orphan))
1275                 dump_orphan_list(sb, sbi);
1276         ASSERT(list_empty(&sbi->s_orphan));
1277
1278         sync_blockdev(sb->s_bdev);
1279         invalidate_bdev(sb->s_bdev);
1280         if (sbi->s_journal_bdev && sbi->s_journal_bdev != sb->s_bdev) {
1281                 /*
1282                  * Invalidate the journal device's buffers.  We don't want them
1283                  * floating about in memory - the physical journal device may
1284                  * hotswapped, and it breaks the `ro-after' testing code.
1285                  */
1286                 sync_blockdev(sbi->s_journal_bdev);
1287                 invalidate_bdev(sbi->s_journal_bdev);
1288                 ext4_blkdev_remove(sbi);
1289         }
1290
1291         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
1292         sbi->s_ea_inode_cache = NULL;
1293
1294         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
1295         sbi->s_ea_block_cache = NULL;
1296
1297         ext4_stop_mmpd(sbi);
1298
1299         brelse(sbi->s_sbh);
1300         sb->s_fs_info = NULL;
1301         /*
1302          * Now that we are completely done shutting down the
1303          * superblock, we need to actually destroy the kobject.
1304          */
1305         kobject_put(&sbi->s_kobj);
1306         wait_for_completion(&sbi->s_kobj_unregister);
1307         if (sbi->s_chksum_driver)
1308                 crypto_free_shash(sbi->s_chksum_driver);
1309         kfree(sbi->s_blockgroup_lock);
1310         fs_put_dax(sbi->s_daxdev);
1311         fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy);
1312 #if IS_ENABLED(CONFIG_UNICODE)
1313         utf8_unload(sb->s_encoding);
1314 #endif
1315         kfree(sbi);
1316 }
1317
1318 static struct kmem_cache *ext4_inode_cachep;
1319
1320 /*
1321  * Called inside transaction, so use GFP_NOFS
1322  */
1323 static struct inode *ext4_alloc_inode(struct super_block *sb)
1324 {
1325         struct ext4_inode_info *ei;
1326
1327         ei = alloc_inode_sb(sb, ext4_inode_cachep, GFP_NOFS);
1328         if (!ei)
1329                 return NULL;
1330
1331         inode_set_iversion(&ei->vfs_inode, 1);
1332         spin_lock_init(&ei->i_raw_lock);
1333         INIT_LIST_HEAD(&ei->i_prealloc_list);
1334         atomic_set(&ei->i_prealloc_active, 0);
1335         spin_lock_init(&ei->i_prealloc_lock);
1336         ext4_es_init_tree(&ei->i_es_tree);
1337         rwlock_init(&ei->i_es_lock);
1338         INIT_LIST_HEAD(&ei->i_es_list);
1339         ei->i_es_all_nr = 0;
1340         ei->i_es_shk_nr = 0;
1341         ei->i_es_shrink_lblk = 0;
1342         ei->i_reserved_data_blocks = 0;
1343         spin_lock_init(&(ei->i_block_reservation_lock));
1344         ext4_init_pending_tree(&ei->i_pending_tree);
1345 #ifdef CONFIG_QUOTA
1346         ei->i_reserved_quota = 0;
1347         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
1348 #endif
1349         ei->jinode = NULL;
1350         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
1351         spin_lock_init(&ei->i_completed_io_lock);
1352         ei->i_sync_tid = 0;
1353         ei->i_datasync_tid = 0;
1354         atomic_set(&ei->i_unwritten, 0);
1355         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
1356         ext4_fc_init_inode(&ei->vfs_inode);
1357         mutex_init(&ei->i_fc_lock);
1358         return &ei->vfs_inode;
1359 }
1360
1361 static int ext4_drop_inode(struct inode *inode)
1362 {
1363         int drop = generic_drop_inode(inode);
1364
1365         if (!drop)
1366                 drop = fscrypt_drop_inode(inode);
1367
1368         trace_ext4_drop_inode(inode, drop);
1369         return drop;
1370 }
1371
1372 static void ext4_free_in_core_inode(struct inode *inode)
1373 {
1374         fscrypt_free_inode(inode);
1375         if (!list_empty(&(EXT4_I(inode)->i_fc_list))) {
1376                 pr_warn("%s: inode %ld still in fc list",
1377                         __func__, inode->i_ino);
1378         }
1379         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1380 }
1381
1382 static void ext4_destroy_inode(struct inode *inode)
1383 {
1384         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1385                 ext4_msg(inode->i_sb, KERN_ERR,
1386                          "Inode %lu (%p): orphan list check failed!",
1387                          inode->i_ino, EXT4_I(inode));
1388                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1389                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1390                                 true);
1391                 dump_stack();
1392         }
1393
1394         if (EXT4_I(inode)->i_reserved_data_blocks)
1395                 ext4_msg(inode->i_sb, KERN_ERR,
1396                          "Inode %lu (%p): i_reserved_data_blocks (%u) not cleared!",
1397                          inode->i_ino, EXT4_I(inode),
1398                          EXT4_I(inode)->i_reserved_data_blocks);
1399 }
1400
1401 static void init_once(void *foo)
1402 {
1403         struct ext4_inode_info *ei = foo;
1404
1405         INIT_LIST_HEAD(&ei->i_orphan);
1406         init_rwsem(&ei->xattr_sem);
1407         init_rwsem(&ei->i_data_sem);
1408         inode_init_once(&ei->vfs_inode);
1409         ext4_fc_init_inode(&ei->vfs_inode);
1410 }
1411
1412 static int __init init_inodecache(void)
1413 {
1414         ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache",
1415                                 sizeof(struct ext4_inode_info), 0,
1416                                 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
1417                                         SLAB_ACCOUNT),
1418                                 offsetof(struct ext4_inode_info, i_data),
1419                                 sizeof_field(struct ext4_inode_info, i_data),
1420                                 init_once);
1421         if (ext4_inode_cachep == NULL)
1422                 return -ENOMEM;
1423         return 0;
1424 }
1425
1426 static void destroy_inodecache(void)
1427 {
1428         /*
1429          * Make sure all delayed rcu free inodes are flushed before we
1430          * destroy cache.
1431          */
1432         rcu_barrier();
1433         kmem_cache_destroy(ext4_inode_cachep);
1434 }
1435
1436 void ext4_clear_inode(struct inode *inode)
1437 {
1438         ext4_fc_del(inode);
1439         invalidate_inode_buffers(inode);
1440         clear_inode(inode);
1441         ext4_discard_preallocations(inode, 0);
1442         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1443         dquot_drop(inode);
1444         if (EXT4_I(inode)->jinode) {
1445                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1446                                                EXT4_I(inode)->jinode);
1447                 jbd2_free_inode(EXT4_I(inode)->jinode);
1448                 EXT4_I(inode)->jinode = NULL;
1449         }
1450         fscrypt_put_encryption_info(inode);
1451         fsverity_cleanup_inode(inode);
1452 }
1453
1454 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1455                                         u64 ino, u32 generation)
1456 {
1457         struct inode *inode;
1458
1459         /*
1460          * Currently we don't know the generation for parent directory, so
1461          * a generation of 0 means "accept any"
1462          */
1463         inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE);
1464         if (IS_ERR(inode))
1465                 return ERR_CAST(inode);
1466         if (generation && inode->i_generation != generation) {
1467                 iput(inode);
1468                 return ERR_PTR(-ESTALE);
1469         }
1470
1471         return inode;
1472 }
1473
1474 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1475                                         int fh_len, int fh_type)
1476 {
1477         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1478                                     ext4_nfs_get_inode);
1479 }
1480
1481 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1482                                         int fh_len, int fh_type)
1483 {
1484         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1485                                     ext4_nfs_get_inode);
1486 }
1487
1488 static int ext4_nfs_commit_metadata(struct inode *inode)
1489 {
1490         struct writeback_control wbc = {
1491                 .sync_mode = WB_SYNC_ALL
1492         };
1493
1494         trace_ext4_nfs_commit_metadata(inode);
1495         return ext4_write_inode(inode, &wbc);
1496 }
1497
1498 #ifdef CONFIG_QUOTA
1499 static const char * const quotatypes[] = INITQFNAMES;
1500 #define QTYPE2NAME(t) (quotatypes[t])
1501
1502 static int ext4_write_dquot(struct dquot *dquot);
1503 static int ext4_acquire_dquot(struct dquot *dquot);
1504 static int ext4_release_dquot(struct dquot *dquot);
1505 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1506 static int ext4_write_info(struct super_block *sb, int type);
1507 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1508                          const struct path *path);
1509 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1510                                size_t len, loff_t off);
1511 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1512                                 const char *data, size_t len, loff_t off);
1513 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1514                              unsigned int flags);
1515
1516 static struct dquot **ext4_get_dquots(struct inode *inode)
1517 {
1518         return EXT4_I(inode)->i_dquot;
1519 }
1520
1521 static const struct dquot_operations ext4_quota_operations = {
1522         .get_reserved_space     = ext4_get_reserved_space,
1523         .write_dquot            = ext4_write_dquot,
1524         .acquire_dquot          = ext4_acquire_dquot,
1525         .release_dquot          = ext4_release_dquot,
1526         .mark_dirty             = ext4_mark_dquot_dirty,
1527         .write_info             = ext4_write_info,
1528         .alloc_dquot            = dquot_alloc,
1529         .destroy_dquot          = dquot_destroy,
1530         .get_projid             = ext4_get_projid,
1531         .get_inode_usage        = ext4_get_inode_usage,
1532         .get_next_id            = dquot_get_next_id,
1533 };
1534
1535 static const struct quotactl_ops ext4_qctl_operations = {
1536         .quota_on       = ext4_quota_on,
1537         .quota_off      = ext4_quota_off,
1538         .quota_sync     = dquot_quota_sync,
1539         .get_state      = dquot_get_state,
1540         .set_info       = dquot_set_dqinfo,
1541         .get_dqblk      = dquot_get_dqblk,
1542         .set_dqblk      = dquot_set_dqblk,
1543         .get_nextdqblk  = dquot_get_next_dqblk,
1544 };
1545 #endif
1546
1547 static const struct super_operations ext4_sops = {
1548         .alloc_inode    = ext4_alloc_inode,
1549         .free_inode     = ext4_free_in_core_inode,
1550         .destroy_inode  = ext4_destroy_inode,
1551         .write_inode    = ext4_write_inode,
1552         .dirty_inode    = ext4_dirty_inode,
1553         .drop_inode     = ext4_drop_inode,
1554         .evict_inode    = ext4_evict_inode,
1555         .put_super      = ext4_put_super,
1556         .sync_fs        = ext4_sync_fs,
1557         .freeze_fs      = ext4_freeze,
1558         .unfreeze_fs    = ext4_unfreeze,
1559         .statfs         = ext4_statfs,
1560         .show_options   = ext4_show_options,
1561 #ifdef CONFIG_QUOTA
1562         .quota_read     = ext4_quota_read,
1563         .quota_write    = ext4_quota_write,
1564         .get_dquots     = ext4_get_dquots,
1565 #endif
1566 };
1567
1568 static const struct export_operations ext4_export_ops = {
1569         .fh_to_dentry = ext4_fh_to_dentry,
1570         .fh_to_parent = ext4_fh_to_parent,
1571         .get_parent = ext4_get_parent,
1572         .commit_metadata = ext4_nfs_commit_metadata,
1573 };
1574
1575 enum {
1576         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1577         Opt_resgid, Opt_resuid, Opt_sb,
1578         Opt_nouid32, Opt_debug, Opt_removed,
1579         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1580         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1581         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1582         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1583         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1584         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1585         Opt_inlinecrypt,
1586         Opt_usrjquota, Opt_grpjquota, Opt_quota,
1587         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1588         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version,
1589         Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never,
1590         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
1591         Opt_nowarn_on_error, Opt_mblk_io_submit, Opt_debug_want_extra_isize,
1592         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1593         Opt_inode_readahead_blks, Opt_journal_ioprio,
1594         Opt_dioread_nolock, Opt_dioread_lock,
1595         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1596         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1597         Opt_no_prefetch_block_bitmaps, Opt_mb_optimize_scan,
1598         Opt_errors, Opt_data, Opt_data_err, Opt_jqfmt, Opt_dax_type,
1599 #ifdef CONFIG_EXT4_DEBUG
1600         Opt_fc_debug_max_replay, Opt_fc_debug_force
1601 #endif
1602 };
1603
1604 static const struct constant_table ext4_param_errors[] = {
1605         {"continue",    EXT4_MOUNT_ERRORS_CONT},
1606         {"panic",       EXT4_MOUNT_ERRORS_PANIC},
1607         {"remount-ro",  EXT4_MOUNT_ERRORS_RO},
1608         {}
1609 };
1610
1611 static const struct constant_table ext4_param_data[] = {
1612         {"journal",     EXT4_MOUNT_JOURNAL_DATA},
1613         {"ordered",     EXT4_MOUNT_ORDERED_DATA},
1614         {"writeback",   EXT4_MOUNT_WRITEBACK_DATA},
1615         {}
1616 };
1617
1618 static const struct constant_table ext4_param_data_err[] = {
1619         {"abort",       Opt_data_err_abort},
1620         {"ignore",      Opt_data_err_ignore},
1621         {}
1622 };
1623
1624 static const struct constant_table ext4_param_jqfmt[] = {
1625         {"vfsold",      QFMT_VFS_OLD},
1626         {"vfsv0",       QFMT_VFS_V0},
1627         {"vfsv1",       QFMT_VFS_V1},
1628         {}
1629 };
1630
1631 static const struct constant_table ext4_param_dax[] = {
1632         {"always",      Opt_dax_always},
1633         {"inode",       Opt_dax_inode},
1634         {"never",       Opt_dax_never},
1635         {}
1636 };
1637
1638 /* String parameter that allows empty argument */
1639 #define fsparam_string_empty(NAME, OPT) \
1640         __fsparam(fs_param_is_string, NAME, OPT, fs_param_can_be_empty, NULL)
1641
1642 /*
1643  * Mount option specification
1644  * We don't use fsparam_flag_no because of the way we set the
1645  * options and the way we show them in _ext4_show_options(). To
1646  * keep the changes to a minimum, let's keep the negative options
1647  * separate for now.
1648  */
1649 static const struct fs_parameter_spec ext4_param_specs[] = {
1650         fsparam_flag    ("bsddf",               Opt_bsd_df),
1651         fsparam_flag    ("minixdf",             Opt_minix_df),
1652         fsparam_flag    ("grpid",               Opt_grpid),
1653         fsparam_flag    ("bsdgroups",           Opt_grpid),
1654         fsparam_flag    ("nogrpid",             Opt_nogrpid),
1655         fsparam_flag    ("sysvgroups",          Opt_nogrpid),
1656         fsparam_u32     ("resgid",              Opt_resgid),
1657         fsparam_u32     ("resuid",              Opt_resuid),
1658         fsparam_u32     ("sb",                  Opt_sb),
1659         fsparam_enum    ("errors",              Opt_errors, ext4_param_errors),
1660         fsparam_flag    ("nouid32",             Opt_nouid32),
1661         fsparam_flag    ("debug",               Opt_debug),
1662         fsparam_flag    ("oldalloc",            Opt_removed),
1663         fsparam_flag    ("orlov",               Opt_removed),
1664         fsparam_flag    ("user_xattr",          Opt_user_xattr),
1665         fsparam_flag    ("nouser_xattr",        Opt_nouser_xattr),
1666         fsparam_flag    ("acl",                 Opt_acl),
1667         fsparam_flag    ("noacl",               Opt_noacl),
1668         fsparam_flag    ("norecovery",          Opt_noload),
1669         fsparam_flag    ("noload",              Opt_noload),
1670         fsparam_flag    ("bh",                  Opt_removed),
1671         fsparam_flag    ("nobh",                Opt_removed),
1672         fsparam_u32     ("commit",              Opt_commit),
1673         fsparam_u32     ("min_batch_time",      Opt_min_batch_time),
1674         fsparam_u32     ("max_batch_time",      Opt_max_batch_time),
1675         fsparam_u32     ("journal_dev",         Opt_journal_dev),
1676         fsparam_bdev    ("journal_path",        Opt_journal_path),
1677         fsparam_flag    ("journal_checksum",    Opt_journal_checksum),
1678         fsparam_flag    ("nojournal_checksum",  Opt_nojournal_checksum),
1679         fsparam_flag    ("journal_async_commit",Opt_journal_async_commit),
1680         fsparam_flag    ("abort",               Opt_abort),
1681         fsparam_enum    ("data",                Opt_data, ext4_param_data),
1682         fsparam_enum    ("data_err",            Opt_data_err,
1683                                                 ext4_param_data_err),
1684         fsparam_string_empty
1685                         ("usrjquota",           Opt_usrjquota),
1686         fsparam_string_empty
1687                         ("grpjquota",           Opt_grpjquota),
1688         fsparam_enum    ("jqfmt",               Opt_jqfmt, ext4_param_jqfmt),
1689         fsparam_flag    ("grpquota",            Opt_grpquota),
1690         fsparam_flag    ("quota",               Opt_quota),
1691         fsparam_flag    ("noquota",             Opt_noquota),
1692         fsparam_flag    ("usrquota",            Opt_usrquota),
1693         fsparam_flag    ("prjquota",            Opt_prjquota),
1694         fsparam_flag    ("barrier",             Opt_barrier),
1695         fsparam_u32     ("barrier",             Opt_barrier),
1696         fsparam_flag    ("nobarrier",           Opt_nobarrier),
1697         fsparam_flag    ("i_version",           Opt_i_version),
1698         fsparam_flag    ("dax",                 Opt_dax),
1699         fsparam_enum    ("dax",                 Opt_dax_type, ext4_param_dax),
1700         fsparam_u32     ("stripe",              Opt_stripe),
1701         fsparam_flag    ("delalloc",            Opt_delalloc),
1702         fsparam_flag    ("nodelalloc",          Opt_nodelalloc),
1703         fsparam_flag    ("warn_on_error",       Opt_warn_on_error),
1704         fsparam_flag    ("nowarn_on_error",     Opt_nowarn_on_error),
1705         fsparam_u32     ("debug_want_extra_isize",
1706                                                 Opt_debug_want_extra_isize),
1707         fsparam_flag    ("mblk_io_submit",      Opt_removed),
1708         fsparam_flag    ("nomblk_io_submit",    Opt_removed),
1709         fsparam_flag    ("block_validity",      Opt_block_validity),
1710         fsparam_flag    ("noblock_validity",    Opt_noblock_validity),
1711         fsparam_u32     ("inode_readahead_blks",
1712                                                 Opt_inode_readahead_blks),
1713         fsparam_u32     ("journal_ioprio",      Opt_journal_ioprio),
1714         fsparam_u32     ("auto_da_alloc",       Opt_auto_da_alloc),
1715         fsparam_flag    ("auto_da_alloc",       Opt_auto_da_alloc),
1716         fsparam_flag    ("noauto_da_alloc",     Opt_noauto_da_alloc),
1717         fsparam_flag    ("dioread_nolock",      Opt_dioread_nolock),
1718         fsparam_flag    ("nodioread_nolock",    Opt_dioread_lock),
1719         fsparam_flag    ("dioread_lock",        Opt_dioread_lock),
1720         fsparam_flag    ("discard",             Opt_discard),
1721         fsparam_flag    ("nodiscard",           Opt_nodiscard),
1722         fsparam_u32     ("init_itable",         Opt_init_itable),
1723         fsparam_flag    ("init_itable",         Opt_init_itable),
1724         fsparam_flag    ("noinit_itable",       Opt_noinit_itable),
1725 #ifdef CONFIG_EXT4_DEBUG
1726         fsparam_flag    ("fc_debug_force",      Opt_fc_debug_force),
1727         fsparam_u32     ("fc_debug_max_replay", Opt_fc_debug_max_replay),
1728 #endif
1729         fsparam_u32     ("max_dir_size_kb",     Opt_max_dir_size_kb),
1730         fsparam_flag    ("test_dummy_encryption",
1731                                                 Opt_test_dummy_encryption),
1732         fsparam_string  ("test_dummy_encryption",
1733                                                 Opt_test_dummy_encryption),
1734         fsparam_flag    ("inlinecrypt",         Opt_inlinecrypt),
1735         fsparam_flag    ("nombcache",           Opt_nombcache),
1736         fsparam_flag    ("no_mbcache",          Opt_nombcache), /* for backward compatibility */
1737         fsparam_flag    ("prefetch_block_bitmaps",
1738                                                 Opt_removed),
1739         fsparam_flag    ("no_prefetch_block_bitmaps",
1740                                                 Opt_no_prefetch_block_bitmaps),
1741         fsparam_s32     ("mb_optimize_scan",    Opt_mb_optimize_scan),
1742         fsparam_string  ("check",               Opt_removed),   /* mount option from ext2/3 */
1743         fsparam_flag    ("nocheck",             Opt_removed),   /* mount option from ext2/3 */
1744         fsparam_flag    ("reservation",         Opt_removed),   /* mount option from ext2/3 */
1745         fsparam_flag    ("noreservation",       Opt_removed),   /* mount option from ext2/3 */
1746         fsparam_u32     ("journal",             Opt_removed),   /* mount option from ext2/3 */
1747         {}
1748 };
1749
1750 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1751
1752 static const char deprecated_msg[] =
1753         "Mount option \"%s\" will be removed by %s\n"
1754         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1755
1756 #define MOPT_SET        0x0001
1757 #define MOPT_CLEAR      0x0002
1758 #define MOPT_NOSUPPORT  0x0004
1759 #define MOPT_EXPLICIT   0x0008
1760 #ifdef CONFIG_QUOTA
1761 #define MOPT_Q          0
1762 #define MOPT_QFMT       0x0010
1763 #else
1764 #define MOPT_Q          MOPT_NOSUPPORT
1765 #define MOPT_QFMT       MOPT_NOSUPPORT
1766 #endif
1767 #define MOPT_NO_EXT2    0x0020
1768 #define MOPT_NO_EXT3    0x0040
1769 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1770 #define MOPT_SKIP       0x0080
1771 #define MOPT_2          0x0100
1772
1773 static const struct mount_opts {
1774         int     token;
1775         int     mount_opt;
1776         int     flags;
1777 } ext4_mount_opts[] = {
1778         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1779         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1780         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1781         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1782         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1783         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1784         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1785          MOPT_EXT4_ONLY | MOPT_SET},
1786         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1787          MOPT_EXT4_ONLY | MOPT_CLEAR},
1788         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1789         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1790         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1791          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1792         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1793          MOPT_EXT4_ONLY | MOPT_CLEAR},
1794         {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET},
1795         {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR},
1796         {Opt_commit, 0, MOPT_NO_EXT2},
1797         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1798          MOPT_EXT4_ONLY | MOPT_CLEAR},
1799         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1800          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1801         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1802                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1803          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1804         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1805         {Opt_data_err, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_NO_EXT2},
1806         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1807         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1808         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1809         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1810         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1811         {Opt_dax_type, 0, MOPT_EXT4_ONLY},
1812         {Opt_journal_dev, 0, MOPT_NO_EXT2},
1813         {Opt_journal_path, 0, MOPT_NO_EXT2},
1814         {Opt_journal_ioprio, 0, MOPT_NO_EXT2},
1815         {Opt_data, 0, MOPT_NO_EXT2},
1816         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1817         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1818 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1819         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1820         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1821 #else
1822         {Opt_acl, 0, MOPT_NOSUPPORT},
1823         {Opt_noacl, 0, MOPT_NOSUPPORT},
1824 #endif
1825         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1826         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1827         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1828         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1829                                                         MOPT_SET | MOPT_Q},
1830         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1831                                                         MOPT_SET | MOPT_Q},
1832         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1833                                                         MOPT_SET | MOPT_Q},
1834         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1835                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1836                                                         MOPT_CLEAR | MOPT_Q},
1837         {Opt_usrjquota, 0, MOPT_Q},
1838         {Opt_grpjquota, 0, MOPT_Q},
1839         {Opt_jqfmt, 0, MOPT_QFMT},
1840         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1841         {Opt_no_prefetch_block_bitmaps, EXT4_MOUNT_NO_PREFETCH_BLOCK_BITMAPS,
1842          MOPT_SET},
1843 #ifdef CONFIG_EXT4_DEBUG
1844         {Opt_fc_debug_force, EXT4_MOUNT2_JOURNAL_FAST_COMMIT,
1845          MOPT_SET | MOPT_2 | MOPT_EXT4_ONLY},
1846 #endif
1847         {Opt_err, 0, 0}
1848 };
1849
1850 #if IS_ENABLED(CONFIG_UNICODE)
1851 static const struct ext4_sb_encodings {
1852         __u16 magic;
1853         char *name;
1854         unsigned int version;
1855 } ext4_sb_encoding_map[] = {
1856         {EXT4_ENC_UTF8_12_1, "utf8", UNICODE_AGE(12, 1, 0)},
1857 };
1858
1859 static const struct ext4_sb_encodings *
1860 ext4_sb_read_encoding(const struct ext4_super_block *es)
1861 {
1862         __u16 magic = le16_to_cpu(es->s_encoding);
1863         int i;
1864
1865         for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++)
1866                 if (magic == ext4_sb_encoding_map[i].magic)
1867                         return &ext4_sb_encoding_map[i];
1868
1869         return NULL;
1870 }
1871 #endif
1872
1873 #define EXT4_SPEC_JQUOTA                        (1 <<  0)
1874 #define EXT4_SPEC_JQFMT                         (1 <<  1)
1875 #define EXT4_SPEC_DATAJ                         (1 <<  2)
1876 #define EXT4_SPEC_SB_BLOCK                      (1 <<  3)
1877 #define EXT4_SPEC_JOURNAL_DEV                   (1 <<  4)
1878 #define EXT4_SPEC_JOURNAL_IOPRIO                (1 <<  5)
1879 #define EXT4_SPEC_s_want_extra_isize            (1 <<  7)
1880 #define EXT4_SPEC_s_max_batch_time              (1 <<  8)
1881 #define EXT4_SPEC_s_min_batch_time              (1 <<  9)
1882 #define EXT4_SPEC_s_inode_readahead_blks        (1 << 10)
1883 #define EXT4_SPEC_s_li_wait_mult                (1 << 11)
1884 #define EXT4_SPEC_s_max_dir_size_kb             (1 << 12)
1885 #define EXT4_SPEC_s_stripe                      (1 << 13)
1886 #define EXT4_SPEC_s_resuid                      (1 << 14)
1887 #define EXT4_SPEC_s_resgid                      (1 << 15)
1888 #define EXT4_SPEC_s_commit_interval             (1 << 16)
1889 #define EXT4_SPEC_s_fc_debug_max_replay         (1 << 17)
1890 #define EXT4_SPEC_s_sb_block                    (1 << 18)
1891 #define EXT4_SPEC_mb_optimize_scan              (1 << 19)
1892
1893 struct ext4_fs_context {
1894         char            *s_qf_names[EXT4_MAXQUOTAS];
1895         struct fscrypt_dummy_policy dummy_enc_policy;
1896         int             s_jquota_fmt;   /* Format of quota to use */
1897 #ifdef CONFIG_EXT4_DEBUG
1898         int s_fc_debug_max_replay;
1899 #endif
1900         unsigned short  qname_spec;
1901         unsigned long   vals_s_flags;   /* Bits to set in s_flags */
1902         unsigned long   mask_s_flags;   /* Bits changed in s_flags */
1903         unsigned long   journal_devnum;
1904         unsigned long   s_commit_interval;
1905         unsigned long   s_stripe;
1906         unsigned int    s_inode_readahead_blks;
1907         unsigned int    s_want_extra_isize;
1908         unsigned int    s_li_wait_mult;
1909         unsigned int    s_max_dir_size_kb;
1910         unsigned int    journal_ioprio;
1911         unsigned int    vals_s_mount_opt;
1912         unsigned int    mask_s_mount_opt;
1913         unsigned int    vals_s_mount_opt2;
1914         unsigned int    mask_s_mount_opt2;
1915         unsigned long   vals_s_mount_flags;
1916         unsigned long   mask_s_mount_flags;
1917         unsigned int    opt_flags;      /* MOPT flags */
1918         unsigned int    spec;
1919         u32             s_max_batch_time;
1920         u32             s_min_batch_time;
1921         kuid_t          s_resuid;
1922         kgid_t          s_resgid;
1923         ext4_fsblk_t    s_sb_block;
1924 };
1925
1926 static void ext4_fc_free(struct fs_context *fc)
1927 {
1928         struct ext4_fs_context *ctx = fc->fs_private;
1929         int i;
1930
1931         if (!ctx)
1932                 return;
1933
1934         for (i = 0; i < EXT4_MAXQUOTAS; i++)
1935                 kfree(ctx->s_qf_names[i]);
1936
1937         fscrypt_free_dummy_policy(&ctx->dummy_enc_policy);
1938         kfree(ctx);
1939 }
1940
1941 int ext4_init_fs_context(struct fs_context *fc)
1942 {
1943         struct ext4_fs_context *ctx;
1944
1945         ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL);
1946         if (!ctx)
1947                 return -ENOMEM;
1948
1949         fc->fs_private = ctx;
1950         fc->ops = &ext4_context_ops;
1951
1952         return 0;
1953 }
1954
1955 #ifdef CONFIG_QUOTA
1956 /*
1957  * Note the name of the specified quota file.
1958  */
1959 static int note_qf_name(struct fs_context *fc, int qtype,
1960                        struct fs_parameter *param)
1961 {
1962         struct ext4_fs_context *ctx = fc->fs_private;
1963         char *qname;
1964
1965         if (param->size < 1) {
1966                 ext4_msg(NULL, KERN_ERR, "Missing quota name");
1967                 return -EINVAL;
1968         }
1969         if (strchr(param->string, '/')) {
1970                 ext4_msg(NULL, KERN_ERR,
1971                          "quotafile must be on filesystem root");
1972                 return -EINVAL;
1973         }
1974         if (ctx->s_qf_names[qtype]) {
1975                 if (strcmp(ctx->s_qf_names[qtype], param->string) != 0) {
1976                         ext4_msg(NULL, KERN_ERR,
1977                                  "%s quota file already specified",
1978                                  QTYPE2NAME(qtype));
1979                         return -EINVAL;
1980                 }
1981                 return 0;
1982         }
1983
1984         qname = kmemdup_nul(param->string, param->size, GFP_KERNEL);
1985         if (!qname) {
1986                 ext4_msg(NULL, KERN_ERR,
1987                          "Not enough memory for storing quotafile name");
1988                 return -ENOMEM;
1989         }
1990         ctx->s_qf_names[qtype] = qname;
1991         ctx->qname_spec |= 1 << qtype;
1992         ctx->spec |= EXT4_SPEC_JQUOTA;
1993         return 0;
1994 }
1995
1996 /*
1997  * Clear the name of the specified quota file.
1998  */
1999 static int unnote_qf_name(struct fs_context *fc, int qtype)
2000 {
2001         struct ext4_fs_context *ctx = fc->fs_private;
2002
2003         if (ctx->s_qf_names[qtype])
2004                 kfree(ctx->s_qf_names[qtype]);
2005
2006         ctx->s_qf_names[qtype] = NULL;
2007         ctx->qname_spec |= 1 << qtype;
2008         ctx->spec |= EXT4_SPEC_JQUOTA;
2009         return 0;
2010 }
2011 #endif
2012
2013 static int ext4_parse_test_dummy_encryption(const struct fs_parameter *param,
2014                                             struct ext4_fs_context *ctx)
2015 {
2016         int err;
2017
2018         if (!IS_ENABLED(CONFIG_FS_ENCRYPTION)) {
2019                 ext4_msg(NULL, KERN_WARNING,
2020                          "test_dummy_encryption option not supported");
2021                 return -EINVAL;
2022         }
2023         err = fscrypt_parse_test_dummy_encryption(param,
2024                                                   &ctx->dummy_enc_policy);
2025         if (err == -EINVAL) {
2026                 ext4_msg(NULL, KERN_WARNING,
2027                          "Value of option \"%s\" is unrecognized", param->key);
2028         } else if (err == -EEXIST) {
2029                 ext4_msg(NULL, KERN_WARNING,
2030                          "Conflicting test_dummy_encryption options");
2031                 return -EINVAL;
2032         }
2033         return err;
2034 }
2035
2036 #define EXT4_SET_CTX(name)                                              \
2037 static inline void ctx_set_##name(struct ext4_fs_context *ctx,          \
2038                                   unsigned long flag)                   \
2039 {                                                                       \
2040         ctx->mask_s_##name |= flag;                                     \
2041         ctx->vals_s_##name |= flag;                                     \
2042 }
2043
2044 #define EXT4_CLEAR_CTX(name)                                            \
2045 static inline void ctx_clear_##name(struct ext4_fs_context *ctx,        \
2046                                     unsigned long flag)                 \
2047 {                                                                       \
2048         ctx->mask_s_##name |= flag;                                     \
2049         ctx->vals_s_##name &= ~flag;                                    \
2050 }
2051
2052 #define EXT4_TEST_CTX(name)                                             \
2053 static inline unsigned long                                             \
2054 ctx_test_##name(struct ext4_fs_context *ctx, unsigned long flag)        \
2055 {                                                                       \
2056         return (ctx->vals_s_##name & flag);                             \
2057 }
2058
2059 EXT4_SET_CTX(flags); /* set only */
2060 EXT4_SET_CTX(mount_opt);
2061 EXT4_CLEAR_CTX(mount_opt);
2062 EXT4_TEST_CTX(mount_opt);
2063 EXT4_SET_CTX(mount_opt2);
2064 EXT4_CLEAR_CTX(mount_opt2);
2065 EXT4_TEST_CTX(mount_opt2);
2066
2067 static inline void ctx_set_mount_flag(struct ext4_fs_context *ctx, int bit)
2068 {
2069         set_bit(bit, &ctx->mask_s_mount_flags);
2070         set_bit(bit, &ctx->vals_s_mount_flags);
2071 }
2072
2073 static int ext4_parse_param(struct fs_context *fc, struct fs_parameter *param)
2074 {
2075         struct ext4_fs_context *ctx = fc->fs_private;
2076         struct fs_parse_result result;
2077         const struct mount_opts *m;
2078         int is_remount;
2079         kuid_t uid;
2080         kgid_t gid;
2081         int token;
2082
2083         token = fs_parse(fc, ext4_param_specs, param, &result);
2084         if (token < 0)
2085                 return token;
2086         is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE;
2087
2088         for (m = ext4_mount_opts; m->token != Opt_err; m++)
2089                 if (token == m->token)
2090                         break;
2091
2092         ctx->opt_flags |= m->flags;
2093
2094         if (m->flags & MOPT_EXPLICIT) {
2095                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
2096                         ctx_set_mount_opt2(ctx, EXT4_MOUNT2_EXPLICIT_DELALLOC);
2097                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
2098                         ctx_set_mount_opt2(ctx,
2099                                        EXT4_MOUNT2_EXPLICIT_JOURNAL_CHECKSUM);
2100                 } else
2101                         return -EINVAL;
2102         }
2103
2104         if (m->flags & MOPT_NOSUPPORT) {
2105                 ext4_msg(NULL, KERN_ERR, "%s option not supported",
2106                          param->key);
2107                 return 0;
2108         }
2109
2110         switch (token) {
2111 #ifdef CONFIG_QUOTA
2112         case Opt_usrjquota:
2113                 if (!*param->string)
2114                         return unnote_qf_name(fc, USRQUOTA);
2115                 else
2116                         return note_qf_name(fc, USRQUOTA, param);
2117         case Opt_grpjquota:
2118                 if (!*param->string)
2119                         return unnote_qf_name(fc, GRPQUOTA);
2120                 else
2121                         return note_qf_name(fc, GRPQUOTA, param);
2122 #endif
2123         case Opt_noacl:
2124         case Opt_nouser_xattr:
2125                 ext4_msg(NULL, KERN_WARNING, deprecated_msg, param->key, "3.5");
2126                 break;
2127         case Opt_sb:
2128                 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
2129                         ext4_msg(NULL, KERN_WARNING,
2130                                  "Ignoring %s option on remount", param->key);
2131                 } else {
2132                         ctx->s_sb_block = result.uint_32;
2133                         ctx->spec |= EXT4_SPEC_s_sb_block;
2134                 }
2135                 return 0;
2136         case Opt_removed:
2137                 ext4_msg(NULL, KERN_WARNING, "Ignoring removed %s option",
2138                          param->key);
2139                 return 0;
2140         case Opt_abort:
2141                 ctx_set_mount_flag(ctx, EXT4_MF_FS_ABORTED);
2142                 return 0;
2143         case Opt_i_version:
2144                 ext4_msg(NULL, KERN_WARNING, deprecated_msg, param->key, "5.20");
2145                 ext4_msg(NULL, KERN_WARNING, "Use iversion instead\n");
2146                 ctx_set_flags(ctx, SB_I_VERSION);
2147                 return 0;
2148         case Opt_inlinecrypt:
2149 #ifdef CONFIG_FS_ENCRYPTION_INLINE_CRYPT
2150                 ctx_set_flags(ctx, SB_INLINECRYPT);
2151 #else
2152                 ext4_msg(NULL, KERN_ERR, "inline encryption not supported");
2153 #endif
2154                 return 0;
2155         case Opt_errors:
2156                 ctx_clear_mount_opt(ctx, EXT4_MOUNT_ERRORS_MASK);
2157                 ctx_set_mount_opt(ctx, result.uint_32);
2158                 return 0;
2159 #ifdef CONFIG_QUOTA
2160         case Opt_jqfmt:
2161                 ctx->s_jquota_fmt = result.uint_32;
2162                 ctx->spec |= EXT4_SPEC_JQFMT;
2163                 return 0;
2164 #endif
2165         case Opt_data:
2166                 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS);
2167                 ctx_set_mount_opt(ctx, result.uint_32);
2168                 ctx->spec |= EXT4_SPEC_DATAJ;
2169                 return 0;
2170         case Opt_commit:
2171                 if (result.uint_32 == 0)
2172                         ctx->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE;
2173                 else if (result.uint_32 > INT_MAX / HZ) {
2174                         ext4_msg(NULL, KERN_ERR,
2175                                  "Invalid commit interval %d, "
2176                                  "must be smaller than %d",
2177                                  result.uint_32, INT_MAX / HZ);
2178                         return -EINVAL;
2179                 }
2180                 ctx->s_commit_interval = HZ * result.uint_32;
2181                 ctx->spec |= EXT4_SPEC_s_commit_interval;
2182                 return 0;
2183         case Opt_debug_want_extra_isize:
2184                 if ((result.uint_32 & 1) || (result.uint_32 < 4)) {
2185                         ext4_msg(NULL, KERN_ERR,
2186                                  "Invalid want_extra_isize %d", result.uint_32);
2187                         return -EINVAL;
2188                 }
2189                 ctx->s_want_extra_isize = result.uint_32;
2190                 ctx->spec |= EXT4_SPEC_s_want_extra_isize;
2191                 return 0;
2192         case Opt_max_batch_time:
2193                 ctx->s_max_batch_time = result.uint_32;
2194                 ctx->spec |= EXT4_SPEC_s_max_batch_time;
2195                 return 0;
2196         case Opt_min_batch_time:
2197                 ctx->s_min_batch_time = result.uint_32;
2198                 ctx->spec |= EXT4_SPEC_s_min_batch_time;
2199                 return 0;
2200         case Opt_inode_readahead_blks:
2201                 if (result.uint_32 &&
2202                     (result.uint_32 > (1 << 30) ||
2203                      !is_power_of_2(result.uint_32))) {
2204                         ext4_msg(NULL, KERN_ERR,
2205                                  "EXT4-fs: inode_readahead_blks must be "
2206                                  "0 or a power of 2 smaller than 2^31");
2207                         return -EINVAL;
2208                 }
2209                 ctx->s_inode_readahead_blks = result.uint_32;
2210                 ctx->spec |= EXT4_SPEC_s_inode_readahead_blks;
2211                 return 0;
2212         case Opt_init_itable:
2213                 ctx_set_mount_opt(ctx, EXT4_MOUNT_INIT_INODE_TABLE);
2214                 ctx->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
2215                 if (param->type == fs_value_is_string)
2216                         ctx->s_li_wait_mult = result.uint_32;
2217                 ctx->spec |= EXT4_SPEC_s_li_wait_mult;
2218                 return 0;
2219         case Opt_max_dir_size_kb:
2220                 ctx->s_max_dir_size_kb = result.uint_32;
2221                 ctx->spec |= EXT4_SPEC_s_max_dir_size_kb;
2222                 return 0;
2223 #ifdef CONFIG_EXT4_DEBUG
2224         case Opt_fc_debug_max_replay:
2225                 ctx->s_fc_debug_max_replay = result.uint_32;
2226                 ctx->spec |= EXT4_SPEC_s_fc_debug_max_replay;
2227                 return 0;
2228 #endif
2229         case Opt_stripe:
2230                 ctx->s_stripe = result.uint_32;
2231                 ctx->spec |= EXT4_SPEC_s_stripe;
2232                 return 0;
2233         case Opt_resuid:
2234                 uid = make_kuid(current_user_ns(), result.uint_32);
2235                 if (!uid_valid(uid)) {
2236                         ext4_msg(NULL, KERN_ERR, "Invalid uid value %d",
2237                                  result.uint_32);
2238                         return -EINVAL;
2239                 }
2240                 ctx->s_resuid = uid;
2241                 ctx->spec |= EXT4_SPEC_s_resuid;
2242                 return 0;
2243         case Opt_resgid:
2244                 gid = make_kgid(current_user_ns(), result.uint_32);
2245                 if (!gid_valid(gid)) {
2246                         ext4_msg(NULL, KERN_ERR, "Invalid gid value %d",
2247                                  result.uint_32);
2248                         return -EINVAL;
2249                 }
2250                 ctx->s_resgid = gid;
2251                 ctx->spec |= EXT4_SPEC_s_resgid;
2252                 return 0;
2253         case Opt_journal_dev:
2254                 if (is_remount) {
2255                         ext4_msg(NULL, KERN_ERR,
2256                                  "Cannot specify journal on remount");
2257                         return -EINVAL;
2258                 }
2259                 ctx->journal_devnum = result.uint_32;
2260                 ctx->spec |= EXT4_SPEC_JOURNAL_DEV;
2261                 return 0;
2262         case Opt_journal_path:
2263         {
2264                 struct inode *journal_inode;
2265                 struct path path;
2266                 int error;
2267
2268                 if (is_remount) {
2269                         ext4_msg(NULL, KERN_ERR,
2270                                  "Cannot specify journal on remount");
2271                         return -EINVAL;
2272                 }
2273
2274                 error = fs_lookup_param(fc, param, 1, &path);
2275                 if (error) {
2276                         ext4_msg(NULL, KERN_ERR, "error: could not find "
2277                                  "journal device path");
2278                         return -EINVAL;
2279                 }
2280
2281                 journal_inode = d_inode(path.dentry);
2282                 ctx->journal_devnum = new_encode_dev(journal_inode->i_rdev);
2283                 ctx->spec |= EXT4_SPEC_JOURNAL_DEV;
2284                 path_put(&path);
2285                 return 0;
2286         }
2287         case Opt_journal_ioprio:
2288                 if (result.uint_32 > 7) {
2289                         ext4_msg(NULL, KERN_ERR, "Invalid journal IO priority"
2290                                  " (must be 0-7)");
2291                         return -EINVAL;
2292                 }
2293                 ctx->journal_ioprio =
2294                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, result.uint_32);
2295                 ctx->spec |= EXT4_SPEC_JOURNAL_IOPRIO;
2296                 return 0;
2297         case Opt_test_dummy_encryption:
2298                 return ext4_parse_test_dummy_encryption(param, ctx);
2299         case Opt_dax:
2300         case Opt_dax_type:
2301 #ifdef CONFIG_FS_DAX
2302         {
2303                 int type = (token == Opt_dax) ?
2304                            Opt_dax : result.uint_32;
2305
2306                 switch (type) {
2307                 case Opt_dax:
2308                 case Opt_dax_always:
2309                         ctx_set_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS);
2310                         ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER);
2311                         break;
2312                 case Opt_dax_never:
2313                         ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER);
2314                         ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS);
2315                         break;
2316                 case Opt_dax_inode:
2317                         ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS);
2318                         ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER);
2319                         /* Strictly for printing options */
2320                         ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE);
2321                         break;
2322                 }
2323                 return 0;
2324         }
2325 #else
2326                 ext4_msg(NULL, KERN_INFO, "dax option not supported");
2327                 return -EINVAL;
2328 #endif
2329         case Opt_data_err:
2330                 if (result.uint_32 == Opt_data_err_abort)
2331                         ctx_set_mount_opt(ctx, m->mount_opt);
2332                 else if (result.uint_32 == Opt_data_err_ignore)
2333                         ctx_clear_mount_opt(ctx, m->mount_opt);
2334                 return 0;
2335         case Opt_mb_optimize_scan:
2336                 if (result.int_32 == 1) {
2337                         ctx_set_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN);
2338                         ctx->spec |= EXT4_SPEC_mb_optimize_scan;
2339                 } else if (result.int_32 == 0) {
2340                         ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN);
2341                         ctx->spec |= EXT4_SPEC_mb_optimize_scan;
2342                 } else {
2343                         ext4_msg(NULL, KERN_WARNING,
2344                                  "mb_optimize_scan should be set to 0 or 1.");
2345                         return -EINVAL;
2346                 }
2347                 return 0;
2348         }
2349
2350         /*
2351          * At this point we should only be getting options requiring MOPT_SET,
2352          * or MOPT_CLEAR. Anything else is a bug
2353          */
2354         if (m->token == Opt_err) {
2355                 ext4_msg(NULL, KERN_WARNING, "buggy handling of option %s",
2356                          param->key);
2357                 WARN_ON(1);
2358                 return -EINVAL;
2359         }
2360
2361         else {
2362                 unsigned int set = 0;
2363
2364                 if ((param->type == fs_value_is_flag) ||
2365                     result.uint_32 > 0)
2366                         set = 1;
2367
2368                 if (m->flags & MOPT_CLEAR)
2369                         set = !set;
2370                 else if (unlikely(!(m->flags & MOPT_SET))) {
2371                         ext4_msg(NULL, KERN_WARNING,
2372                                  "buggy handling of option %s",
2373                                  param->key);
2374                         WARN_ON(1);
2375                         return -EINVAL;
2376                 }
2377                 if (m->flags & MOPT_2) {
2378                         if (set != 0)
2379                                 ctx_set_mount_opt2(ctx, m->mount_opt);
2380                         else
2381                                 ctx_clear_mount_opt2(ctx, m->mount_opt);
2382                 } else {
2383                         if (set != 0)
2384                                 ctx_set_mount_opt(ctx, m->mount_opt);
2385                         else
2386                                 ctx_clear_mount_opt(ctx, m->mount_opt);
2387                 }
2388         }
2389
2390         return 0;
2391 }
2392
2393 static int parse_options(struct fs_context *fc, char *options)
2394 {
2395         struct fs_parameter param;
2396         int ret;
2397         char *key;
2398
2399         if (!options)
2400                 return 0;
2401
2402         while ((key = strsep(&options, ",")) != NULL) {
2403                 if (*key) {
2404                         size_t v_len = 0;
2405                         char *value = strchr(key, '=');
2406
2407                         param.type = fs_value_is_flag;
2408                         param.string = NULL;
2409
2410                         if (value) {
2411                                 if (value == key)
2412                                         continue;
2413
2414                                 *value++ = 0;
2415                                 v_len = strlen(value);
2416                                 param.string = kmemdup_nul(value, v_len,
2417                                                            GFP_KERNEL);
2418                                 if (!param.string)
2419                                         return -ENOMEM;
2420                                 param.type = fs_value_is_string;
2421                         }
2422
2423                         param.key = key;
2424                         param.size = v_len;
2425
2426                         ret = ext4_parse_param(fc, &param);
2427                         if (param.string)
2428                                 kfree(param.string);
2429                         if (ret < 0)
2430                                 return ret;
2431                 }
2432         }
2433
2434         ret = ext4_validate_options(fc);
2435         if (ret < 0)
2436                 return ret;
2437
2438         return 0;
2439 }
2440
2441 static int parse_apply_sb_mount_options(struct super_block *sb,
2442                                         struct ext4_fs_context *m_ctx)
2443 {
2444         struct ext4_sb_info *sbi = EXT4_SB(sb);
2445         char *s_mount_opts = NULL;
2446         struct ext4_fs_context *s_ctx = NULL;
2447         struct fs_context *fc = NULL;
2448         int ret = -ENOMEM;
2449
2450         if (!sbi->s_es->s_mount_opts[0])
2451                 return 0;
2452
2453         s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
2454                                 sizeof(sbi->s_es->s_mount_opts),
2455                                 GFP_KERNEL);
2456         if (!s_mount_opts)
2457                 return ret;
2458
2459         fc = kzalloc(sizeof(struct fs_context), GFP_KERNEL);
2460         if (!fc)
2461                 goto out_free;
2462
2463         s_ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL);
2464         if (!s_ctx)
2465                 goto out_free;
2466
2467         fc->fs_private = s_ctx;
2468         fc->s_fs_info = sbi;
2469
2470         ret = parse_options(fc, s_mount_opts);
2471         if (ret < 0)
2472                 goto parse_failed;
2473
2474         ret = ext4_check_opt_consistency(fc, sb);
2475         if (ret < 0) {
2476 parse_failed:
2477                 ext4_msg(sb, KERN_WARNING,
2478                          "failed to parse options in superblock: %s",
2479                          s_mount_opts);
2480                 ret = 0;
2481                 goto out_free;
2482         }
2483
2484         if (s_ctx->spec & EXT4_SPEC_JOURNAL_DEV)
2485                 m_ctx->journal_devnum = s_ctx->journal_devnum;
2486         if (s_ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)
2487                 m_ctx->journal_ioprio = s_ctx->journal_ioprio;
2488
2489         ext4_apply_options(fc, sb);
2490         ret = 0;
2491
2492 out_free:
2493         if (fc) {
2494                 ext4_fc_free(fc);
2495                 kfree(fc);
2496         }
2497         kfree(s_mount_opts);
2498         return ret;
2499 }
2500
2501 static void ext4_apply_quota_options(struct fs_context *fc,
2502                                      struct super_block *sb)
2503 {
2504 #ifdef CONFIG_QUOTA
2505         bool quota_feature = ext4_has_feature_quota(sb);
2506         struct ext4_fs_context *ctx = fc->fs_private;
2507         struct ext4_sb_info *sbi = EXT4_SB(sb);
2508         char *qname;
2509         int i;
2510
2511         if (quota_feature)
2512                 return;
2513
2514         if (ctx->spec & EXT4_SPEC_JQUOTA) {
2515                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2516                         if (!(ctx->qname_spec & (1 << i)))
2517                                 continue;
2518
2519                         qname = ctx->s_qf_names[i]; /* May be NULL */
2520                         if (qname)
2521                                 set_opt(sb, QUOTA);
2522                         ctx->s_qf_names[i] = NULL;
2523                         qname = rcu_replace_pointer(sbi->s_qf_names[i], qname,
2524                                                 lockdep_is_held(&sb->s_umount));
2525                         if (qname)
2526                                 kfree_rcu(qname);
2527                 }
2528         }
2529
2530         if (ctx->spec & EXT4_SPEC_JQFMT)
2531                 sbi->s_jquota_fmt = ctx->s_jquota_fmt;
2532 #endif
2533 }
2534
2535 /*
2536  * Check quota settings consistency.
2537  */
2538 static int ext4_check_quota_consistency(struct fs_context *fc,
2539                                         struct super_block *sb)
2540 {
2541 #ifdef CONFIG_QUOTA
2542         struct ext4_fs_context *ctx = fc->fs_private;
2543         struct ext4_sb_info *sbi = EXT4_SB(sb);
2544         bool quota_feature = ext4_has_feature_quota(sb);
2545         bool quota_loaded = sb_any_quota_loaded(sb);
2546         bool usr_qf_name, grp_qf_name, usrquota, grpquota;
2547         int quota_flags, i;
2548
2549         /*
2550          * We do the test below only for project quotas. 'usrquota' and
2551          * 'grpquota' mount options are allowed even without quota feature
2552          * to support legacy quotas in quota files.
2553          */
2554         if (ctx_test_mount_opt(ctx, EXT4_MOUNT_PRJQUOTA) &&
2555             !ext4_has_feature_project(sb)) {
2556                 ext4_msg(NULL, KERN_ERR, "Project quota feature not enabled. "
2557                          "Cannot enable project quota enforcement.");
2558                 return -EINVAL;
2559         }
2560
2561         quota_flags = EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
2562                       EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA;
2563         if (quota_loaded &&
2564             ctx->mask_s_mount_opt & quota_flags &&
2565             !ctx_test_mount_opt(ctx, quota_flags))
2566                 goto err_quota_change;
2567
2568         if (ctx->spec & EXT4_SPEC_JQUOTA) {
2569
2570                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2571                         if (!(ctx->qname_spec & (1 << i)))
2572                                 continue;
2573
2574                         if (quota_loaded &&
2575                             !!sbi->s_qf_names[i] != !!ctx->s_qf_names[i])
2576                                 goto err_jquota_change;
2577
2578                         if (sbi->s_qf_names[i] && ctx->s_qf_names[i] &&
2579                             strcmp(get_qf_name(sb, sbi, i),
2580                                    ctx->s_qf_names[i]) != 0)
2581                                 goto err_jquota_specified;
2582                 }
2583
2584                 if (quota_feature) {
2585                         ext4_msg(NULL, KERN_INFO,
2586                                  "Journaled quota options ignored when "
2587                                  "QUOTA feature is enabled");
2588                         return 0;
2589                 }
2590         }
2591
2592         if (ctx->spec & EXT4_SPEC_JQFMT) {
2593                 if (sbi->s_jquota_fmt != ctx->s_jquota_fmt && quota_loaded)
2594                         goto err_jquota_change;
2595                 if (quota_feature) {
2596                         ext4_msg(NULL, KERN_INFO, "Quota format mount options "
2597                                  "ignored when QUOTA feature is enabled");
2598                         return 0;
2599                 }
2600         }
2601
2602         /* Make sure we don't mix old and new quota format */
2603         usr_qf_name = (get_qf_name(sb, sbi, USRQUOTA) ||
2604                        ctx->s_qf_names[USRQUOTA]);
2605         grp_qf_name = (get_qf_name(sb, sbi, GRPQUOTA) ||
2606                        ctx->s_qf_names[GRPQUOTA]);
2607
2608         usrquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) ||
2609                     test_opt(sb, USRQUOTA));
2610
2611         grpquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) ||
2612                     test_opt(sb, GRPQUOTA));
2613
2614         if (usr_qf_name) {
2615                 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA);
2616                 usrquota = false;
2617         }
2618         if (grp_qf_name) {
2619                 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA);
2620                 grpquota = false;
2621         }
2622
2623         if (usr_qf_name || grp_qf_name) {
2624                 if (usrquota || grpquota) {
2625                         ext4_msg(NULL, KERN_ERR, "old and new quota "
2626                                  "format mixing");
2627                         return -EINVAL;
2628                 }
2629
2630                 if (!(ctx->spec & EXT4_SPEC_JQFMT || sbi->s_jquota_fmt)) {
2631                         ext4_msg(NULL, KERN_ERR, "journaled quota format "
2632                                  "not specified");
2633                         return -EINVAL;
2634                 }
2635         }
2636
2637         return 0;
2638
2639 err_quota_change:
2640         ext4_msg(NULL, KERN_ERR,
2641                  "Cannot change quota options when quota turned on");
2642         return -EINVAL;
2643 err_jquota_change:
2644         ext4_msg(NULL, KERN_ERR, "Cannot change journaled quota "
2645                  "options when quota turned on");
2646         return -EINVAL;
2647 err_jquota_specified:
2648         ext4_msg(NULL, KERN_ERR, "%s quota file already specified",
2649                  QTYPE2NAME(i));
2650         return -EINVAL;
2651 #else
2652         return 0;
2653 #endif
2654 }
2655
2656 static int ext4_check_test_dummy_encryption(const struct fs_context *fc,
2657                                             struct super_block *sb)
2658 {
2659         const struct ext4_fs_context *ctx = fc->fs_private;
2660         const struct ext4_sb_info *sbi = EXT4_SB(sb);
2661         int err;
2662
2663         if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy))
2664                 return 0;
2665
2666         if (!ext4_has_feature_encrypt(sb)) {
2667                 ext4_msg(NULL, KERN_WARNING,
2668                          "test_dummy_encryption requires encrypt feature");
2669                 return -EINVAL;
2670         }
2671         /*
2672          * This mount option is just for testing, and it's not worthwhile to
2673          * implement the extra complexity (e.g. RCU protection) that would be
2674          * needed to allow it to be set or changed during remount.  We do allow
2675          * it to be specified during remount, but only if there is no change.
2676          */
2677         if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
2678                 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy,
2679                                                  &ctx->dummy_enc_policy))
2680                         return 0;
2681                 ext4_msg(NULL, KERN_WARNING,
2682                          "Can't set or change test_dummy_encryption on remount");
2683                 return -EINVAL;
2684         }
2685         /* Also make sure s_mount_opts didn't contain a conflicting value. */
2686         if (fscrypt_is_dummy_policy_set(&sbi->s_dummy_enc_policy)) {
2687                 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy,
2688                                                  &ctx->dummy_enc_policy))
2689                         return 0;
2690                 ext4_msg(NULL, KERN_WARNING,
2691                          "Conflicting test_dummy_encryption options");
2692                 return -EINVAL;
2693         }
2694         /*
2695          * fscrypt_add_test_dummy_key() technically changes the super_block, so
2696          * technically it should be delayed until ext4_apply_options() like the
2697          * other changes.  But since we never get here for remounts (see above),
2698          * and this is the last chance to report errors, we do it here.
2699          */
2700         err = fscrypt_add_test_dummy_key(sb, &ctx->dummy_enc_policy);
2701         if (err)
2702                 ext4_msg(NULL, KERN_WARNING,
2703                          "Error adding test dummy encryption key [%d]", err);
2704         return err;
2705 }
2706
2707 static void ext4_apply_test_dummy_encryption(struct ext4_fs_context *ctx,
2708                                              struct super_block *sb)
2709 {
2710         if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy) ||
2711             /* if already set, it was already verified to be the same */
2712             fscrypt_is_dummy_policy_set(&EXT4_SB(sb)->s_dummy_enc_policy))
2713                 return;
2714         EXT4_SB(sb)->s_dummy_enc_policy = ctx->dummy_enc_policy;
2715         memset(&ctx->dummy_enc_policy, 0, sizeof(ctx->dummy_enc_policy));
2716         ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled");
2717 }
2718
2719 static int ext4_check_opt_consistency(struct fs_context *fc,
2720                                       struct super_block *sb)
2721 {
2722         struct ext4_fs_context *ctx = fc->fs_private;
2723         struct ext4_sb_info *sbi = fc->s_fs_info;
2724         int is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE;
2725         int err;
2726
2727         if ((ctx->opt_flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
2728                 ext4_msg(NULL, KERN_ERR,
2729                          "Mount option(s) incompatible with ext2");
2730                 return -EINVAL;
2731         }
2732         if ((ctx->opt_flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
2733                 ext4_msg(NULL, KERN_ERR,
2734                          "Mount option(s) incompatible with ext3");
2735                 return -EINVAL;
2736         }
2737
2738         if (ctx->s_want_extra_isize >
2739             (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE)) {
2740                 ext4_msg(NULL, KERN_ERR,
2741                          "Invalid want_extra_isize %d",
2742                          ctx->s_want_extra_isize);
2743                 return -EINVAL;
2744         }
2745
2746         if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DIOREAD_NOLOCK)) {
2747                 int blocksize =
2748                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
2749                 if (blocksize < PAGE_SIZE)
2750                         ext4_msg(NULL, KERN_WARNING, "Warning: mounting with an "
2751                                  "experimental mount option 'dioread_nolock' "
2752                                  "for blocksize < PAGE_SIZE");
2753         }
2754
2755         err = ext4_check_test_dummy_encryption(fc, sb);
2756         if (err)
2757                 return err;
2758
2759         if ((ctx->spec & EXT4_SPEC_DATAJ) && is_remount) {
2760                 if (!sbi->s_journal) {
2761                         ext4_msg(NULL, KERN_WARNING,
2762                                  "Remounting file system with no journal "
2763                                  "so ignoring journalled data option");
2764                         ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS);
2765                 } else if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS) !=
2766                            test_opt(sb, DATA_FLAGS)) {
2767                         ext4_msg(NULL, KERN_ERR, "Cannot change data mode "
2768                                  "on remount");
2769                         return -EINVAL;
2770                 }
2771         }
2772
2773         if (is_remount) {
2774                 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) &&
2775                     (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) {
2776                         ext4_msg(NULL, KERN_ERR, "can't mount with "
2777                                  "both data=journal and dax");
2778                         return -EINVAL;
2779                 }
2780
2781                 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) &&
2782                     (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2783                      (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) {
2784 fail_dax_change_remount:
2785                         ext4_msg(NULL, KERN_ERR, "can't change "
2786                                  "dax mount option while remounting");
2787                         return -EINVAL;
2788                 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER) &&
2789                          (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2790                           (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS))) {
2791                         goto fail_dax_change_remount;
2792                 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE) &&
2793                            ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2794                             (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2795                             !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE))) {
2796                         goto fail_dax_change_remount;
2797                 }
2798         }
2799
2800         return ext4_check_quota_consistency(fc, sb);
2801 }
2802
2803 static void ext4_apply_options(struct fs_context *fc, struct super_block *sb)
2804 {
2805         struct ext4_fs_context *ctx = fc->fs_private;
2806         struct ext4_sb_info *sbi = fc->s_fs_info;
2807
2808         sbi->s_mount_opt &= ~ctx->mask_s_mount_opt;
2809         sbi->s_mount_opt |= ctx->vals_s_mount_opt;
2810         sbi->s_mount_opt2 &= ~ctx->mask_s_mount_opt2;
2811         sbi->s_mount_opt2 |= ctx->vals_s_mount_opt2;
2812         sbi->s_mount_flags &= ~ctx->mask_s_mount_flags;
2813         sbi->s_mount_flags |= ctx->vals_s_mount_flags;
2814         sb->s_flags &= ~ctx->mask_s_flags;
2815         sb->s_flags |= ctx->vals_s_flags;
2816
2817         /*
2818          * i_version differs from common mount option iversion so we have
2819          * to let vfs know that it was set, otherwise it would get cleared
2820          * on remount
2821          */
2822         if (ctx->mask_s_flags & SB_I_VERSION)
2823                 fc->sb_flags |= SB_I_VERSION;
2824
2825 #define APPLY(X) ({ if (ctx->spec & EXT4_SPEC_##X) sbi->X = ctx->X; })
2826         APPLY(s_commit_interval);
2827         APPLY(s_stripe);
2828         APPLY(s_max_batch_time);
2829         APPLY(s_min_batch_time);
2830         APPLY(s_want_extra_isize);
2831         APPLY(s_inode_readahead_blks);
2832         APPLY(s_max_dir_size_kb);
2833         APPLY(s_li_wait_mult);
2834         APPLY(s_resgid);
2835         APPLY(s_resuid);
2836
2837 #ifdef CONFIG_EXT4_DEBUG
2838         APPLY(s_fc_debug_max_replay);
2839 #endif
2840
2841         ext4_apply_quota_options(fc, sb);
2842         ext4_apply_test_dummy_encryption(ctx, sb);
2843 }
2844
2845
2846 static int ext4_validate_options(struct fs_context *fc)
2847 {
2848 #ifdef CONFIG_QUOTA
2849         struct ext4_fs_context *ctx = fc->fs_private;
2850         char *usr_qf_name, *grp_qf_name;
2851
2852         usr_qf_name = ctx->s_qf_names[USRQUOTA];
2853         grp_qf_name = ctx->s_qf_names[GRPQUOTA];
2854
2855         if (usr_qf_name || grp_qf_name) {
2856                 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) && usr_qf_name)
2857                         ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA);
2858
2859                 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) && grp_qf_name)
2860                         ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA);
2861
2862                 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) ||
2863                     ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA)) {
2864                         ext4_msg(NULL, KERN_ERR, "old and new quota "
2865                                  "format mixing");
2866                         return -EINVAL;
2867                 }
2868         }
2869 #endif
2870         return 1;
2871 }
2872
2873 static inline void ext4_show_quota_options(struct seq_file *seq,
2874                                            struct super_block *sb)
2875 {
2876 #if defined(CONFIG_QUOTA)
2877         struct ext4_sb_info *sbi = EXT4_SB(sb);
2878         char *usr_qf_name, *grp_qf_name;
2879
2880         if (sbi->s_jquota_fmt) {
2881                 char *fmtname = "";
2882
2883                 switch (sbi->s_jquota_fmt) {
2884                 case QFMT_VFS_OLD:
2885                         fmtname = "vfsold";
2886                         break;
2887                 case QFMT_VFS_V0:
2888                         fmtname = "vfsv0";
2889                         break;
2890                 case QFMT_VFS_V1:
2891                         fmtname = "vfsv1";
2892                         break;
2893                 }
2894                 seq_printf(seq, ",jqfmt=%s", fmtname);
2895         }
2896
2897         rcu_read_lock();
2898         usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]);
2899         grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]);
2900         if (usr_qf_name)
2901                 seq_show_option(seq, "usrjquota", usr_qf_name);
2902         if (grp_qf_name)
2903                 seq_show_option(seq, "grpjquota", grp_qf_name);
2904         rcu_read_unlock();
2905 #endif
2906 }
2907
2908 static const char *token2str(int token)
2909 {
2910         const struct fs_parameter_spec *spec;
2911
2912         for (spec = ext4_param_specs; spec->name != NULL; spec++)
2913                 if (spec->opt == token && !spec->type)
2914                         break;
2915         return spec->name;
2916 }
2917
2918 /*
2919  * Show an option if
2920  *  - it's set to a non-default value OR
2921  *  - if the per-sb default is different from the global default
2922  */
2923 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2924                               int nodefs)
2925 {
2926         struct ext4_sb_info *sbi = EXT4_SB(sb);
2927         struct ext4_super_block *es = sbi->s_es;
2928         int def_errors, def_mount_opt = sbi->s_def_mount_opt;
2929         const struct mount_opts *m;
2930         char sep = nodefs ? '\n' : ',';
2931
2932 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2933 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2934
2935         if (sbi->s_sb_block != 1)
2936                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2937
2938         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2939                 int want_set = m->flags & MOPT_SET;
2940                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2941                     m->flags & MOPT_SKIP)
2942                         continue;
2943                 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2944                         continue; /* skip if same as the default */
2945                 if ((want_set &&
2946                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2947                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2948                         continue; /* select Opt_noFoo vs Opt_Foo */
2949                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2950         }
2951
2952         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2953             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2954                 SEQ_OPTS_PRINT("resuid=%u",
2955                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2956         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2957             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2958                 SEQ_OPTS_PRINT("resgid=%u",
2959                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2960         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2961         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2962                 SEQ_OPTS_PUTS("errors=remount-ro");
2963         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2964                 SEQ_OPTS_PUTS("errors=continue");
2965         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2966                 SEQ_OPTS_PUTS("errors=panic");
2967         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2968                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2969         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2970                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2971         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2972                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2973         if (sb->s_flags & SB_I_VERSION)
2974                 SEQ_OPTS_PUTS("i_version");
2975         if (nodefs || sbi->s_stripe)
2976                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2977         if (nodefs || EXT4_MOUNT_DATA_FLAGS &
2978                         (sbi->s_mount_opt ^ def_mount_opt)) {
2979                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2980                         SEQ_OPTS_PUTS("data=journal");
2981                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2982                         SEQ_OPTS_PUTS("data=ordered");
2983                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2984                         SEQ_OPTS_PUTS("data=writeback");
2985         }
2986         if (nodefs ||
2987             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2988                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2989                                sbi->s_inode_readahead_blks);
2990
2991         if (test_opt(sb, INIT_INODE_TABLE) && (nodefs ||
2992                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2993                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2994         if (nodefs || sbi->s_max_dir_size_kb)
2995                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2996         if (test_opt(sb, DATA_ERR_ABORT))
2997                 SEQ_OPTS_PUTS("data_err=abort");
2998
2999         fscrypt_show_test_dummy_encryption(seq, sep, sb);
3000
3001         if (sb->s_flags & SB_INLINECRYPT)
3002                 SEQ_OPTS_PUTS("inlinecrypt");
3003
3004         if (test_opt(sb, DAX_ALWAYS)) {
3005                 if (IS_EXT2_SB(sb))
3006                         SEQ_OPTS_PUTS("dax");
3007                 else
3008                         SEQ_OPTS_PUTS("dax=always");
3009         } else if (test_opt2(sb, DAX_NEVER)) {
3010                 SEQ_OPTS_PUTS("dax=never");
3011         } else if (test_opt2(sb, DAX_INODE)) {
3012                 SEQ_OPTS_PUTS("dax=inode");
3013         }
3014
3015         if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD &&
3016                         !test_opt2(sb, MB_OPTIMIZE_SCAN)) {
3017                 SEQ_OPTS_PUTS("mb_optimize_scan=0");
3018         } else if (sbi->s_groups_count < MB_DEFAULT_LINEAR_SCAN_THRESHOLD &&
3019                         test_opt2(sb, MB_OPTIMIZE_SCAN)) {
3020                 SEQ_OPTS_PUTS("mb_optimize_scan=1");
3021         }
3022
3023         ext4_show_quota_options(seq, sb);
3024         return 0;
3025 }
3026
3027 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
3028 {
3029         return _ext4_show_options(seq, root->d_sb, 0);
3030 }
3031
3032 int ext4_seq_options_show(struct seq_file *seq, void *offset)
3033 {
3034         struct super_block *sb = seq->private;
3035         int rc;
3036
3037         seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
3038         rc = _ext4_show_options(seq, sb, 1);
3039         seq_puts(seq, "\n");
3040         return rc;
3041 }
3042
3043 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
3044                             int read_only)
3045 {
3046         struct ext4_sb_info *sbi = EXT4_SB(sb);
3047         int err = 0;
3048
3049         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
3050                 ext4_msg(sb, KERN_ERR, "revision level too high, "
3051                          "forcing read-only mode");
3052                 err = -EROFS;
3053                 goto done;
3054         }
3055         if (read_only)
3056                 goto done;
3057         if (!(sbi->s_mount_state & EXT4_VALID_FS))
3058                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
3059                          "running e2fsck is recommended");
3060         else if (sbi->s_mount_state & EXT4_ERROR_FS)
3061                 ext4_msg(sb, KERN_WARNING,
3062                          "warning: mounting fs with errors, "
3063                          "running e2fsck is recommended");
3064         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
3065                  le16_to_cpu(es->s_mnt_count) >=
3066                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
3067                 ext4_msg(sb, KERN_WARNING,
3068                          "warning: maximal mount count reached, "
3069                          "running e2fsck is recommended");
3070         else if (le32_to_cpu(es->s_checkinterval) &&
3071                  (ext4_get_tstamp(es, s_lastcheck) +
3072                   le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds()))
3073                 ext4_msg(sb, KERN_WARNING,
3074                          "warning: checktime reached, "
3075                          "running e2fsck is recommended");
3076         if (!sbi->s_journal)
3077                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
3078         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
3079                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
3080         le16_add_cpu(&es->s_mnt_count, 1);
3081         ext4_update_tstamp(es, s_mtime);
3082         if (sbi->s_journal) {
3083                 ext4_set_feature_journal_needs_recovery(sb);
3084                 if (ext4_has_feature_orphan_file(sb))
3085                         ext4_set_feature_orphan_present(sb);
3086         }
3087
3088         err = ext4_commit_super(sb);
3089 done:
3090         if (test_opt(sb, DEBUG))
3091                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
3092                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
3093                         sb->s_blocksize,
3094                         sbi->s_groups_count,
3095                         EXT4_BLOCKS_PER_GROUP(sb),
3096                         EXT4_INODES_PER_GROUP(sb),
3097                         sbi->s_mount_opt, sbi->s_mount_opt2);
3098         return err;
3099 }
3100
3101 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
3102 {
3103         struct ext4_sb_info *sbi = EXT4_SB(sb);
3104         struct flex_groups **old_groups, **new_groups;
3105         int size, i, j;
3106
3107         if (!sbi->s_log_groups_per_flex)
3108                 return 0;
3109
3110         size = ext4_flex_group(sbi, ngroup - 1) + 1;
3111         if (size <= sbi->s_flex_groups_allocated)
3112                 return 0;
3113
3114         new_groups = kvzalloc(roundup_pow_of_two(size *
3115                               sizeof(*sbi->s_flex_groups)), GFP_KERNEL);
3116         if (!new_groups) {
3117                 ext4_msg(sb, KERN_ERR,
3118                          "not enough memory for %d flex group pointers", size);
3119                 return -ENOMEM;
3120         }
3121         for (i = sbi->s_flex_groups_allocated; i < size; i++) {
3122                 new_groups[i] = kvzalloc(roundup_pow_of_two(
3123                                          sizeof(struct flex_groups)),
3124                                          GFP_KERNEL);
3125                 if (!new_groups[i]) {
3126                         for (j = sbi->s_flex_groups_allocated; j < i; j++)
3127                                 kvfree(new_groups[j]);
3128                         kvfree(new_groups);
3129                         ext4_msg(sb, KERN_ERR,
3130                                  "not enough memory for %d flex groups", size);
3131                         return -ENOMEM;
3132                 }
3133         }
3134         rcu_read_lock();
3135         old_groups = rcu_dereference(sbi->s_flex_groups);
3136         if (old_groups)
3137                 memcpy(new_groups, old_groups,
3138                        (sbi->s_flex_groups_allocated *
3139                         sizeof(struct flex_groups *)));
3140         rcu_read_unlock();
3141         rcu_assign_pointer(sbi->s_flex_groups, new_groups);
3142         sbi->s_flex_groups_allocated = size;
3143         if (old_groups)
3144                 ext4_kvfree_array_rcu(old_groups);
3145         return 0;
3146 }
3147
3148 static int ext4_fill_flex_info(struct super_block *sb)
3149 {
3150         struct ext4_sb_info *sbi = EXT4_SB(sb);
3151         struct ext4_group_desc *gdp = NULL;
3152         struct flex_groups *fg;
3153         ext4_group_t flex_group;
3154         int i, err;
3155
3156         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
3157         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
3158                 sbi->s_log_groups_per_flex = 0;
3159                 return 1;
3160         }
3161
3162         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
3163         if (err)
3164                 goto failed;
3165
3166         for (i = 0; i < sbi->s_groups_count; i++) {
3167                 gdp = ext4_get_group_desc(sb, i, NULL);
3168
3169                 flex_group = ext4_flex_group(sbi, i);
3170                 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
3171                 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes);
3172                 atomic64_add(ext4_free_group_clusters(sb, gdp),
3173                              &fg->free_clusters);
3174                 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs);
3175         }
3176
3177         return 1;
3178 failed:
3179         return 0;
3180 }
3181
3182 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
3183                                    struct ext4_group_desc *gdp)
3184 {
3185         int offset = offsetof(struct ext4_group_desc, bg_checksum);
3186         __u16 crc = 0;
3187         __le32 le_group = cpu_to_le32(block_group);
3188         struct ext4_sb_info *sbi = EXT4_SB(sb);
3189
3190         if (ext4_has_metadata_csum(sbi->s_sb)) {
3191                 /* Use new metadata_csum algorithm */
3192                 __u32 csum32;
3193                 __u16 dummy_csum = 0;
3194
3195                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
3196                                      sizeof(le_group));
3197                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
3198                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
3199                                      sizeof(dummy_csum));
3200                 offset += sizeof(dummy_csum);
3201                 if (offset < sbi->s_desc_size)
3202                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
3203                                              sbi->s_desc_size - offset);
3204
3205                 crc = csum32 & 0xFFFF;
3206                 goto out;
3207         }
3208
3209         /* old crc16 code */
3210         if (!ext4_has_feature_gdt_csum(sb))
3211                 return 0;
3212
3213         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
3214         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
3215         crc = crc16(crc, (__u8 *)gdp, offset);
3216         offset += sizeof(gdp->bg_checksum); /* skip checksum */
3217         /* for checksum of struct ext4_group_desc do the rest...*/
3218         if (ext4_has_feature_64bit(sb) &&
3219             offset < le16_to_cpu(sbi->s_es->s_desc_size))
3220                 crc = crc16(crc, (__u8 *)gdp + offset,
3221                             le16_to_cpu(sbi->s_es->s_desc_size) -
3222                                 offset);
3223
3224 out:
3225         return cpu_to_le16(crc);
3226 }
3227
3228 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
3229                                 struct ext4_group_desc *gdp)
3230 {
3231         if (ext4_has_group_desc_csum(sb) &&
3232             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
3233                 return 0;
3234
3235         return 1;
3236 }
3237
3238 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
3239                               struct ext4_group_desc *gdp)
3240 {
3241         if (!ext4_has_group_desc_csum(sb))
3242                 return;
3243         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
3244 }
3245
3246 /* Called at mount-time, super-block is locked */
3247 static int ext4_check_descriptors(struct super_block *sb,
3248                                   ext4_fsblk_t sb_block,
3249                                   ext4_group_t *first_not_zeroed)
3250 {
3251         struct ext4_sb_info *sbi = EXT4_SB(sb);
3252         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
3253         ext4_fsblk_t last_block;
3254         ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0);
3255         ext4_fsblk_t block_bitmap;
3256         ext4_fsblk_t inode_bitmap;
3257         ext4_fsblk_t inode_table;
3258         int flexbg_flag = 0;
3259         ext4_group_t i, grp = sbi->s_groups_count;
3260
3261         if (ext4_has_feature_flex_bg(sb))
3262                 flexbg_flag = 1;
3263
3264         ext4_debug("Checking group descriptors");
3265
3266         for (i = 0; i < sbi->s_groups_count; i++) {
3267                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
3268
3269                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
3270                         last_block = ext4_blocks_count(sbi->s_es) - 1;
3271                 else
3272                         last_block = first_block +
3273                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
3274
3275                 if ((grp == sbi->s_groups_count) &&
3276                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3277                         grp = i;
3278
3279                 block_bitmap = ext4_block_bitmap(sb, gdp);
3280                 if (block_bitmap == sb_block) {
3281                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3282                                  "Block bitmap for group %u overlaps "
3283                                  "superblock", i);
3284                         if (!sb_rdonly(sb))
3285                                 return 0;
3286                 }
3287                 if (block_bitmap >= sb_block + 1 &&
3288                     block_bitmap <= last_bg_block) {
3289                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3290                                  "Block bitmap for group %u overlaps "
3291                                  "block group descriptors", i);
3292                         if (!sb_rdonly(sb))
3293                                 return 0;
3294                 }
3295                 if (block_bitmap < first_block || block_bitmap > last_block) {
3296                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3297                                "Block bitmap for group %u not in group "
3298                                "(block %llu)!", i, block_bitmap);
3299                         return 0;
3300                 }
3301                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
3302                 if (inode_bitmap == sb_block) {
3303                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3304                                  "Inode bitmap for group %u overlaps "
3305                                  "superblock", i);
3306                         if (!sb_rdonly(sb))
3307                                 return 0;
3308                 }
3309                 if (inode_bitmap >= sb_block + 1 &&
3310                     inode_bitmap <= last_bg_block) {
3311                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3312                                  "Inode bitmap for group %u overlaps "
3313                                  "block group descriptors", i);
3314                         if (!sb_rdonly(sb))
3315                                 return 0;
3316                 }
3317                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
3318                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3319                                "Inode bitmap for group %u not in group "
3320                                "(block %llu)!", i, inode_bitmap);
3321                         return 0;
3322                 }
3323                 inode_table = ext4_inode_table(sb, gdp);
3324                 if (inode_table == sb_block) {
3325                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3326                                  "Inode table for group %u overlaps "
3327                                  "superblock", i);
3328                         if (!sb_rdonly(sb))
3329                                 return 0;
3330                 }
3331                 if (inode_table >= sb_block + 1 &&
3332                     inode_table <= last_bg_block) {
3333                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3334                                  "Inode table for group %u overlaps "
3335                                  "block group descriptors", i);
3336                         if (!sb_rdonly(sb))
3337                                 return 0;
3338                 }
3339                 if (inode_table < first_block ||
3340                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
3341                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3342                                "Inode table for group %u not in group "
3343                                "(block %llu)!", i, inode_table);
3344                         return 0;
3345                 }
3346                 ext4_lock_group(sb, i);
3347                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
3348                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3349                                  "Checksum for group %u failed (%u!=%u)",
3350                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
3351                                      gdp)), le16_to_cpu(gdp->bg_checksum));
3352                         if (!sb_rdonly(sb)) {
3353                                 ext4_unlock_group(sb, i);
3354                                 return 0;
3355                         }
3356                 }
3357                 ext4_unlock_group(sb, i);
3358                 if (!flexbg_flag)
3359                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
3360         }
3361         if (NULL != first_not_zeroed)
3362                 *first_not_zeroed = grp;
3363         return 1;
3364 }
3365
3366 /*
3367  * Maximal extent format file size.
3368  * Resulting logical blkno at s_maxbytes must fit in our on-disk
3369  * extent format containers, within a sector_t, and within i_blocks
3370  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
3371  * so that won't be a limiting factor.
3372  *
3373  * However there is other limiting factor. We do store extents in the form
3374  * of starting block and length, hence the resulting length of the extent
3375  * covering maximum file size must fit into on-disk format containers as
3376  * well. Given that length is always by 1 unit bigger than max unit (because
3377  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
3378  *
3379  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
3380  */
3381 static loff_t ext4_max_size(int blkbits, int has_huge_files)
3382 {
3383         loff_t res;
3384         loff_t upper_limit = MAX_LFS_FILESIZE;
3385
3386         BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64));
3387
3388         if (!has_huge_files) {
3389                 upper_limit = (1LL << 32) - 1;
3390
3391                 /* total blocks in file system block size */
3392                 upper_limit >>= (blkbits - 9);
3393                 upper_limit <<= blkbits;
3394         }
3395
3396         /*
3397          * 32-bit extent-start container, ee_block. We lower the maxbytes
3398          * by one fs block, so ee_len can cover the extent of maximum file
3399          * size
3400          */
3401         res = (1LL << 32) - 1;
3402         res <<= blkbits;
3403
3404         /* Sanity check against vm- & vfs- imposed limits */
3405         if (res > upper_limit)
3406                 res = upper_limit;
3407
3408         return res;
3409 }
3410
3411 /*
3412  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
3413  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
3414  * We need to be 1 filesystem block less than the 2^48 sector limit.
3415  */
3416 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
3417 {
3418         loff_t upper_limit, res = EXT4_NDIR_BLOCKS;
3419         int meta_blocks;
3420         unsigned int ppb = 1 << (bits - 2);
3421
3422         /*
3423          * This is calculated to be the largest file size for a dense, block
3424          * mapped file such that the file's total number of 512-byte sectors,
3425          * including data and all indirect blocks, does not exceed (2^48 - 1).
3426          *
3427          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
3428          * number of 512-byte sectors of the file.
3429          */
3430         if (!has_huge_files) {
3431                 /*
3432                  * !has_huge_files or implies that the inode i_block field
3433                  * represents total file blocks in 2^32 512-byte sectors ==
3434                  * size of vfs inode i_blocks * 8
3435                  */
3436                 upper_limit = (1LL << 32) - 1;
3437
3438                 /* total blocks in file system block size */
3439                 upper_limit >>= (bits - 9);
3440
3441         } else {
3442                 /*
3443                  * We use 48 bit ext4_inode i_blocks
3444                  * With EXT4_HUGE_FILE_FL set the i_blocks
3445                  * represent total number of blocks in
3446                  * file system block size
3447                  */
3448                 upper_limit = (1LL << 48) - 1;
3449
3450         }
3451
3452         /* Compute how many blocks we can address by block tree */
3453         res += ppb;
3454         res += ppb * ppb;
3455         res += ((loff_t)ppb) * ppb * ppb;
3456         /* Compute how many metadata blocks are needed */
3457         meta_blocks = 1;
3458         meta_blocks += 1 + ppb;
3459         meta_blocks += 1 + ppb + ppb * ppb;
3460         /* Does block tree limit file size? */
3461         if (res + meta_blocks <= upper_limit)
3462                 goto check_lfs;
3463
3464         res = upper_limit;
3465         /* How many metadata blocks are needed for addressing upper_limit? */
3466         upper_limit -= EXT4_NDIR_BLOCKS;
3467         /* indirect blocks */
3468         meta_blocks = 1;
3469         upper_limit -= ppb;
3470         /* double indirect blocks */
3471         if (upper_limit < ppb * ppb) {
3472                 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb);
3473                 res -= meta_blocks;
3474                 goto check_lfs;
3475         }
3476         meta_blocks += 1 + ppb;
3477         upper_limit -= ppb * ppb;
3478         /* tripple indirect blocks for the rest */
3479         meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb) +
3480                 DIV_ROUND_UP_ULL(upper_limit, ppb*ppb);
3481         res -= meta_blocks;
3482 check_lfs:
3483         res <<= bits;
3484         if (res > MAX_LFS_FILESIZE)
3485                 res = MAX_LFS_FILESIZE;
3486
3487         return res;
3488 }
3489
3490 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
3491                                    ext4_fsblk_t logical_sb_block, int nr)
3492 {
3493         struct ext4_sb_info *sbi = EXT4_SB(sb);
3494         ext4_group_t bg, first_meta_bg;
3495         int has_super = 0;
3496
3497         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
3498
3499         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
3500                 return logical_sb_block + nr + 1;
3501         bg = sbi->s_desc_per_block * nr;
3502         if (ext4_bg_has_super(sb, bg))
3503                 has_super = 1;
3504
3505         /*
3506          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
3507          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
3508          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
3509          * compensate.
3510          */
3511         if (sb->s_blocksize == 1024 && nr == 0 &&
3512             le32_to_cpu(sbi->s_es->s_first_data_block) == 0)
3513                 has_super++;
3514
3515         return (has_super + ext4_group_first_block_no(sb, bg));
3516 }
3517
3518 /**
3519  * ext4_get_stripe_size: Get the stripe size.
3520  * @sbi: In memory super block info
3521  *
3522  * If we have specified it via mount option, then
3523  * use the mount option value. If the value specified at mount time is
3524  * greater than the blocks per group use the super block value.
3525  * If the super block value is greater than blocks per group return 0.
3526  * Allocator needs it be less than blocks per group.
3527  *
3528  */
3529 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
3530 {
3531         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
3532         unsigned long stripe_width =
3533                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
3534         int ret;
3535
3536         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
3537                 ret = sbi->s_stripe;
3538         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
3539                 ret = stripe_width;
3540         else if (stride && stride <= sbi->s_blocks_per_group)
3541                 ret = stride;
3542         else
3543                 ret = 0;
3544
3545         /*
3546          * If the stripe width is 1, this makes no sense and
3547          * we set it to 0 to turn off stripe handling code.
3548          */
3549         if (ret <= 1)
3550                 ret = 0;
3551
3552         return ret;
3553 }
3554
3555 /*
3556  * Check whether this filesystem can be mounted based on
3557  * the features present and the RDONLY/RDWR mount requested.
3558  * Returns 1 if this filesystem can be mounted as requested,
3559  * 0 if it cannot be.
3560  */
3561 int ext4_feature_set_ok(struct super_block *sb, int readonly)
3562 {
3563         if (ext4_has_unknown_ext4_incompat_features(sb)) {
3564                 ext4_msg(sb, KERN_ERR,
3565                         "Couldn't mount because of "
3566                         "unsupported optional features (%x)",
3567                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
3568                         ~EXT4_FEATURE_INCOMPAT_SUPP));
3569                 return 0;
3570         }
3571
3572 #if !IS_ENABLED(CONFIG_UNICODE)
3573         if (ext4_has_feature_casefold(sb)) {
3574                 ext4_msg(sb, KERN_ERR,
3575                          "Filesystem with casefold feature cannot be "
3576                          "mounted without CONFIG_UNICODE");
3577                 return 0;
3578         }
3579 #endif
3580
3581         if (readonly)
3582                 return 1;
3583
3584         if (ext4_has_feature_readonly(sb)) {
3585                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
3586                 sb->s_flags |= SB_RDONLY;
3587                 return 1;
3588         }
3589
3590         /* Check that feature set is OK for a read-write mount */
3591         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
3592                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
3593                          "unsupported optional features (%x)",
3594                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
3595                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
3596                 return 0;
3597         }
3598         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
3599                 ext4_msg(sb, KERN_ERR,
3600                          "Can't support bigalloc feature without "
3601                          "extents feature\n");
3602                 return 0;
3603         }
3604
3605 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2)
3606         if (!readonly && (ext4_has_feature_quota(sb) ||
3607                           ext4_has_feature_project(sb))) {
3608                 ext4_msg(sb, KERN_ERR,
3609                          "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2");
3610                 return 0;
3611         }
3612 #endif  /* CONFIG_QUOTA */
3613         return 1;
3614 }
3615
3616 /*
3617  * This function is called once a day if we have errors logged
3618  * on the file system
3619  */
3620 static void print_daily_error_info(struct timer_list *t)
3621 {
3622         struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report);
3623         struct super_block *sb = sbi->s_sb;
3624         struct ext4_super_block *es = sbi->s_es;
3625
3626         if (es->s_error_count)
3627                 /* fsck newer than v1.41.13 is needed to clean this condition. */
3628                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
3629                          le32_to_cpu(es->s_error_count));
3630         if (es->s_first_error_time) {
3631                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d",
3632                        sb->s_id,
3633                        ext4_get_tstamp(es, s_first_error_time),
3634                        (int) sizeof(es->s_first_error_func),
3635                        es->s_first_error_func,
3636                        le32_to_cpu(es->s_first_error_line));
3637                 if (es->s_first_error_ino)
3638                         printk(KERN_CONT ": inode %u",
3639                                le32_to_cpu(es->s_first_error_ino));
3640                 if (es->s_first_error_block)
3641                         printk(KERN_CONT ": block %llu", (unsigned long long)
3642                                le64_to_cpu(es->s_first_error_block));
3643                 printk(KERN_CONT "\n");
3644         }
3645         if (es->s_last_error_time) {
3646                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d",
3647                        sb->s_id,
3648                        ext4_get_tstamp(es, s_last_error_time),
3649                        (int) sizeof(es->s_last_error_func),
3650                        es->s_last_error_func,
3651                        le32_to_cpu(es->s_last_error_line));
3652                 if (es->s_last_error_ino)
3653                         printk(KERN_CONT ": inode %u",
3654                                le32_to_cpu(es->s_last_error_ino));
3655                 if (es->s_last_error_block)
3656                         printk(KERN_CONT ": block %llu", (unsigned long long)
3657                                le64_to_cpu(es->s_last_error_block));
3658                 printk(KERN_CONT "\n");
3659         }
3660         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
3661 }
3662
3663 /* Find next suitable group and run ext4_init_inode_table */
3664 static int ext4_run_li_request(struct ext4_li_request *elr)
3665 {
3666         struct ext4_group_desc *gdp = NULL;
3667         struct super_block *sb = elr->lr_super;
3668         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3669         ext4_group_t group = elr->lr_next_group;
3670         unsigned int prefetch_ios = 0;
3671         int ret = 0;
3672         u64 start_time;
3673
3674         if (elr->lr_mode == EXT4_LI_MODE_PREFETCH_BBITMAP) {
3675                 elr->lr_next_group = ext4_mb_prefetch(sb, group,
3676                                 EXT4_SB(sb)->s_mb_prefetch, &prefetch_ios);
3677                 if (prefetch_ios)
3678                         ext4_mb_prefetch_fini(sb, elr->lr_next_group,
3679                                               prefetch_ios);
3680                 trace_ext4_prefetch_bitmaps(sb, group, elr->lr_next_group,
3681                                             prefetch_ios);
3682                 if (group >= elr->lr_next_group) {
3683                         ret = 1;
3684                         if (elr->lr_first_not_zeroed != ngroups &&
3685                             !sb_rdonly(sb) && test_opt(sb, INIT_INODE_TABLE)) {
3686                                 elr->lr_next_group = elr->lr_first_not_zeroed;
3687                                 elr->lr_mode = EXT4_LI_MODE_ITABLE;
3688                                 ret = 0;
3689                         }
3690                 }
3691                 return ret;
3692         }
3693
3694         for (; group < ngroups; group++) {
3695                 gdp = ext4_get_group_desc(sb, group, NULL);
3696                 if (!gdp) {
3697                         ret = 1;
3698                         break;
3699                 }
3700
3701                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3702                         break;
3703         }
3704
3705         if (group >= ngroups)
3706                 ret = 1;
3707
3708         if (!ret) {
3709                 start_time = ktime_get_real_ns();
3710                 ret = ext4_init_inode_table(sb, group,
3711                                             elr->lr_timeout ? 0 : 1);
3712                 trace_ext4_lazy_itable_init(sb, group);
3713                 if (elr->lr_timeout == 0) {
3714                         elr->lr_timeout = nsecs_to_jiffies((ktime_get_real_ns() - start_time) *
3715                                 EXT4_SB(elr->lr_super)->s_li_wait_mult);
3716                 }
3717                 elr->lr_next_sched = jiffies + elr->lr_timeout;
3718                 elr->lr_next_group = group + 1;
3719         }
3720         return ret;
3721 }
3722
3723 /*
3724  * Remove lr_request from the list_request and free the
3725  * request structure. Should be called with li_list_mtx held
3726  */
3727 static void ext4_remove_li_request(struct ext4_li_request *elr)
3728 {
3729         if (!elr)
3730                 return;
3731
3732         list_del(&elr->lr_request);
3733         EXT4_SB(elr->lr_super)->s_li_request = NULL;
3734         kfree(elr);
3735 }
3736
3737 static void ext4_unregister_li_request(struct super_block *sb)
3738 {
3739         mutex_lock(&ext4_li_mtx);
3740         if (!ext4_li_info) {
3741                 mutex_unlock(&ext4_li_mtx);
3742                 return;
3743         }
3744
3745         mutex_lock(&ext4_li_info->li_list_mtx);
3746         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
3747         mutex_unlock(&ext4_li_info->li_list_mtx);
3748         mutex_unlock(&ext4_li_mtx);
3749 }
3750
3751 static struct task_struct *ext4_lazyinit_task;
3752
3753 /*
3754  * This is the function where ext4lazyinit thread lives. It walks
3755  * through the request list searching for next scheduled filesystem.
3756  * When such a fs is found, run the lazy initialization request
3757  * (ext4_rn_li_request) and keep track of the time spend in this
3758  * function. Based on that time we compute next schedule time of
3759  * the request. When walking through the list is complete, compute
3760  * next waking time and put itself into sleep.
3761  */
3762 static int ext4_lazyinit_thread(void *arg)
3763 {
3764         struct ext4_lazy_init *eli = arg;
3765         struct list_head *pos, *n;
3766         struct ext4_li_request *elr;
3767         unsigned long next_wakeup, cur;
3768
3769         BUG_ON(NULL == eli);
3770
3771 cont_thread:
3772         while (true) {
3773                 next_wakeup = MAX_JIFFY_OFFSET;
3774
3775                 mutex_lock(&eli->li_list_mtx);
3776                 if (list_empty(&eli->li_request_list)) {
3777                         mutex_unlock(&eli->li_list_mtx);
3778                         goto exit_thread;
3779                 }
3780                 list_for_each_safe(pos, n, &eli->li_request_list) {
3781                         int err = 0;
3782                         int progress = 0;
3783                         elr = list_entry(pos, struct ext4_li_request,
3784                                          lr_request);
3785
3786                         if (time_before(jiffies, elr->lr_next_sched)) {
3787                                 if (time_before(elr->lr_next_sched, next_wakeup))
3788                                         next_wakeup = elr->lr_next_sched;
3789                                 continue;
3790                         }
3791                         if (down_read_trylock(&elr->lr_super->s_umount)) {
3792                                 if (sb_start_write_trylock(elr->lr_super)) {
3793                                         progress = 1;
3794                                         /*
3795                                          * We hold sb->s_umount, sb can not
3796                                          * be removed from the list, it is
3797                                          * now safe to drop li_list_mtx
3798                                          */
3799                                         mutex_unlock(&eli->li_list_mtx);
3800                                         err = ext4_run_li_request(elr);
3801                                         sb_end_write(elr->lr_super);
3802                                         mutex_lock(&eli->li_list_mtx);
3803                                         n = pos->next;
3804                                 }
3805                                 up_read((&elr->lr_super->s_umount));
3806                         }
3807                         /* error, remove the lazy_init job */
3808                         if (err) {
3809                                 ext4_remove_li_request(elr);
3810                                 continue;
3811                         }
3812                         if (!progress) {
3813                                 elr->lr_next_sched = jiffies +
3814                                         (prandom_u32()
3815                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3816                         }
3817                         if (time_before(elr->lr_next_sched, next_wakeup))
3818                                 next_wakeup = elr->lr_next_sched;
3819                 }
3820                 mutex_unlock(&eli->li_list_mtx);
3821
3822                 try_to_freeze();
3823
3824                 cur = jiffies;
3825                 if ((time_after_eq(cur, next_wakeup)) ||
3826                     (MAX_JIFFY_OFFSET == next_wakeup)) {
3827                         cond_resched();
3828                         continue;
3829                 }
3830
3831                 schedule_timeout_interruptible(next_wakeup - cur);
3832
3833                 if (kthread_should_stop()) {
3834                         ext4_clear_request_list();
3835                         goto exit_thread;
3836                 }
3837         }
3838
3839 exit_thread:
3840         /*
3841          * It looks like the request list is empty, but we need
3842          * to check it under the li_list_mtx lock, to prevent any
3843          * additions into it, and of course we should lock ext4_li_mtx
3844          * to atomically free the list and ext4_li_info, because at
3845          * this point another ext4 filesystem could be registering
3846          * new one.
3847          */
3848         mutex_lock(&ext4_li_mtx);
3849         mutex_lock(&eli->li_list_mtx);
3850         if (!list_empty(&eli->li_request_list)) {
3851                 mutex_unlock(&eli->li_list_mtx);
3852                 mutex_unlock(&ext4_li_mtx);
3853                 goto cont_thread;
3854         }
3855         mutex_unlock(&eli->li_list_mtx);
3856         kfree(ext4_li_info);
3857         ext4_li_info = NULL;
3858         mutex_unlock(&ext4_li_mtx);
3859
3860         return 0;
3861 }
3862
3863 static void ext4_clear_request_list(void)
3864 {
3865         struct list_head *pos, *n;
3866         struct ext4_li_request *elr;
3867
3868         mutex_lock(&ext4_li_info->li_list_mtx);
3869         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3870                 elr = list_entry(pos, struct ext4_li_request,
3871                                  lr_request);
3872                 ext4_remove_li_request(elr);
3873         }
3874         mutex_unlock(&ext4_li_info->li_list_mtx);
3875 }
3876
3877 static int ext4_run_lazyinit_thread(void)
3878 {
3879         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3880                                          ext4_li_info, "ext4lazyinit");
3881         if (IS_ERR(ext4_lazyinit_task)) {
3882                 int err = PTR_ERR(ext4_lazyinit_task);
3883                 ext4_clear_request_list();
3884                 kfree(ext4_li_info);
3885                 ext4_li_info = NULL;
3886                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3887                                  "initialization thread\n",
3888                                  err);
3889                 return err;
3890         }
3891         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3892         return 0;
3893 }
3894
3895 /*
3896  * Check whether it make sense to run itable init. thread or not.
3897  * If there is at least one uninitialized inode table, return
3898  * corresponding group number, else the loop goes through all
3899  * groups and return total number of groups.
3900  */
3901 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3902 {
3903         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3904         struct ext4_group_desc *gdp = NULL;
3905
3906         if (!ext4_has_group_desc_csum(sb))
3907                 return ngroups;
3908
3909         for (group = 0; group < ngroups; group++) {
3910                 gdp = ext4_get_group_desc(sb, group, NULL);
3911                 if (!gdp)
3912                         continue;
3913
3914                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3915                         break;
3916         }
3917
3918         return group;
3919 }
3920
3921 static int ext4_li_info_new(void)
3922 {
3923         struct ext4_lazy_init *eli = NULL;
3924
3925         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3926         if (!eli)
3927                 return -ENOMEM;
3928
3929         INIT_LIST_HEAD(&eli->li_request_list);
3930         mutex_init(&eli->li_list_mtx);
3931
3932         eli->li_state |= EXT4_LAZYINIT_QUIT;
3933
3934         ext4_li_info = eli;
3935
3936         return 0;
3937 }
3938
3939 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3940                                             ext4_group_t start)
3941 {
3942         struct ext4_li_request *elr;
3943
3944         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3945         if (!elr)
3946                 return NULL;
3947
3948         elr->lr_super = sb;
3949         elr->lr_first_not_zeroed = start;
3950         if (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS)) {
3951                 elr->lr_mode = EXT4_LI_MODE_ITABLE;
3952                 elr->lr_next_group = start;
3953         } else {
3954                 elr->lr_mode = EXT4_LI_MODE_PREFETCH_BBITMAP;
3955         }
3956
3957         /*
3958          * Randomize first schedule time of the request to
3959          * spread the inode table initialization requests
3960          * better.
3961          */
3962         elr->lr_next_sched = jiffies + (prandom_u32() %
3963                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3964         return elr;
3965 }
3966
3967 int ext4_register_li_request(struct super_block *sb,
3968                              ext4_group_t first_not_zeroed)
3969 {
3970         struct ext4_sb_info *sbi = EXT4_SB(sb);
3971         struct ext4_li_request *elr = NULL;
3972         ext4_group_t ngroups = sbi->s_groups_count;
3973         int ret = 0;
3974
3975         mutex_lock(&ext4_li_mtx);
3976         if (sbi->s_li_request != NULL) {
3977                 /*
3978                  * Reset timeout so it can be computed again, because
3979                  * s_li_wait_mult might have changed.
3980                  */
3981                 sbi->s_li_request->lr_timeout = 0;
3982                 goto out;
3983         }
3984
3985         if (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS) &&
3986             (first_not_zeroed == ngroups || sb_rdonly(sb) ||
3987              !test_opt(sb, INIT_INODE_TABLE)))
3988                 goto out;
3989
3990         elr = ext4_li_request_new(sb, first_not_zeroed);
3991         if (!elr) {
3992                 ret = -ENOMEM;
3993                 goto out;
3994         }
3995
3996         if (NULL == ext4_li_info) {
3997                 ret = ext4_li_info_new();
3998                 if (ret)
3999                         goto out;
4000         }
4001
4002         mutex_lock(&ext4_li_info->li_list_mtx);
4003         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
4004         mutex_unlock(&ext4_li_info->li_list_mtx);
4005
4006         sbi->s_li_request = elr;
4007         /*
4008          * set elr to NULL here since it has been inserted to
4009          * the request_list and the removal and free of it is
4010          * handled by ext4_clear_request_list from now on.
4011          */
4012         elr = NULL;
4013
4014         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
4015                 ret = ext4_run_lazyinit_thread();
4016                 if (ret)
4017                         goto out;
4018         }
4019 out:
4020         mutex_unlock(&ext4_li_mtx);
4021         if (ret)
4022                 kfree(elr);
4023         return ret;
4024 }
4025
4026 /*
4027  * We do not need to lock anything since this is called on
4028  * module unload.
4029  */
4030 static void ext4_destroy_lazyinit_thread(void)
4031 {
4032         /*
4033          * If thread exited earlier
4034          * there's nothing to be done.
4035          */
4036         if (!ext4_li_info || !ext4_lazyinit_task)
4037                 return;
4038
4039         kthread_stop(ext4_lazyinit_task);
4040 }
4041
4042 static int set_journal_csum_feature_set(struct super_block *sb)
4043 {
4044         int ret = 1;
4045         int compat, incompat;
4046         struct ext4_sb_info *sbi = EXT4_SB(sb);
4047
4048         if (ext4_has_metadata_csum(sb)) {
4049                 /* journal checksum v3 */
4050                 compat = 0;
4051                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
4052         } else {
4053                 /* journal checksum v1 */
4054                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
4055                 incompat = 0;
4056         }
4057
4058         jbd2_journal_clear_features(sbi->s_journal,
4059                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
4060                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
4061                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
4062         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4063                 ret = jbd2_journal_set_features(sbi->s_journal,
4064                                 compat, 0,
4065                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
4066                                 incompat);
4067         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
4068                 ret = jbd2_journal_set_features(sbi->s_journal,
4069                                 compat, 0,
4070                                 incompat);
4071                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
4072                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
4073         } else {
4074                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
4075                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
4076         }
4077
4078         return ret;
4079 }
4080
4081 /*
4082  * Note: calculating the overhead so we can be compatible with
4083  * historical BSD practice is quite difficult in the face of
4084  * clusters/bigalloc.  This is because multiple metadata blocks from
4085  * different block group can end up in the same allocation cluster.
4086  * Calculating the exact overhead in the face of clustered allocation
4087  * requires either O(all block bitmaps) in memory or O(number of block
4088  * groups**2) in time.  We will still calculate the superblock for
4089  * older file systems --- and if we come across with a bigalloc file
4090  * system with zero in s_overhead_clusters the estimate will be close to
4091  * correct especially for very large cluster sizes --- but for newer
4092  * file systems, it's better to calculate this figure once at mkfs
4093  * time, and store it in the superblock.  If the superblock value is
4094  * present (even for non-bigalloc file systems), we will use it.
4095  */
4096 static int count_overhead(struct super_block *sb, ext4_group_t grp,
4097                           char *buf)
4098 {
4099         struct ext4_sb_info     *sbi = EXT4_SB(sb);
4100         struct ext4_group_desc  *gdp;
4101         ext4_fsblk_t            first_block, last_block, b;
4102         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
4103         int                     s, j, count = 0;
4104         int                     has_super = ext4_bg_has_super(sb, grp);
4105
4106         if (!ext4_has_feature_bigalloc(sb))
4107                 return (has_super + ext4_bg_num_gdb(sb, grp) +
4108                         (has_super ? le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0) +
4109                         sbi->s_itb_per_group + 2);
4110
4111         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
4112                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
4113         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
4114         for (i = 0; i < ngroups; i++) {
4115                 gdp = ext4_get_group_desc(sb, i, NULL);
4116                 b = ext4_block_bitmap(sb, gdp);
4117                 if (b >= first_block && b <= last_block) {
4118                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
4119                         count++;
4120                 }
4121                 b = ext4_inode_bitmap(sb, gdp);
4122                 if (b >= first_block && b <= last_block) {
4123                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
4124                         count++;
4125                 }
4126                 b = ext4_inode_table(sb, gdp);
4127                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
4128                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
4129                                 int c = EXT4_B2C(sbi, b - first_block);
4130                                 ext4_set_bit(c, buf);
4131                                 count++;
4132                         }
4133                 if (i != grp)
4134                         continue;
4135                 s = 0;
4136                 if (ext4_bg_has_super(sb, grp)) {
4137                         ext4_set_bit(s++, buf);
4138                         count++;
4139                 }
4140                 j = ext4_bg_num_gdb(sb, grp);
4141                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
4142                         ext4_error(sb, "Invalid number of block group "
4143                                    "descriptor blocks: %d", j);
4144                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
4145                 }
4146                 count += j;
4147                 for (; j > 0; j--)
4148                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
4149         }
4150         if (!count)
4151                 return 0;
4152         return EXT4_CLUSTERS_PER_GROUP(sb) -
4153                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
4154 }
4155
4156 /*
4157  * Compute the overhead and stash it in sbi->s_overhead
4158  */
4159 int ext4_calculate_overhead(struct super_block *sb)
4160 {
4161         struct ext4_sb_info *sbi = EXT4_SB(sb);
4162         struct ext4_super_block *es = sbi->s_es;
4163         struct inode *j_inode;
4164         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
4165         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4166         ext4_fsblk_t overhead = 0;
4167         char *buf = (char *) get_zeroed_page(GFP_NOFS);
4168
4169         if (!buf)
4170                 return -ENOMEM;
4171
4172         /*
4173          * Compute the overhead (FS structures).  This is constant
4174          * for a given filesystem unless the number of block groups
4175          * changes so we cache the previous value until it does.
4176          */
4177
4178         /*
4179          * All of the blocks before first_data_block are overhead
4180          */
4181         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
4182
4183         /*
4184          * Add the overhead found in each block group
4185          */
4186         for (i = 0; i < ngroups; i++) {
4187                 int blks;
4188
4189                 blks = count_overhead(sb, i, buf);
4190                 overhead += blks;
4191                 if (blks)
4192                         memset(buf, 0, PAGE_SIZE);
4193                 cond_resched();
4194         }
4195
4196         /*
4197          * Add the internal journal blocks whether the journal has been
4198          * loaded or not
4199          */
4200         if (sbi->s_journal && !sbi->s_journal_bdev)
4201                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_total_len);
4202         else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) {
4203                 /* j_inum for internal journal is non-zero */
4204                 j_inode = ext4_get_journal_inode(sb, j_inum);
4205                 if (j_inode) {
4206                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
4207                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
4208                         iput(j_inode);
4209                 } else {
4210                         ext4_msg(sb, KERN_ERR, "can't get journal size");
4211                 }
4212         }
4213         sbi->s_overhead = overhead;
4214         smp_wmb();
4215         free_page((unsigned long) buf);
4216         return 0;
4217 }
4218
4219 static void ext4_set_resv_clusters(struct super_block *sb)
4220 {
4221         ext4_fsblk_t resv_clusters;
4222         struct ext4_sb_info *sbi = EXT4_SB(sb);
4223
4224         /*
4225          * There's no need to reserve anything when we aren't using extents.
4226          * The space estimates are exact, there are no unwritten extents,
4227          * hole punching doesn't need new metadata... This is needed especially
4228          * to keep ext2/3 backward compatibility.
4229          */
4230         if (!ext4_has_feature_extents(sb))
4231                 return;
4232         /*
4233          * By default we reserve 2% or 4096 clusters, whichever is smaller.
4234          * This should cover the situations where we can not afford to run
4235          * out of space like for example punch hole, or converting
4236          * unwritten extents in delalloc path. In most cases such
4237          * allocation would require 1, or 2 blocks, higher numbers are
4238          * very rare.
4239          */
4240         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
4241                          sbi->s_cluster_bits);
4242
4243         do_div(resv_clusters, 50);
4244         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
4245
4246         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
4247 }
4248
4249 static const char *ext4_quota_mode(struct super_block *sb)
4250 {
4251 #ifdef CONFIG_QUOTA
4252         if (!ext4_quota_capable(sb))
4253                 return "none";
4254
4255         if (EXT4_SB(sb)->s_journal && ext4_is_quota_journalled(sb))
4256                 return "journalled";
4257         else
4258                 return "writeback";
4259 #else
4260         return "disabled";
4261 #endif
4262 }
4263
4264 static void ext4_setup_csum_trigger(struct super_block *sb,
4265                                     enum ext4_journal_trigger_type type,
4266                                     void (*trigger)(
4267                                         struct jbd2_buffer_trigger_type *type,
4268                                         struct buffer_head *bh,
4269                                         void *mapped_data,
4270                                         size_t size))
4271 {
4272         struct ext4_sb_info *sbi = EXT4_SB(sb);
4273
4274         sbi->s_journal_triggers[type].sb = sb;
4275         sbi->s_journal_triggers[type].tr_triggers.t_frozen = trigger;
4276 }
4277
4278 static void ext4_free_sbi(struct ext4_sb_info *sbi)
4279 {
4280         if (!sbi)
4281                 return;
4282
4283         kfree(sbi->s_blockgroup_lock);
4284         fs_put_dax(sbi->s_daxdev);
4285         kfree(sbi);
4286 }
4287
4288 static struct ext4_sb_info *ext4_alloc_sbi(struct super_block *sb)
4289 {
4290         struct ext4_sb_info *sbi;
4291
4292         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
4293         if (!sbi)
4294                 return NULL;
4295
4296         sbi->s_daxdev = fs_dax_get_by_bdev(sb->s_bdev, &sbi->s_dax_part_off);
4297
4298         sbi->s_blockgroup_lock =
4299                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
4300
4301         if (!sbi->s_blockgroup_lock)
4302                 goto err_out;
4303
4304         sb->s_fs_info = sbi;
4305         sbi->s_sb = sb;
4306         return sbi;
4307 err_out:
4308         fs_put_dax(sbi->s_daxdev);
4309         kfree(sbi);
4310         return NULL;
4311 }
4312
4313 static int __ext4_fill_super(struct fs_context *fc, struct super_block *sb)
4314 {
4315         struct buffer_head *bh, **group_desc;
4316         struct ext4_super_block *es = NULL;
4317         struct ext4_sb_info *sbi = EXT4_SB(sb);
4318         struct flex_groups **flex_groups;
4319         ext4_fsblk_t block;
4320         ext4_fsblk_t logical_sb_block;
4321         unsigned long offset = 0;
4322         unsigned long def_mount_opts;
4323         struct inode *root;
4324         int ret = -ENOMEM;
4325         int blocksize, clustersize;
4326         unsigned int db_count;
4327         unsigned int i;
4328         int needs_recovery, has_huge_files;
4329         __u64 blocks_count;
4330         int err = 0;
4331         ext4_group_t first_not_zeroed;
4332         struct ext4_fs_context *ctx = fc->fs_private;
4333         int silent = fc->sb_flags & SB_SILENT;
4334
4335         /* Set defaults for the variables that will be set during parsing */
4336         if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO))
4337                 ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4338
4339         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
4340         sbi->s_sectors_written_start =
4341                 part_stat_read(sb->s_bdev, sectors[STAT_WRITE]);
4342
4343         /* -EINVAL is default */
4344         ret = -EINVAL;
4345         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
4346         if (!blocksize) {
4347                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
4348                 goto out_fail;
4349         }
4350
4351         /*
4352          * The ext4 superblock will not be buffer aligned for other than 1kB
4353          * block sizes.  We need to calculate the offset from buffer start.
4354          */
4355         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
4356                 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE;
4357                 offset = do_div(logical_sb_block, blocksize);
4358         } else {
4359                 logical_sb_block = sbi->s_sb_block;
4360         }
4361
4362         bh = ext4_sb_bread_unmovable(sb, logical_sb_block);
4363         if (IS_ERR(bh)) {
4364                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
4365                 ret = PTR_ERR(bh);
4366                 goto out_fail;
4367         }
4368         /*
4369          * Note: s_es must be initialized as soon as possible because
4370          *       some ext4 macro-instructions depend on its value
4371          */
4372         es = (struct ext4_super_block *) (bh->b_data + offset);
4373         sbi->s_es = es;
4374         sb->s_magic = le16_to_cpu(es->s_magic);
4375         if (sb->s_magic != EXT4_SUPER_MAGIC)
4376                 goto cantfind_ext4;
4377         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
4378
4379         /* Warn if metadata_csum and gdt_csum are both set. */
4380         if (ext4_has_feature_metadata_csum(sb) &&
4381             ext4_has_feature_gdt_csum(sb))
4382                 ext4_warning(sb, "metadata_csum and uninit_bg are "
4383                              "redundant flags; please run fsck.");
4384
4385         /* Check for a known checksum algorithm */
4386         if (!ext4_verify_csum_type(sb, es)) {
4387                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
4388                          "unknown checksum algorithm.");
4389                 silent = 1;
4390                 goto cantfind_ext4;
4391         }
4392         ext4_setup_csum_trigger(sb, EXT4_JTR_ORPHAN_FILE,
4393                                 ext4_orphan_file_block_trigger);
4394
4395         /* Load the checksum driver */
4396         sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
4397         if (IS_ERR(sbi->s_chksum_driver)) {
4398                 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
4399                 ret = PTR_ERR(sbi->s_chksum_driver);
4400                 sbi->s_chksum_driver = NULL;
4401                 goto failed_mount;
4402         }
4403
4404         /* Check superblock checksum */
4405         if (!ext4_superblock_csum_verify(sb, es)) {
4406                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
4407                          "invalid superblock checksum.  Run e2fsck?");
4408                 silent = 1;
4409                 ret = -EFSBADCRC;
4410                 goto cantfind_ext4;
4411         }
4412
4413         /* Precompute checksum seed for all metadata */
4414         if (ext4_has_feature_csum_seed(sb))
4415                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
4416         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
4417                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
4418                                                sizeof(es->s_uuid));
4419
4420         /* Set defaults before we parse the mount options */
4421         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
4422         set_opt(sb, INIT_INODE_TABLE);
4423         if (def_mount_opts & EXT4_DEFM_DEBUG)
4424                 set_opt(sb, DEBUG);
4425         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
4426                 set_opt(sb, GRPID);
4427         if (def_mount_opts & EXT4_DEFM_UID16)
4428                 set_opt(sb, NO_UID32);
4429         /* xattr user namespace & acls are now defaulted on */
4430         set_opt(sb, XATTR_USER);
4431 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4432         set_opt(sb, POSIX_ACL);
4433 #endif
4434         if (ext4_has_feature_fast_commit(sb))
4435                 set_opt2(sb, JOURNAL_FAST_COMMIT);
4436         /* don't forget to enable journal_csum when metadata_csum is enabled. */
4437         if (ext4_has_metadata_csum(sb))
4438                 set_opt(sb, JOURNAL_CHECKSUM);
4439
4440         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
4441                 set_opt(sb, JOURNAL_DATA);
4442         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
4443                 set_opt(sb, ORDERED_DATA);
4444         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
4445                 set_opt(sb, WRITEBACK_DATA);
4446
4447         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
4448                 set_opt(sb, ERRORS_PANIC);
4449         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
4450                 set_opt(sb, ERRORS_CONT);
4451         else
4452                 set_opt(sb, ERRORS_RO);
4453         /* block_validity enabled by default; disable with noblock_validity */
4454         set_opt(sb, BLOCK_VALIDITY);
4455         if (def_mount_opts & EXT4_DEFM_DISCARD)
4456                 set_opt(sb, DISCARD);
4457
4458         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
4459         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
4460         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
4461         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
4462         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
4463
4464         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
4465                 set_opt(sb, BARRIER);
4466
4467         /*
4468          * enable delayed allocation by default
4469          * Use -o nodelalloc to turn it off
4470          */
4471         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
4472             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
4473                 set_opt(sb, DELALLOC);
4474
4475         /*
4476          * set default s_li_wait_mult for lazyinit, for the case there is
4477          * no mount option specified.
4478          */
4479         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
4480
4481         if (le32_to_cpu(es->s_log_block_size) >
4482             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4483                 ext4_msg(sb, KERN_ERR,
4484                          "Invalid log block size: %u",
4485                          le32_to_cpu(es->s_log_block_size));
4486                 goto failed_mount;
4487         }
4488         if (le32_to_cpu(es->s_log_cluster_size) >
4489             (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4490                 ext4_msg(sb, KERN_ERR,
4491                          "Invalid log cluster size: %u",
4492                          le32_to_cpu(es->s_log_cluster_size));
4493                 goto failed_mount;
4494         }
4495
4496         blocksize = EXT4_MIN_BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
4497
4498         if (blocksize == PAGE_SIZE)
4499                 set_opt(sb, DIOREAD_NOLOCK);
4500
4501         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
4502                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
4503                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
4504         } else {
4505                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
4506                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
4507                 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) {
4508                         ext4_msg(sb, KERN_ERR, "invalid first ino: %u",
4509                                  sbi->s_first_ino);
4510                         goto failed_mount;
4511                 }
4512                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
4513                     (!is_power_of_2(sbi->s_inode_size)) ||
4514                     (sbi->s_inode_size > blocksize)) {
4515                         ext4_msg(sb, KERN_ERR,
4516                                "unsupported inode size: %d",
4517                                sbi->s_inode_size);
4518                         ext4_msg(sb, KERN_ERR, "blocksize: %d", blocksize);
4519                         goto failed_mount;
4520                 }
4521                 /*
4522                  * i_atime_extra is the last extra field available for
4523                  * [acm]times in struct ext4_inode. Checking for that
4524                  * field should suffice to ensure we have extra space
4525                  * for all three.
4526                  */
4527                 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) +
4528                         sizeof(((struct ext4_inode *)0)->i_atime_extra)) {
4529                         sb->s_time_gran = 1;
4530                         sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX;
4531                 } else {
4532                         sb->s_time_gran = NSEC_PER_SEC;
4533                         sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX;
4534                 }
4535                 sb->s_time_min = EXT4_TIMESTAMP_MIN;
4536         }
4537         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
4538                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4539                         EXT4_GOOD_OLD_INODE_SIZE;
4540                 if (ext4_has_feature_extra_isize(sb)) {
4541                         unsigned v, max = (sbi->s_inode_size -
4542                                            EXT4_GOOD_OLD_INODE_SIZE);
4543
4544                         v = le16_to_cpu(es->s_want_extra_isize);
4545                         if (v > max) {
4546                                 ext4_msg(sb, KERN_ERR,
4547                                          "bad s_want_extra_isize: %d", v);
4548                                 goto failed_mount;
4549                         }
4550                         if (sbi->s_want_extra_isize < v)
4551                                 sbi->s_want_extra_isize = v;
4552
4553                         v = le16_to_cpu(es->s_min_extra_isize);
4554                         if (v > max) {
4555                                 ext4_msg(sb, KERN_ERR,
4556                                          "bad s_min_extra_isize: %d", v);
4557                                 goto failed_mount;
4558                         }
4559                         if (sbi->s_want_extra_isize < v)
4560                                 sbi->s_want_extra_isize = v;
4561                 }
4562         }
4563
4564         err = parse_apply_sb_mount_options(sb, ctx);
4565         if (err < 0)
4566                 goto failed_mount;
4567
4568         sbi->s_def_mount_opt = sbi->s_mount_opt;
4569
4570         err = ext4_check_opt_consistency(fc, sb);
4571         if (err < 0)
4572                 goto failed_mount;
4573
4574         ext4_apply_options(fc, sb);
4575
4576 #if IS_ENABLED(CONFIG_UNICODE)
4577         if (ext4_has_feature_casefold(sb) && !sb->s_encoding) {
4578                 const struct ext4_sb_encodings *encoding_info;
4579                 struct unicode_map *encoding;
4580                 __u16 encoding_flags = le16_to_cpu(es->s_encoding_flags);
4581
4582                 encoding_info = ext4_sb_read_encoding(es);
4583                 if (!encoding_info) {
4584                         ext4_msg(sb, KERN_ERR,
4585                                  "Encoding requested by superblock is unknown");
4586                         goto failed_mount;
4587                 }
4588
4589                 encoding = utf8_load(encoding_info->version);
4590                 if (IS_ERR(encoding)) {
4591                         ext4_msg(sb, KERN_ERR,
4592                                  "can't mount with superblock charset: %s-%u.%u.%u "
4593                                  "not supported by the kernel. flags: 0x%x.",
4594                                  encoding_info->name,
4595                                  unicode_major(encoding_info->version),
4596                                  unicode_minor(encoding_info->version),
4597                                  unicode_rev(encoding_info->version),
4598                                  encoding_flags);
4599                         goto failed_mount;
4600                 }
4601                 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: "
4602                          "%s-%u.%u.%u with flags 0x%hx", encoding_info->name,
4603                          unicode_major(encoding_info->version),
4604                          unicode_minor(encoding_info->version),
4605                          unicode_rev(encoding_info->version),
4606                          encoding_flags);
4607
4608                 sb->s_encoding = encoding;
4609                 sb->s_encoding_flags = encoding_flags;
4610         }
4611 #endif
4612
4613         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4614                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with data=journal disables delayed allocation, dioread_nolock, O_DIRECT and fast_commit support!\n");
4615                 /* can't mount with both data=journal and dioread_nolock. */
4616                 clear_opt(sb, DIOREAD_NOLOCK);
4617                 clear_opt2(sb, JOURNAL_FAST_COMMIT);
4618                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4619                         ext4_msg(sb, KERN_ERR, "can't mount with "
4620                                  "both data=journal and delalloc");
4621                         goto failed_mount;
4622                 }
4623                 if (test_opt(sb, DAX_ALWAYS)) {
4624                         ext4_msg(sb, KERN_ERR, "can't mount with "
4625                                  "both data=journal and dax");
4626                         goto failed_mount;
4627                 }
4628                 if (ext4_has_feature_encrypt(sb)) {
4629                         ext4_msg(sb, KERN_WARNING,
4630                                  "encrypted files will use data=ordered "
4631                                  "instead of data journaling mode");
4632                 }
4633                 if (test_opt(sb, DELALLOC))
4634                         clear_opt(sb, DELALLOC);
4635         } else {
4636                 sb->s_iflags |= SB_I_CGROUPWB;
4637         }
4638
4639         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
4640                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
4641
4642         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
4643             (ext4_has_compat_features(sb) ||
4644              ext4_has_ro_compat_features(sb) ||
4645              ext4_has_incompat_features(sb)))
4646                 ext4_msg(sb, KERN_WARNING,
4647                        "feature flags set on rev 0 fs, "
4648                        "running e2fsck is recommended");
4649
4650         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
4651                 set_opt2(sb, HURD_COMPAT);
4652                 if (ext4_has_feature_64bit(sb)) {
4653                         ext4_msg(sb, KERN_ERR,
4654                                  "The Hurd can't support 64-bit file systems");
4655                         goto failed_mount;
4656                 }
4657
4658                 /*
4659                  * ea_inode feature uses l_i_version field which is not
4660                  * available in HURD_COMPAT mode.
4661                  */
4662                 if (ext4_has_feature_ea_inode(sb)) {
4663                         ext4_msg(sb, KERN_ERR,
4664                                  "ea_inode feature is not supported for Hurd");
4665                         goto failed_mount;
4666                 }
4667         }
4668
4669         if (IS_EXT2_SB(sb)) {
4670                 if (ext2_feature_set_ok(sb))
4671                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
4672                                  "using the ext4 subsystem");
4673                 else {
4674                         /*
4675                          * If we're probing be silent, if this looks like
4676                          * it's actually an ext[34] filesystem.
4677                          */
4678                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4679                                 goto failed_mount;
4680                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
4681                                  "to feature incompatibilities");
4682                         goto failed_mount;
4683                 }
4684         }
4685
4686         if (IS_EXT3_SB(sb)) {
4687                 if (ext3_feature_set_ok(sb))
4688                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
4689                                  "using the ext4 subsystem");
4690                 else {
4691                         /*
4692                          * If we're probing be silent, if this looks like
4693                          * it's actually an ext4 filesystem.
4694                          */
4695                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4696                                 goto failed_mount;
4697                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
4698                                  "to feature incompatibilities");
4699                         goto failed_mount;
4700                 }
4701         }
4702
4703         /*
4704          * Check feature flags regardless of the revision level, since we
4705          * previously didn't change the revision level when setting the flags,
4706          * so there is a chance incompat flags are set on a rev 0 filesystem.
4707          */
4708         if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
4709                 goto failed_mount;
4710
4711         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
4712                 ext4_msg(sb, KERN_ERR,
4713                          "Number of reserved GDT blocks insanely large: %d",
4714                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
4715                 goto failed_mount;
4716         }
4717
4718         if (sbi->s_daxdev) {
4719                 if (blocksize == PAGE_SIZE)
4720                         set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags);
4721                 else
4722                         ext4_msg(sb, KERN_ERR, "unsupported blocksize for DAX\n");
4723         }
4724
4725         if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) {
4726                 if (ext4_has_feature_inline_data(sb)) {
4727                         ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
4728                                         " that may contain inline data");
4729                         goto failed_mount;
4730                 }
4731                 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) {
4732                         ext4_msg(sb, KERN_ERR,
4733                                 "DAX unsupported by block device.");
4734                         goto failed_mount;
4735                 }
4736         }
4737
4738         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
4739                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
4740                          es->s_encryption_level);
4741                 goto failed_mount;
4742         }
4743
4744         if (sb->s_blocksize != blocksize) {
4745                 /*
4746                  * bh must be released before kill_bdev(), otherwise
4747                  * it won't be freed and its page also. kill_bdev()
4748                  * is called by sb_set_blocksize().
4749                  */
4750                 brelse(bh);
4751                 /* Validate the filesystem blocksize */
4752                 if (!sb_set_blocksize(sb, blocksize)) {
4753                         ext4_msg(sb, KERN_ERR, "bad block size %d",
4754                                         blocksize);
4755                         bh = NULL;
4756                         goto failed_mount;
4757                 }
4758
4759                 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE;
4760                 offset = do_div(logical_sb_block, blocksize);
4761                 bh = ext4_sb_bread_unmovable(sb, logical_sb_block);
4762                 if (IS_ERR(bh)) {
4763                         ext4_msg(sb, KERN_ERR,
4764                                "Can't read superblock on 2nd try");
4765                         ret = PTR_ERR(bh);
4766                         bh = NULL;
4767                         goto failed_mount;
4768                 }
4769                 es = (struct ext4_super_block *)(bh->b_data + offset);
4770                 sbi->s_es = es;
4771                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
4772                         ext4_msg(sb, KERN_ERR,
4773                                "Magic mismatch, very weird!");
4774                         goto failed_mount;
4775                 }
4776         }
4777
4778         has_huge_files = ext4_has_feature_huge_file(sb);
4779         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
4780                                                       has_huge_files);
4781         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
4782
4783         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
4784         if (ext4_has_feature_64bit(sb)) {
4785                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
4786                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
4787                     !is_power_of_2(sbi->s_desc_size)) {
4788                         ext4_msg(sb, KERN_ERR,
4789                                "unsupported descriptor size %lu",
4790                                sbi->s_desc_size);
4791                         goto failed_mount;
4792                 }
4793         } else
4794                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
4795
4796         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
4797         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
4798
4799         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
4800         if (sbi->s_inodes_per_block == 0)
4801                 goto cantfind_ext4;
4802         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
4803             sbi->s_inodes_per_group > blocksize * 8) {
4804                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
4805                          sbi->s_inodes_per_group);
4806                 goto failed_mount;
4807         }
4808         sbi->s_itb_per_group = sbi->s_inodes_per_group /
4809                                         sbi->s_inodes_per_block;
4810         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
4811         sbi->s_sbh = bh;
4812         sbi->s_mount_state = le16_to_cpu(es->s_state) & ~EXT4_FC_REPLAY;
4813         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
4814         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
4815
4816         for (i = 0; i < 4; i++)
4817                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
4818         sbi->s_def_hash_version = es->s_def_hash_version;
4819         if (ext4_has_feature_dir_index(sb)) {
4820                 i = le32_to_cpu(es->s_flags);
4821                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
4822                         sbi->s_hash_unsigned = 3;
4823                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
4824 #ifdef __CHAR_UNSIGNED__
4825                         if (!sb_rdonly(sb))
4826                                 es->s_flags |=
4827                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
4828                         sbi->s_hash_unsigned = 3;
4829 #else
4830                         if (!sb_rdonly(sb))
4831                                 es->s_flags |=
4832                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
4833 #endif
4834                 }
4835         }
4836
4837         /* Handle clustersize */
4838         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
4839         if (ext4_has_feature_bigalloc(sb)) {
4840                 if (clustersize < blocksize) {
4841                         ext4_msg(sb, KERN_ERR,
4842                                  "cluster size (%d) smaller than "
4843                                  "block size (%d)", clustersize, blocksize);
4844                         goto failed_mount;
4845                 }
4846                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
4847                         le32_to_cpu(es->s_log_block_size);
4848                 sbi->s_clusters_per_group =
4849                         le32_to_cpu(es->s_clusters_per_group);
4850                 if (sbi->s_clusters_per_group > blocksize * 8) {
4851                         ext4_msg(sb, KERN_ERR,
4852                                  "#clusters per group too big: %lu",
4853                                  sbi->s_clusters_per_group);
4854                         goto failed_mount;
4855                 }
4856                 if (sbi->s_blocks_per_group !=
4857                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
4858                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
4859                                  "clusters per group (%lu) inconsistent",
4860                                  sbi->s_blocks_per_group,
4861                                  sbi->s_clusters_per_group);
4862                         goto failed_mount;
4863                 }
4864         } else {
4865                 if (clustersize != blocksize) {
4866                         ext4_msg(sb, KERN_ERR,
4867                                  "fragment/cluster size (%d) != "
4868                                  "block size (%d)", clustersize, blocksize);
4869                         goto failed_mount;
4870                 }
4871                 if (sbi->s_blocks_per_group > blocksize * 8) {
4872                         ext4_msg(sb, KERN_ERR,
4873                                  "#blocks per group too big: %lu",
4874                                  sbi->s_blocks_per_group);
4875                         goto failed_mount;
4876                 }
4877                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
4878                 sbi->s_cluster_bits = 0;
4879         }
4880         sbi->s_cluster_ratio = clustersize / blocksize;
4881
4882         /* Do we have standard group size of clustersize * 8 blocks ? */
4883         if (sbi->s_blocks_per_group == clustersize << 3)
4884                 set_opt2(sb, STD_GROUP_SIZE);
4885
4886         /*
4887          * Test whether we have more sectors than will fit in sector_t,
4888          * and whether the max offset is addressable by the page cache.
4889          */
4890         err = generic_check_addressable(sb->s_blocksize_bits,
4891                                         ext4_blocks_count(es));
4892         if (err) {
4893                 ext4_msg(sb, KERN_ERR, "filesystem"
4894                          " too large to mount safely on this system");
4895                 goto failed_mount;
4896         }
4897
4898         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
4899                 goto cantfind_ext4;
4900
4901         /* check blocks count against device size */
4902         blocks_count = sb_bdev_nr_blocks(sb);
4903         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
4904                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
4905                        "exceeds size of device (%llu blocks)",
4906                        ext4_blocks_count(es), blocks_count);
4907                 goto failed_mount;
4908         }
4909
4910         /*
4911          * It makes no sense for the first data block to be beyond the end
4912          * of the filesystem.
4913          */
4914         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
4915                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4916                          "block %u is beyond end of filesystem (%llu)",
4917                          le32_to_cpu(es->s_first_data_block),
4918                          ext4_blocks_count(es));
4919                 goto failed_mount;
4920         }
4921         if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) &&
4922             (sbi->s_cluster_ratio == 1)) {
4923                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4924                          "block is 0 with a 1k block and cluster size");
4925                 goto failed_mount;
4926         }
4927
4928         blocks_count = (ext4_blocks_count(es) -
4929                         le32_to_cpu(es->s_first_data_block) +
4930                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
4931         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
4932         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
4933                 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu "
4934                        "(block count %llu, first data block %u, "
4935                        "blocks per group %lu)", blocks_count,
4936                        ext4_blocks_count(es),
4937                        le32_to_cpu(es->s_first_data_block),
4938                        EXT4_BLOCKS_PER_GROUP(sb));
4939                 goto failed_mount;
4940         }
4941         sbi->s_groups_count = blocks_count;
4942         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
4943                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
4944         if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) !=
4945             le32_to_cpu(es->s_inodes_count)) {
4946                 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu",
4947                          le32_to_cpu(es->s_inodes_count),
4948                          ((u64)sbi->s_groups_count * sbi->s_inodes_per_group));
4949                 ret = -EINVAL;
4950                 goto failed_mount;
4951         }
4952         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
4953                    EXT4_DESC_PER_BLOCK(sb);
4954         if (ext4_has_feature_meta_bg(sb)) {
4955                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
4956                         ext4_msg(sb, KERN_WARNING,
4957                                  "first meta block group too large: %u "
4958                                  "(group descriptor block count %u)",
4959                                  le32_to_cpu(es->s_first_meta_bg), db_count);
4960                         goto failed_mount;
4961                 }
4962         }
4963         rcu_assign_pointer(sbi->s_group_desc,
4964                            kvmalloc_array(db_count,
4965                                           sizeof(struct buffer_head *),
4966                                           GFP_KERNEL));
4967         if (sbi->s_group_desc == NULL) {
4968                 ext4_msg(sb, KERN_ERR, "not enough memory");
4969                 ret = -ENOMEM;
4970                 goto failed_mount;
4971         }
4972
4973         bgl_lock_init(sbi->s_blockgroup_lock);
4974
4975         /* Pre-read the descriptors into the buffer cache */
4976         for (i = 0; i < db_count; i++) {
4977                 block = descriptor_loc(sb, logical_sb_block, i);
4978                 ext4_sb_breadahead_unmovable(sb, block);
4979         }
4980
4981         for (i = 0; i < db_count; i++) {
4982                 struct buffer_head *bh;
4983
4984                 block = descriptor_loc(sb, logical_sb_block, i);
4985                 bh = ext4_sb_bread_unmovable(sb, block);
4986                 if (IS_ERR(bh)) {
4987                         ext4_msg(sb, KERN_ERR,
4988                                "can't read group descriptor %d", i);
4989                         db_count = i;
4990                         ret = PTR_ERR(bh);
4991                         goto failed_mount2;
4992                 }
4993                 rcu_read_lock();
4994                 rcu_dereference(sbi->s_group_desc)[i] = bh;
4995                 rcu_read_unlock();
4996         }
4997         sbi->s_gdb_count = db_count;
4998         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
4999                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
5000                 ret = -EFSCORRUPTED;
5001                 goto failed_mount2;
5002         }
5003
5004         timer_setup(&sbi->s_err_report, print_daily_error_info, 0);
5005         spin_lock_init(&sbi->s_error_lock);
5006         INIT_WORK(&sbi->s_error_work, flush_stashed_error_work);
5007
5008         /* Register extent status tree shrinker */
5009         if (ext4_es_register_shrinker(sbi))
5010                 goto failed_mount3;
5011
5012         sbi->s_stripe = ext4_get_stripe_size(sbi);
5013         sbi->s_extent_max_zeroout_kb = 32;
5014
5015         /*
5016          * set up enough so that it can read an inode
5017          */
5018         sb->s_op = &ext4_sops;
5019         sb->s_export_op = &ext4_export_ops;
5020         sb->s_xattr = ext4_xattr_handlers;
5021 #ifdef CONFIG_FS_ENCRYPTION
5022         sb->s_cop = &ext4_cryptops;
5023 #endif
5024 #ifdef CONFIG_FS_VERITY
5025         sb->s_vop = &ext4_verityops;
5026 #endif
5027 #ifdef CONFIG_QUOTA
5028         sb->dq_op = &ext4_quota_operations;
5029         if (ext4_has_feature_quota(sb))
5030                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
5031         else
5032                 sb->s_qcop = &ext4_qctl_operations;
5033         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
5034 #endif
5035         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
5036
5037         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
5038         mutex_init(&sbi->s_orphan_lock);
5039
5040         /* Initialize fast commit stuff */
5041         atomic_set(&sbi->s_fc_subtid, 0);
5042         INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_MAIN]);
5043         INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_STAGING]);
5044         INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_MAIN]);
5045         INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_STAGING]);
5046         sbi->s_fc_bytes = 0;
5047         ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE);
5048         sbi->s_fc_ineligible_tid = 0;
5049         spin_lock_init(&sbi->s_fc_lock);
5050         memset(&sbi->s_fc_stats, 0, sizeof(sbi->s_fc_stats));
5051         sbi->s_fc_replay_state.fc_regions = NULL;
5052         sbi->s_fc_replay_state.fc_regions_size = 0;
5053         sbi->s_fc_replay_state.fc_regions_used = 0;
5054         sbi->s_fc_replay_state.fc_regions_valid = 0;
5055         sbi->s_fc_replay_state.fc_modified_inodes = NULL;
5056         sbi->s_fc_replay_state.fc_modified_inodes_size = 0;
5057         sbi->s_fc_replay_state.fc_modified_inodes_used = 0;
5058
5059         sb->s_root = NULL;
5060
5061         needs_recovery = (es->s_last_orphan != 0 ||
5062                           ext4_has_feature_orphan_present(sb) ||
5063                           ext4_has_feature_journal_needs_recovery(sb));
5064
5065         if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb))
5066                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
5067                         goto failed_mount3a;
5068
5069         /*
5070          * The first inode we look at is the journal inode.  Don't try
5071          * root first: it may be modified in the journal!
5072          */
5073         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
5074                 err = ext4_load_journal(sb, es, ctx->journal_devnum);
5075                 if (err)
5076                         goto failed_mount3a;
5077         } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
5078                    ext4_has_feature_journal_needs_recovery(sb)) {
5079                 ext4_msg(sb, KERN_ERR, "required journal recovery "
5080                        "suppressed and not mounted read-only");
5081                 goto failed_mount_wq;
5082         } else {
5083                 /* Nojournal mode, all journal mount options are illegal */
5084                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
5085                         ext4_msg(sb, KERN_ERR, "can't mount with "
5086                                  "journal_checksum, fs mounted w/o journal");
5087                         goto failed_mount_wq;
5088                 }
5089                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5090                         ext4_msg(sb, KERN_ERR, "can't mount with "
5091                                  "journal_async_commit, fs mounted w/o journal");
5092                         goto failed_mount_wq;
5093                 }
5094                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
5095                         ext4_msg(sb, KERN_ERR, "can't mount with "
5096                                  "commit=%lu, fs mounted w/o journal",
5097                                  sbi->s_commit_interval / HZ);
5098                         goto failed_mount_wq;
5099                 }
5100                 if (EXT4_MOUNT_DATA_FLAGS &
5101                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
5102                         ext4_msg(sb, KERN_ERR, "can't mount with "
5103                                  "data=, fs mounted w/o journal");
5104                         goto failed_mount_wq;
5105                 }
5106                 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM;
5107                 clear_opt(sb, JOURNAL_CHECKSUM);
5108                 clear_opt(sb, DATA_FLAGS);
5109                 clear_opt2(sb, JOURNAL_FAST_COMMIT);
5110                 sbi->s_journal = NULL;
5111                 needs_recovery = 0;
5112                 goto no_journal;
5113         }
5114
5115         if (ext4_has_feature_64bit(sb) &&
5116             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
5117                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
5118                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
5119                 goto failed_mount_wq;
5120         }
5121
5122         if (!set_journal_csum_feature_set(sb)) {
5123                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
5124                          "feature set");
5125                 goto failed_mount_wq;
5126         }
5127
5128         if (test_opt2(sb, JOURNAL_FAST_COMMIT) &&
5129                 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
5130                                           JBD2_FEATURE_INCOMPAT_FAST_COMMIT)) {
5131                 ext4_msg(sb, KERN_ERR,
5132                         "Failed to set fast commit journal feature");
5133                 goto failed_mount_wq;
5134         }
5135
5136         /* We have now updated the journal if required, so we can
5137          * validate the data journaling mode. */
5138         switch (test_opt(sb, DATA_FLAGS)) {
5139         case 0:
5140                 /* No mode set, assume a default based on the journal
5141                  * capabilities: ORDERED_DATA if the journal can
5142                  * cope, else JOURNAL_DATA
5143                  */
5144                 if (jbd2_journal_check_available_features
5145                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
5146                         set_opt(sb, ORDERED_DATA);
5147                         sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
5148                 } else {
5149                         set_opt(sb, JOURNAL_DATA);
5150                         sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
5151                 }
5152                 break;
5153
5154         case EXT4_MOUNT_ORDERED_DATA:
5155         case EXT4_MOUNT_WRITEBACK_DATA:
5156                 if (!jbd2_journal_check_available_features
5157                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
5158                         ext4_msg(sb, KERN_ERR, "Journal does not support "
5159                                "requested data journaling mode");
5160                         goto failed_mount_wq;
5161                 }
5162                 break;
5163         default:
5164                 break;
5165         }
5166
5167         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
5168             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5169                 ext4_msg(sb, KERN_ERR, "can't mount with "
5170                         "journal_async_commit in data=ordered mode");
5171                 goto failed_mount_wq;
5172         }
5173
5174         set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio);
5175
5176         sbi->s_journal->j_submit_inode_data_buffers =
5177                 ext4_journal_submit_inode_data_buffers;
5178         sbi->s_journal->j_finish_inode_data_buffers =
5179                 ext4_journal_finish_inode_data_buffers;
5180
5181 no_journal:
5182         if (!test_opt(sb, NO_MBCACHE)) {
5183                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
5184                 if (!sbi->s_ea_block_cache) {
5185                         ext4_msg(sb, KERN_ERR,
5186                                  "Failed to create ea_block_cache");
5187                         goto failed_mount_wq;
5188                 }
5189
5190                 if (ext4_has_feature_ea_inode(sb)) {
5191                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
5192                         if (!sbi->s_ea_inode_cache) {
5193                                 ext4_msg(sb, KERN_ERR,
5194                                          "Failed to create ea_inode_cache");
5195                                 goto failed_mount_wq;
5196                         }
5197                 }
5198         }
5199
5200         if (ext4_has_feature_verity(sb) && blocksize != PAGE_SIZE) {
5201                 ext4_msg(sb, KERN_ERR, "Unsupported blocksize for fs-verity");
5202                 goto failed_mount_wq;
5203         }
5204
5205         /*
5206          * Get the # of file system overhead blocks from the
5207          * superblock if present.
5208          */
5209         sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
5210         /* ignore the precalculated value if it is ridiculous */
5211         if (sbi->s_overhead > ext4_blocks_count(es))
5212                 sbi->s_overhead = 0;
5213         /*
5214          * If the bigalloc feature is not enabled recalculating the
5215          * overhead doesn't take long, so we might as well just redo
5216          * it to make sure we are using the correct value.
5217          */
5218         if (!ext4_has_feature_bigalloc(sb))
5219                 sbi->s_overhead = 0;
5220         if (sbi->s_overhead == 0) {
5221                 err = ext4_calculate_overhead(sb);
5222                 if (err)
5223                         goto failed_mount_wq;
5224         }
5225
5226         /*
5227          * The maximum number of concurrent works can be high and
5228          * concurrency isn't really necessary.  Limit it to 1.
5229          */
5230         EXT4_SB(sb)->rsv_conversion_wq =
5231                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
5232         if (!EXT4_SB(sb)->rsv_conversion_wq) {
5233                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
5234                 ret = -ENOMEM;
5235                 goto failed_mount4;
5236         }
5237
5238         /*
5239          * The jbd2_journal_load will have done any necessary log recovery,
5240          * so we can safely mount the rest of the filesystem now.
5241          */
5242
5243         root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL);
5244         if (IS_ERR(root)) {
5245                 ext4_msg(sb, KERN_ERR, "get root inode failed");
5246                 ret = PTR_ERR(root);
5247                 root = NULL;
5248                 goto failed_mount4;
5249         }
5250         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
5251                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
5252                 iput(root);
5253                 goto failed_mount4;
5254         }
5255
5256         sb->s_root = d_make_root(root);
5257         if (!sb->s_root) {
5258                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
5259                 ret = -ENOMEM;
5260                 goto failed_mount4;
5261         }
5262
5263         ret = ext4_setup_super(sb, es, sb_rdonly(sb));
5264         if (ret == -EROFS) {
5265                 sb->s_flags |= SB_RDONLY;
5266                 ret = 0;
5267         } else if (ret)
5268                 goto failed_mount4a;
5269
5270         ext4_set_resv_clusters(sb);
5271
5272         if (test_opt(sb, BLOCK_VALIDITY)) {
5273                 err = ext4_setup_system_zone(sb);
5274                 if (err) {
5275                         ext4_msg(sb, KERN_ERR, "failed to initialize system "
5276                                  "zone (%d)", err);
5277                         goto failed_mount4a;
5278                 }
5279         }
5280         ext4_fc_replay_cleanup(sb);
5281
5282         ext4_ext_init(sb);
5283
5284         /*
5285          * Enable optimize_scan if number of groups is > threshold. This can be
5286          * turned off by passing "mb_optimize_scan=0". This can also be
5287          * turned on forcefully by passing "mb_optimize_scan=1".
5288          */
5289         if (!(ctx->spec & EXT4_SPEC_mb_optimize_scan)) {
5290                 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD)
5291                         set_opt2(sb, MB_OPTIMIZE_SCAN);
5292                 else
5293                         clear_opt2(sb, MB_OPTIMIZE_SCAN);
5294         }
5295
5296         err = ext4_mb_init(sb);
5297         if (err) {
5298                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
5299                          err);
5300                 goto failed_mount5;
5301         }
5302
5303         /*
5304          * We can only set up the journal commit callback once
5305          * mballoc is initialized
5306          */
5307         if (sbi->s_journal)
5308                 sbi->s_journal->j_commit_callback =
5309                         ext4_journal_commit_callback;
5310
5311         block = ext4_count_free_clusters(sb);
5312         ext4_free_blocks_count_set(sbi->s_es,
5313                                    EXT4_C2B(sbi, block));
5314         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
5315                                   GFP_KERNEL);
5316         if (!err) {
5317                 unsigned long freei = ext4_count_free_inodes(sb);
5318                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
5319                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
5320                                           GFP_KERNEL);
5321         }
5322         if (!err)
5323                 err = percpu_counter_init(&sbi->s_dirs_counter,
5324                                           ext4_count_dirs(sb), GFP_KERNEL);
5325         if (!err)
5326                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
5327                                           GFP_KERNEL);
5328         if (!err)
5329                 err = percpu_counter_init(&sbi->s_sra_exceeded_retry_limit, 0,
5330                                           GFP_KERNEL);
5331         if (!err)
5332                 err = percpu_init_rwsem(&sbi->s_writepages_rwsem);
5333
5334         if (err) {
5335                 ext4_msg(sb, KERN_ERR, "insufficient memory");
5336                 goto failed_mount6;
5337         }
5338
5339         if (ext4_has_feature_flex_bg(sb))
5340                 if (!ext4_fill_flex_info(sb)) {
5341                         ext4_msg(sb, KERN_ERR,
5342                                "unable to initialize "
5343                                "flex_bg meta info!");
5344                         ret = -ENOMEM;
5345                         goto failed_mount6;
5346                 }
5347
5348         err = ext4_register_li_request(sb, first_not_zeroed);
5349         if (err)
5350                 goto failed_mount6;
5351
5352         err = ext4_register_sysfs(sb);
5353         if (err)
5354                 goto failed_mount7;
5355
5356         err = ext4_init_orphan_info(sb);
5357         if (err)
5358                 goto failed_mount8;
5359 #ifdef CONFIG_QUOTA
5360         /* Enable quota usage during mount. */
5361         if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
5362                 err = ext4_enable_quotas(sb);
5363                 if (err)
5364                         goto failed_mount9;
5365         }
5366 #endif  /* CONFIG_QUOTA */
5367
5368         /*
5369          * Save the original bdev mapping's wb_err value which could be
5370          * used to detect the metadata async write error.
5371          */
5372         spin_lock_init(&sbi->s_bdev_wb_lock);
5373         errseq_check_and_advance(&sb->s_bdev->bd_inode->i_mapping->wb_err,
5374                                  &sbi->s_bdev_wb_err);
5375         sb->s_bdev->bd_super = sb;
5376         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
5377         ext4_orphan_cleanup(sb, es);
5378         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
5379         /*
5380          * Update the checksum after updating free space/inode counters and
5381          * ext4_orphan_cleanup. Otherwise the superblock can have an incorrect
5382          * checksum in the buffer cache until it is written out and
5383          * e2fsprogs programs trying to open a file system immediately
5384          * after it is mounted can fail.
5385          */
5386         ext4_superblock_csum_set(sb);
5387         if (needs_recovery) {
5388                 ext4_msg(sb, KERN_INFO, "recovery complete");
5389                 err = ext4_mark_recovery_complete(sb, es);
5390                 if (err)
5391                         goto failed_mount9;
5392         }
5393
5394         if (test_opt(sb, DISCARD) && !bdev_max_discard_sectors(sb->s_bdev))
5395                 ext4_msg(sb, KERN_WARNING,
5396                          "mounting with \"discard\" option, but the device does not support discard");
5397
5398         if (es->s_error_count)
5399                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
5400
5401         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
5402         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
5403         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
5404         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
5405         atomic_set(&sbi->s_warning_count, 0);
5406         atomic_set(&sbi->s_msg_count, 0);
5407
5408         return 0;
5409
5410 cantfind_ext4:
5411         if (!silent)
5412                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
5413         goto failed_mount;
5414
5415 failed_mount9:
5416         ext4_release_orphan_info(sb);
5417 failed_mount8:
5418         ext4_unregister_sysfs(sb);
5419         kobject_put(&sbi->s_kobj);
5420 failed_mount7:
5421         ext4_unregister_li_request(sb);
5422 failed_mount6:
5423         ext4_mb_release(sb);
5424         rcu_read_lock();
5425         flex_groups = rcu_dereference(sbi->s_flex_groups);
5426         if (flex_groups) {
5427                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
5428                         kvfree(flex_groups[i]);
5429                 kvfree(flex_groups);
5430         }
5431         rcu_read_unlock();
5432         percpu_counter_destroy(&sbi->s_freeclusters_counter);
5433         percpu_counter_destroy(&sbi->s_freeinodes_counter);
5434         percpu_counter_destroy(&sbi->s_dirs_counter);
5435         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
5436         percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit);
5437         percpu_free_rwsem(&sbi->s_writepages_rwsem);
5438 failed_mount5:
5439         ext4_ext_release(sb);
5440         ext4_release_system_zone(sb);
5441 failed_mount4a:
5442         dput(sb->s_root);
5443         sb->s_root = NULL;
5444 failed_mount4:
5445         ext4_msg(sb, KERN_ERR, "mount failed");
5446         if (EXT4_SB(sb)->rsv_conversion_wq)
5447                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
5448 failed_mount_wq:
5449         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
5450         sbi->s_ea_inode_cache = NULL;
5451
5452         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
5453         sbi->s_ea_block_cache = NULL;
5454
5455         if (sbi->s_journal) {
5456                 /* flush s_error_work before journal destroy. */
5457                 flush_work(&sbi->s_error_work);
5458                 jbd2_journal_destroy(sbi->s_journal);
5459                 sbi->s_journal = NULL;
5460         }
5461 failed_mount3a:
5462         ext4_es_unregister_shrinker(sbi);
5463 failed_mount3:
5464         /* flush s_error_work before sbi destroy */
5465         flush_work(&sbi->s_error_work);
5466         del_timer_sync(&sbi->s_err_report);
5467         ext4_stop_mmpd(sbi);
5468 failed_mount2:
5469         rcu_read_lock();
5470         group_desc = rcu_dereference(sbi->s_group_desc);
5471         for (i = 0; i < db_count; i++)
5472                 brelse(group_desc[i]);
5473         kvfree(group_desc);
5474         rcu_read_unlock();
5475 failed_mount:
5476         if (sbi->s_chksum_driver)
5477                 crypto_free_shash(sbi->s_chksum_driver);
5478
5479 #if IS_ENABLED(CONFIG_UNICODE)
5480         utf8_unload(sb->s_encoding);
5481 #endif
5482
5483 #ifdef CONFIG_QUOTA
5484         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5485                 kfree(get_qf_name(sb, sbi, i));
5486 #endif
5487         fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy);
5488         /* ext4_blkdev_remove() calls kill_bdev(), release bh before it. */
5489         brelse(bh);
5490         ext4_blkdev_remove(sbi);
5491 out_fail:
5492         sb->s_fs_info = NULL;
5493         return err ? err : ret;
5494 }
5495
5496 static int ext4_fill_super(struct super_block *sb, struct fs_context *fc)
5497 {
5498         struct ext4_fs_context *ctx = fc->fs_private;
5499         struct ext4_sb_info *sbi;
5500         const char *descr;
5501         int ret;
5502
5503         sbi = ext4_alloc_sbi(sb);
5504         if (!sbi)
5505                 return -ENOMEM;
5506
5507         fc->s_fs_info = sbi;
5508
5509         /* Cleanup superblock name */
5510         strreplace(sb->s_id, '/', '!');
5511
5512         sbi->s_sb_block = 1;    /* Default super block location */
5513         if (ctx->spec & EXT4_SPEC_s_sb_block)
5514                 sbi->s_sb_block = ctx->s_sb_block;
5515
5516         ret = __ext4_fill_super(fc, sb);
5517         if (ret < 0)
5518                 goto free_sbi;
5519
5520         if (sbi->s_journal) {
5521                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
5522                         descr = " journalled data mode";
5523                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
5524                         descr = " ordered data mode";
5525                 else
5526                         descr = " writeback data mode";
5527         } else
5528                 descr = "out journal";
5529
5530         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
5531                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
5532                          "Quota mode: %s.", descr, ext4_quota_mode(sb));
5533
5534         /* Update the s_overhead_clusters if necessary */
5535         ext4_update_overhead(sb, false);
5536         return 0;
5537
5538 free_sbi:
5539         ext4_free_sbi(sbi);
5540         fc->s_fs_info = NULL;
5541         return ret;
5542 }
5543
5544 static int ext4_get_tree(struct fs_context *fc)
5545 {
5546         return get_tree_bdev(fc, ext4_fill_super);
5547 }
5548
5549 /*
5550  * Setup any per-fs journal parameters now.  We'll do this both on
5551  * initial mount, once the journal has been initialised but before we've
5552  * done any recovery; and again on any subsequent remount.
5553  */
5554 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
5555 {
5556         struct ext4_sb_info *sbi = EXT4_SB(sb);
5557
5558         journal->j_commit_interval = sbi->s_commit_interval;
5559         journal->j_min_batch_time = sbi->s_min_batch_time;
5560         journal->j_max_batch_time = sbi->s_max_batch_time;
5561         ext4_fc_init(sb, journal);
5562
5563         write_lock(&journal->j_state_lock);
5564         if (test_opt(sb, BARRIER))
5565                 journal->j_flags |= JBD2_BARRIER;
5566         else
5567                 journal->j_flags &= ~JBD2_BARRIER;
5568         if (test_opt(sb, DATA_ERR_ABORT))
5569                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
5570         else
5571                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
5572         write_unlock(&journal->j_state_lock);
5573 }
5574
5575 static struct inode *ext4_get_journal_inode(struct super_block *sb,
5576                                              unsigned int journal_inum)
5577 {
5578         struct inode *journal_inode;
5579
5580         /*
5581          * Test for the existence of a valid inode on disk.  Bad things
5582          * happen if we iget() an unused inode, as the subsequent iput()
5583          * will try to delete it.
5584          */
5585         journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL);
5586         if (IS_ERR(journal_inode)) {
5587                 ext4_msg(sb, KERN_ERR, "no journal found");
5588                 return NULL;
5589         }
5590         if (!journal_inode->i_nlink) {
5591                 make_bad_inode(journal_inode);
5592                 iput(journal_inode);
5593                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
5594                 return NULL;
5595         }
5596
5597         ext4_debug("Journal inode found at %p: %lld bytes\n",
5598                   journal_inode, journal_inode->i_size);
5599         if (!S_ISREG(journal_inode->i_mode)) {
5600                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
5601                 iput(journal_inode);
5602                 return NULL;
5603         }
5604         return journal_inode;
5605 }
5606
5607 static journal_t *ext4_get_journal(struct super_block *sb,
5608                                    unsigned int journal_inum)
5609 {
5610         struct inode *journal_inode;
5611         journal_t *journal;
5612
5613         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5614                 return NULL;
5615
5616         journal_inode = ext4_get_journal_inode(sb, journal_inum);
5617         if (!journal_inode)
5618                 return NULL;
5619
5620         journal = jbd2_journal_init_inode(journal_inode);
5621         if (!journal) {
5622                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
5623                 iput(journal_inode);
5624                 return NULL;
5625         }
5626         journal->j_private = sb;
5627         ext4_init_journal_params(sb, journal);
5628         return journal;
5629 }
5630
5631 static journal_t *ext4_get_dev_journal(struct super_block *sb,
5632                                        dev_t j_dev)
5633 {
5634         struct buffer_head *bh;
5635         journal_t *journal;
5636         ext4_fsblk_t start;
5637         ext4_fsblk_t len;
5638         int hblock, blocksize;
5639         ext4_fsblk_t sb_block;
5640         unsigned long offset;
5641         struct ext4_super_block *es;
5642         struct block_device *bdev;
5643
5644         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5645                 return NULL;
5646
5647         bdev = ext4_blkdev_get(j_dev, sb);
5648         if (bdev == NULL)
5649                 return NULL;
5650
5651         blocksize = sb->s_blocksize;
5652         hblock = bdev_logical_block_size(bdev);
5653         if (blocksize < hblock) {
5654                 ext4_msg(sb, KERN_ERR,
5655                         "blocksize too small for journal device");
5656                 goto out_bdev;
5657         }
5658
5659         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
5660         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
5661         set_blocksize(bdev, blocksize);
5662         if (!(bh = __bread(bdev, sb_block, blocksize))) {
5663                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
5664                        "external journal");
5665                 goto out_bdev;
5666         }
5667
5668         es = (struct ext4_super_block *) (bh->b_data + offset);
5669         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
5670             !(le32_to_cpu(es->s_feature_incompat) &
5671               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
5672                 ext4_msg(sb, KERN_ERR, "external journal has "
5673                                         "bad superblock");
5674                 brelse(bh);
5675                 goto out_bdev;
5676         }
5677
5678         if ((le32_to_cpu(es->s_feature_ro_compat) &
5679              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
5680             es->s_checksum != ext4_superblock_csum(sb, es)) {
5681                 ext4_msg(sb, KERN_ERR, "external journal has "
5682                                        "corrupt superblock");
5683                 brelse(bh);
5684                 goto out_bdev;
5685         }
5686
5687         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
5688                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
5689                 brelse(bh);
5690                 goto out_bdev;
5691         }
5692
5693         len = ext4_blocks_count(es);
5694         start = sb_block + 1;
5695         brelse(bh);     /* we're done with the superblock */
5696
5697         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
5698                                         start, len, blocksize);
5699         if (!journal) {
5700                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
5701                 goto out_bdev;
5702         }
5703         journal->j_private = sb;
5704         if (ext4_read_bh_lock(journal->j_sb_buffer, REQ_META | REQ_PRIO, true)) {
5705                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
5706                 goto out_journal;
5707         }
5708         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
5709                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
5710                                         "user (unsupported) - %d",
5711                         be32_to_cpu(journal->j_superblock->s_nr_users));
5712                 goto out_journal;
5713         }
5714         EXT4_SB(sb)->s_journal_bdev = bdev;
5715         ext4_init_journal_params(sb, journal);
5716         return journal;
5717
5718 out_journal:
5719         jbd2_journal_destroy(journal);
5720 out_bdev:
5721         ext4_blkdev_put(bdev);
5722         return NULL;
5723 }
5724
5725 static int ext4_load_journal(struct super_block *sb,
5726                              struct ext4_super_block *es,
5727                              unsigned long journal_devnum)
5728 {
5729         journal_t *journal;
5730         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
5731         dev_t journal_dev;
5732         int err = 0;
5733         int really_read_only;
5734         int journal_dev_ro;
5735
5736         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5737                 return -EFSCORRUPTED;
5738
5739         if (journal_devnum &&
5740             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5741                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
5742                         "numbers have changed");
5743                 journal_dev = new_decode_dev(journal_devnum);
5744         } else
5745                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
5746
5747         if (journal_inum && journal_dev) {
5748                 ext4_msg(sb, KERN_ERR,
5749                          "filesystem has both journal inode and journal device!");
5750                 return -EINVAL;
5751         }
5752
5753         if (journal_inum) {
5754                 journal = ext4_get_journal(sb, journal_inum);
5755                 if (!journal)
5756                         return -EINVAL;
5757         } else {
5758                 journal = ext4_get_dev_journal(sb, journal_dev);
5759                 if (!journal)
5760                         return -EINVAL;
5761         }
5762
5763         journal_dev_ro = bdev_read_only(journal->j_dev);
5764         really_read_only = bdev_read_only(sb->s_bdev) | journal_dev_ro;
5765
5766         if (journal_dev_ro && !sb_rdonly(sb)) {
5767                 ext4_msg(sb, KERN_ERR,
5768                          "journal device read-only, try mounting with '-o ro'");
5769                 err = -EROFS;
5770                 goto err_out;
5771         }
5772
5773         /*
5774          * Are we loading a blank journal or performing recovery after a
5775          * crash?  For recovery, we need to check in advance whether we
5776          * can get read-write access to the device.
5777          */
5778         if (ext4_has_feature_journal_needs_recovery(sb)) {
5779                 if (sb_rdonly(sb)) {
5780                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
5781                                         "required on readonly filesystem");
5782                         if (really_read_only) {
5783                                 ext4_msg(sb, KERN_ERR, "write access "
5784                                         "unavailable, cannot proceed "
5785                                         "(try mounting with noload)");
5786                                 err = -EROFS;
5787                                 goto err_out;
5788                         }
5789                         ext4_msg(sb, KERN_INFO, "write access will "
5790                                "be enabled during recovery");
5791                 }
5792         }
5793
5794         if (!(journal->j_flags & JBD2_BARRIER))
5795                 ext4_msg(sb, KERN_INFO, "barriers disabled");
5796
5797         if (!ext4_has_feature_journal_needs_recovery(sb))
5798                 err = jbd2_journal_wipe(journal, !really_read_only);
5799         if (!err) {
5800                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
5801                 if (save)
5802                         memcpy(save, ((char *) es) +
5803                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
5804                 err = jbd2_journal_load(journal);
5805                 if (save)
5806                         memcpy(((char *) es) + EXT4_S_ERR_START,
5807                                save, EXT4_S_ERR_LEN);
5808                 kfree(save);
5809         }
5810
5811         if (err) {
5812                 ext4_msg(sb, KERN_ERR, "error loading journal");
5813                 goto err_out;
5814         }
5815
5816         EXT4_SB(sb)->s_journal = journal;
5817         err = ext4_clear_journal_err(sb, es);
5818         if (err) {
5819                 EXT4_SB(sb)->s_journal = NULL;
5820                 jbd2_journal_destroy(journal);
5821                 return err;
5822         }
5823
5824         if (!really_read_only && journal_devnum &&
5825             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5826                 es->s_journal_dev = cpu_to_le32(journal_devnum);
5827
5828                 /* Make sure we flush the recovery flag to disk. */
5829                 ext4_commit_super(sb);
5830         }
5831
5832         return 0;
5833
5834 err_out:
5835         jbd2_journal_destroy(journal);
5836         return err;
5837 }
5838
5839 /* Copy state of EXT4_SB(sb) into buffer for on-disk superblock */
5840 static void ext4_update_super(struct super_block *sb)
5841 {
5842         struct ext4_sb_info *sbi = EXT4_SB(sb);
5843         struct ext4_super_block *es = sbi->s_es;
5844         struct buffer_head *sbh = sbi->s_sbh;
5845
5846         lock_buffer(sbh);
5847         /*
5848          * If the file system is mounted read-only, don't update the
5849          * superblock write time.  This avoids updating the superblock
5850          * write time when we are mounting the root file system
5851          * read/only but we need to replay the journal; at that point,
5852          * for people who are east of GMT and who make their clock
5853          * tick in localtime for Windows bug-for-bug compatibility,
5854          * the clock is set in the future, and this will cause e2fsck
5855          * to complain and force a full file system check.
5856          */
5857         if (!(sb->s_flags & SB_RDONLY))
5858                 ext4_update_tstamp(es, s_wtime);
5859         es->s_kbytes_written =
5860                 cpu_to_le64(sbi->s_kbytes_written +
5861                     ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) -
5862                       sbi->s_sectors_written_start) >> 1));
5863         if (percpu_counter_initialized(&sbi->s_freeclusters_counter))
5864                 ext4_free_blocks_count_set(es,
5865                         EXT4_C2B(sbi, percpu_counter_sum_positive(
5866                                 &sbi->s_freeclusters_counter)));
5867         if (percpu_counter_initialized(&sbi->s_freeinodes_counter))
5868                 es->s_free_inodes_count =
5869                         cpu_to_le32(percpu_counter_sum_positive(
5870                                 &sbi->s_freeinodes_counter));
5871         /* Copy error information to the on-disk superblock */
5872         spin_lock(&sbi->s_error_lock);
5873         if (sbi->s_add_error_count > 0) {
5874                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
5875                 if (!es->s_first_error_time && !es->s_first_error_time_hi) {
5876                         __ext4_update_tstamp(&es->s_first_error_time,
5877                                              &es->s_first_error_time_hi,
5878                                              sbi->s_first_error_time);
5879                         strncpy(es->s_first_error_func, sbi->s_first_error_func,
5880                                 sizeof(es->s_first_error_func));
5881                         es->s_first_error_line =
5882                                 cpu_to_le32(sbi->s_first_error_line);
5883                         es->s_first_error_ino =
5884                                 cpu_to_le32(sbi->s_first_error_ino);
5885                         es->s_first_error_block =
5886                                 cpu_to_le64(sbi->s_first_error_block);
5887                         es->s_first_error_errcode =
5888                                 ext4_errno_to_code(sbi->s_first_error_code);
5889                 }
5890                 __ext4_update_tstamp(&es->s_last_error_time,
5891                                      &es->s_last_error_time_hi,
5892                                      sbi->s_last_error_time);
5893                 strncpy(es->s_last_error_func, sbi->s_last_error_func,
5894                         sizeof(es->s_last_error_func));
5895                 es->s_last_error_line = cpu_to_le32(sbi->s_last_error_line);
5896                 es->s_last_error_ino = cpu_to_le32(sbi->s_last_error_ino);
5897                 es->s_last_error_block = cpu_to_le64(sbi->s_last_error_block);
5898                 es->s_last_error_errcode =
5899                                 ext4_errno_to_code(sbi->s_last_error_code);
5900                 /*
5901                  * Start the daily error reporting function if it hasn't been
5902                  * started already
5903                  */
5904                 if (!es->s_error_count)
5905                         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);
5906                 le32_add_cpu(&es->s_error_count, sbi->s_add_error_count);
5907                 sbi->s_add_error_count = 0;
5908         }
5909         spin_unlock(&sbi->s_error_lock);
5910
5911         ext4_superblock_csum_set(sb);
5912         unlock_buffer(sbh);
5913 }
5914
5915 static int ext4_commit_super(struct super_block *sb)
5916 {
5917         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
5918
5919         if (!sbh)
5920                 return -EINVAL;
5921         if (block_device_ejected(sb))
5922                 return -ENODEV;
5923
5924         ext4_update_super(sb);
5925
5926         lock_buffer(sbh);
5927         /* Buffer got discarded which means block device got invalidated */
5928         if (!buffer_mapped(sbh)) {
5929                 unlock_buffer(sbh);
5930                 return -EIO;
5931         }
5932
5933         if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
5934                 /*
5935                  * Oh, dear.  A previous attempt to write the
5936                  * superblock failed.  This could happen because the
5937                  * USB device was yanked out.  Or it could happen to
5938                  * be a transient write error and maybe the block will
5939                  * be remapped.  Nothing we can do but to retry the
5940                  * write and hope for the best.
5941                  */
5942                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
5943                        "superblock detected");
5944                 clear_buffer_write_io_error(sbh);
5945                 set_buffer_uptodate(sbh);
5946         }
5947         get_bh(sbh);
5948         /* Clear potential dirty bit if it was journalled update */
5949         clear_buffer_dirty(sbh);
5950         sbh->b_end_io = end_buffer_write_sync;
5951         submit_bh(REQ_OP_WRITE | REQ_SYNC |
5952                   (test_opt(sb, BARRIER) ? REQ_FUA : 0), sbh);
5953         wait_on_buffer(sbh);
5954         if (buffer_write_io_error(sbh)) {
5955                 ext4_msg(sb, KERN_ERR, "I/O error while writing "
5956                        "superblock");
5957                 clear_buffer_write_io_error(sbh);
5958                 set_buffer_uptodate(sbh);
5959                 return -EIO;
5960         }
5961         return 0;
5962 }
5963
5964 /*
5965  * Have we just finished recovery?  If so, and if we are mounting (or
5966  * remounting) the filesystem readonly, then we will end up with a
5967  * consistent fs on disk.  Record that fact.
5968  */
5969 static int ext4_mark_recovery_complete(struct super_block *sb,
5970                                        struct ext4_super_block *es)
5971 {
5972         int err;
5973         journal_t *journal = EXT4_SB(sb)->s_journal;
5974
5975         if (!ext4_has_feature_journal(sb)) {
5976                 if (journal != NULL) {
5977                         ext4_error(sb, "Journal got removed while the fs was "
5978                                    "mounted!");
5979                         return -EFSCORRUPTED;
5980                 }
5981                 return 0;
5982         }
5983         jbd2_journal_lock_updates(journal);
5984         err = jbd2_journal_flush(journal, 0);
5985         if (err < 0)
5986                 goto out;
5987
5988         if (sb_rdonly(sb) && (ext4_has_feature_journal_needs_recovery(sb) ||
5989             ext4_has_feature_orphan_present(sb))) {
5990                 if (!ext4_orphan_file_empty(sb)) {
5991                         ext4_error(sb, "Orphan file not empty on read-only fs.");
5992                         err = -EFSCORRUPTED;
5993                         goto out;
5994                 }
5995                 ext4_clear_feature_journal_needs_recovery(sb);
5996                 ext4_clear_feature_orphan_present(sb);
5997                 ext4_commit_super(sb);
5998         }
5999 out:
6000         jbd2_journal_unlock_updates(journal);
6001         return err;
6002 }
6003
6004 /*
6005  * If we are mounting (or read-write remounting) a filesystem whose journal
6006  * has recorded an error from a previous lifetime, move that error to the
6007  * main filesystem now.
6008  */
6009 static int ext4_clear_journal_err(struct super_block *sb,
6010                                    struct ext4_super_block *es)
6011 {
6012         journal_t *journal;
6013         int j_errno;
6014         const char *errstr;
6015
6016         if (!ext4_has_feature_journal(sb)) {
6017                 ext4_error(sb, "Journal got removed while the fs was mounted!");
6018                 return -EFSCORRUPTED;
6019         }
6020
6021         journal = EXT4_SB(sb)->s_journal;
6022
6023         /*
6024          * Now check for any error status which may have been recorded in the
6025          * journal by a prior ext4_error() or ext4_abort()
6026          */
6027
6028         j_errno = jbd2_journal_errno(journal);
6029         if (j_errno) {
6030                 char nbuf[16];
6031
6032                 errstr = ext4_decode_error(sb, j_errno, nbuf);
6033                 ext4_warning(sb, "Filesystem error recorded "
6034                              "from previous mount: %s", errstr);
6035                 ext4_warning(sb, "Marking fs in need of filesystem check.");
6036
6037                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
6038                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
6039                 ext4_commit_super(sb);
6040
6041                 jbd2_journal_clear_err(journal);
6042                 jbd2_journal_update_sb_errno(journal);
6043         }
6044         return 0;
6045 }
6046
6047 /*
6048  * Force the running and committing transactions to commit,
6049  * and wait on the commit.
6050  */
6051 int ext4_force_commit(struct super_block *sb)
6052 {
6053         journal_t *journal;
6054
6055         if (sb_rdonly(sb))
6056                 return 0;
6057
6058         journal = EXT4_SB(sb)->s_journal;
6059         return ext4_journal_force_commit(journal);
6060 }
6061
6062 static int ext4_sync_fs(struct super_block *sb, int wait)
6063 {
6064         int ret = 0;
6065         tid_t target;
6066         bool needs_barrier = false;
6067         struct ext4_sb_info *sbi = EXT4_SB(sb);
6068
6069         if (unlikely(ext4_forced_shutdown(sbi)))
6070                 return 0;
6071
6072         trace_ext4_sync_fs(sb, wait);
6073         flush_workqueue(sbi->rsv_conversion_wq);
6074         /*
6075          * Writeback quota in non-journalled quota case - journalled quota has
6076          * no dirty dquots
6077          */
6078         dquot_writeback_dquots(sb, -1);
6079         /*
6080          * Data writeback is possible w/o journal transaction, so barrier must
6081          * being sent at the end of the function. But we can skip it if
6082          * transaction_commit will do it for us.
6083          */
6084         if (sbi->s_journal) {
6085                 target = jbd2_get_latest_transaction(sbi->s_journal);
6086                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
6087                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
6088                         needs_barrier = true;
6089
6090                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
6091                         if (wait)
6092                                 ret = jbd2_log_wait_commit(sbi->s_journal,
6093                                                            target);
6094                 }
6095         } else if (wait && test_opt(sb, BARRIER))
6096                 needs_barrier = true;
6097         if (needs_barrier) {
6098                 int err;
6099                 err = blkdev_issue_flush(sb->s_bdev);
6100                 if (!ret)
6101                         ret = err;
6102         }
6103
6104         return ret;
6105 }
6106
6107 /*
6108  * LVM calls this function before a (read-only) snapshot is created.  This
6109  * gives us a chance to flush the journal completely and mark the fs clean.
6110  *
6111  * Note that only this function cannot bring a filesystem to be in a clean
6112  * state independently. It relies on upper layer to stop all data & metadata
6113  * modifications.
6114  */
6115 static int ext4_freeze(struct super_block *sb)
6116 {
6117         int error = 0;
6118         journal_t *journal;
6119
6120         if (sb_rdonly(sb))
6121                 return 0;
6122
6123         journal = EXT4_SB(sb)->s_journal;
6124
6125         if (journal) {
6126                 /* Now we set up the journal barrier. */
6127                 jbd2_journal_lock_updates(journal);
6128
6129                 /*
6130                  * Don't clear the needs_recovery flag if we failed to
6131                  * flush the journal.
6132                  */
6133                 error = jbd2_journal_flush(journal, 0);
6134                 if (error < 0)
6135                         goto out;
6136
6137                 /* Journal blocked and flushed, clear needs_recovery flag. */
6138                 ext4_clear_feature_journal_needs_recovery(sb);
6139                 if (ext4_orphan_file_empty(sb))
6140                         ext4_clear_feature_orphan_present(sb);
6141         }
6142
6143         error = ext4_commit_super(sb);
6144 out:
6145         if (journal)
6146                 /* we rely on upper layer to stop further updates */
6147                 jbd2_journal_unlock_updates(journal);
6148         return error;
6149 }
6150
6151 /*
6152  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
6153  * flag here, even though the filesystem is not technically dirty yet.
6154  */
6155 static int ext4_unfreeze(struct super_block *sb)
6156 {
6157         if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
6158                 return 0;
6159
6160         if (EXT4_SB(sb)->s_journal) {
6161                 /* Reset the needs_recovery flag before the fs is unlocked. */
6162                 ext4_set_feature_journal_needs_recovery(sb);
6163                 if (ext4_has_feature_orphan_file(sb))
6164                         ext4_set_feature_orphan_present(sb);
6165         }
6166
6167         ext4_commit_super(sb);
6168         return 0;
6169 }
6170
6171 /*
6172  * Structure to save mount options for ext4_remount's benefit
6173  */
6174 struct ext4_mount_options {
6175         unsigned long s_mount_opt;
6176         unsigned long s_mount_opt2;
6177         kuid_t s_resuid;
6178         kgid_t s_resgid;
6179         unsigned long s_commit_interval;
6180         u32 s_min_batch_time, s_max_batch_time;
6181 #ifdef CONFIG_QUOTA
6182         int s_jquota_fmt;
6183         char *s_qf_names[EXT4_MAXQUOTAS];
6184 #endif
6185 };
6186
6187 static int __ext4_remount(struct fs_context *fc, struct super_block *sb)
6188 {
6189         struct ext4_fs_context *ctx = fc->fs_private;
6190         struct ext4_super_block *es;
6191         struct ext4_sb_info *sbi = EXT4_SB(sb);
6192         unsigned long old_sb_flags;
6193         struct ext4_mount_options old_opts;
6194         ext4_group_t g;
6195         int err = 0;
6196 #ifdef CONFIG_QUOTA
6197         int enable_quota = 0;
6198         int i, j;
6199         char *to_free[EXT4_MAXQUOTAS];
6200 #endif
6201
6202
6203         /* Store the original options */
6204         old_sb_flags = sb->s_flags;
6205         old_opts.s_mount_opt = sbi->s_mount_opt;
6206         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
6207         old_opts.s_resuid = sbi->s_resuid;
6208         old_opts.s_resgid = sbi->s_resgid;
6209         old_opts.s_commit_interval = sbi->s_commit_interval;
6210         old_opts.s_min_batch_time = sbi->s_min_batch_time;
6211         old_opts.s_max_batch_time = sbi->s_max_batch_time;
6212 #ifdef CONFIG_QUOTA
6213         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
6214         for (i = 0; i < EXT4_MAXQUOTAS; i++)
6215                 if (sbi->s_qf_names[i]) {
6216                         char *qf_name = get_qf_name(sb, sbi, i);
6217
6218                         old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL);
6219                         if (!old_opts.s_qf_names[i]) {
6220                                 for (j = 0; j < i; j++)
6221                                         kfree(old_opts.s_qf_names[j]);
6222                                 return -ENOMEM;
6223                         }
6224                 } else
6225                         old_opts.s_qf_names[i] = NULL;
6226 #endif
6227         if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)) {
6228                 if (sbi->s_journal && sbi->s_journal->j_task->io_context)
6229                         ctx->journal_ioprio =
6230                                 sbi->s_journal->j_task->io_context->ioprio;
6231                 else
6232                         ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
6233
6234         }
6235
6236         ext4_apply_options(fc, sb);
6237
6238         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
6239             test_opt(sb, JOURNAL_CHECKSUM)) {
6240                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
6241                          "during remount not supported; ignoring");
6242                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
6243         }
6244
6245         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
6246                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
6247                         ext4_msg(sb, KERN_ERR, "can't mount with "
6248                                  "both data=journal and delalloc");
6249                         err = -EINVAL;
6250                         goto restore_opts;
6251                 }
6252                 if (test_opt(sb, DIOREAD_NOLOCK)) {
6253                         ext4_msg(sb, KERN_ERR, "can't mount with "
6254                                  "both data=journal and dioread_nolock");
6255                         err = -EINVAL;
6256                         goto restore_opts;
6257                 }
6258         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
6259                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
6260                         ext4_msg(sb, KERN_ERR, "can't mount with "
6261                                 "journal_async_commit in data=ordered mode");
6262                         err = -EINVAL;
6263                         goto restore_opts;
6264                 }
6265         }
6266
6267         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
6268                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
6269                 err = -EINVAL;
6270                 goto restore_opts;
6271         }
6272
6273         if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED))
6274                 ext4_abort(sb, ESHUTDOWN, "Abort forced by user");
6275
6276         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
6277                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
6278
6279         es = sbi->s_es;
6280
6281         if (sbi->s_journal) {
6282                 ext4_init_journal_params(sb, sbi->s_journal);
6283                 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio);
6284         }
6285
6286         /* Flush outstanding errors before changing fs state */
6287         flush_work(&sbi->s_error_work);
6288
6289         if ((bool)(fc->sb_flags & SB_RDONLY) != sb_rdonly(sb)) {
6290                 if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED)) {
6291                         err = -EROFS;
6292                         goto restore_opts;
6293                 }
6294
6295                 if (fc->sb_flags & SB_RDONLY) {
6296                         err = sync_filesystem(sb);
6297                         if (err < 0)
6298                                 goto restore_opts;
6299                         err = dquot_suspend(sb, -1);
6300                         if (err < 0)
6301                                 goto restore_opts;
6302
6303                         /*
6304                          * First of all, the unconditional stuff we have to do
6305                          * to disable replay of the journal when we next remount
6306                          */
6307                         sb->s_flags |= SB_RDONLY;
6308
6309                         /*
6310                          * OK, test if we are remounting a valid rw partition
6311                          * readonly, and if so set the rdonly flag and then
6312                          * mark the partition as valid again.
6313                          */
6314                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
6315                             (sbi->s_mount_state & EXT4_VALID_FS))
6316                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
6317
6318                         if (sbi->s_journal) {
6319                                 /*
6320                                  * We let remount-ro finish even if marking fs
6321                                  * as clean failed...
6322                                  */
6323                                 ext4_mark_recovery_complete(sb, es);
6324                         }
6325                 } else {
6326                         /* Make sure we can mount this feature set readwrite */
6327                         if (ext4_has_feature_readonly(sb) ||
6328                             !ext4_feature_set_ok(sb, 0)) {
6329                                 err = -EROFS;
6330                                 goto restore_opts;
6331                         }
6332                         /*
6333                          * Make sure the group descriptor checksums
6334                          * are sane.  If they aren't, refuse to remount r/w.
6335                          */
6336                         for (g = 0; g < sbi->s_groups_count; g++) {
6337                                 struct ext4_group_desc *gdp =
6338                                         ext4_get_group_desc(sb, g, NULL);
6339
6340                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
6341                                         ext4_msg(sb, KERN_ERR,
6342                "ext4_remount: Checksum for group %u failed (%u!=%u)",
6343                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
6344                                                le16_to_cpu(gdp->bg_checksum));
6345                                         err = -EFSBADCRC;
6346                                         goto restore_opts;
6347                                 }
6348                         }
6349
6350                         /*
6351                          * If we have an unprocessed orphan list hanging
6352                          * around from a previously readonly bdev mount,
6353                          * require a full umount/remount for now.
6354                          */
6355                         if (es->s_last_orphan || !ext4_orphan_file_empty(sb)) {
6356                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
6357                                        "remount RDWR because of unprocessed "
6358                                        "orphan inode list.  Please "
6359                                        "umount/remount instead");
6360                                 err = -EINVAL;
6361                                 goto restore_opts;
6362                         }
6363
6364                         /*
6365                          * Mounting a RDONLY partition read-write, so reread
6366                          * and store the current valid flag.  (It may have
6367                          * been changed by e2fsck since we originally mounted
6368                          * the partition.)
6369                          */
6370                         if (sbi->s_journal) {
6371                                 err = ext4_clear_journal_err(sb, es);
6372                                 if (err)
6373                                         goto restore_opts;
6374                         }
6375                         sbi->s_mount_state = (le16_to_cpu(es->s_state) &
6376                                               ~EXT4_FC_REPLAY);
6377
6378                         err = ext4_setup_super(sb, es, 0);
6379                         if (err)
6380                                 goto restore_opts;
6381
6382                         sb->s_flags &= ~SB_RDONLY;
6383                         if (ext4_has_feature_mmp(sb))
6384                                 if (ext4_multi_mount_protect(sb,
6385                                                 le64_to_cpu(es->s_mmp_block))) {
6386                                         err = -EROFS;
6387                                         goto restore_opts;
6388                                 }
6389 #ifdef CONFIG_QUOTA
6390                         enable_quota = 1;
6391 #endif
6392                 }
6393         }
6394
6395         /*
6396          * Reinitialize lazy itable initialization thread based on
6397          * current settings
6398          */
6399         if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
6400                 ext4_unregister_li_request(sb);
6401         else {
6402                 ext4_group_t first_not_zeroed;
6403                 first_not_zeroed = ext4_has_uninit_itable(sb);
6404                 ext4_register_li_request(sb, first_not_zeroed);
6405         }
6406
6407         /*
6408          * Handle creation of system zone data early because it can fail.
6409          * Releasing of existing data is done when we are sure remount will
6410          * succeed.
6411          */
6412         if (test_opt(sb, BLOCK_VALIDITY) && !sbi->s_system_blks) {
6413                 err = ext4_setup_system_zone(sb);
6414                 if (err)
6415                         goto restore_opts;
6416         }
6417
6418         if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) {
6419                 err = ext4_commit_super(sb);
6420                 if (err)
6421                         goto restore_opts;
6422         }
6423
6424 #ifdef CONFIG_QUOTA
6425         /* Release old quota file names */
6426         for (i = 0; i < EXT4_MAXQUOTAS; i++)
6427                 kfree(old_opts.s_qf_names[i]);
6428         if (enable_quota) {
6429                 if (sb_any_quota_suspended(sb))
6430                         dquot_resume(sb, -1);
6431                 else if (ext4_has_feature_quota(sb)) {
6432                         err = ext4_enable_quotas(sb);
6433                         if (err)
6434                                 goto restore_opts;
6435                 }
6436         }
6437 #endif
6438         if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks)
6439                 ext4_release_system_zone(sb);
6440
6441         if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb))
6442                 ext4_stop_mmpd(sbi);
6443
6444         return 0;
6445
6446 restore_opts:
6447         sb->s_flags = old_sb_flags;
6448         sbi->s_mount_opt = old_opts.s_mount_opt;
6449         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
6450         sbi->s_resuid = old_opts.s_resuid;
6451         sbi->s_resgid = old_opts.s_resgid;
6452         sbi->s_commit_interval = old_opts.s_commit_interval;
6453         sbi->s_min_batch_time = old_opts.s_min_batch_time;
6454         sbi->s_max_batch_time = old_opts.s_max_batch_time;
6455         if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks)
6456                 ext4_release_system_zone(sb);
6457 #ifdef CONFIG_QUOTA
6458         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
6459         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
6460                 to_free[i] = get_qf_name(sb, sbi, i);
6461                 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]);
6462         }
6463         synchronize_rcu();
6464         for (i = 0; i < EXT4_MAXQUOTAS; i++)
6465                 kfree(to_free[i]);
6466 #endif
6467         if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb))
6468                 ext4_stop_mmpd(sbi);
6469         return err;
6470 }
6471
6472 static int ext4_reconfigure(struct fs_context *fc)
6473 {
6474         struct super_block *sb = fc->root->d_sb;
6475         int ret;
6476
6477         fc->s_fs_info = EXT4_SB(sb);
6478
6479         ret = ext4_check_opt_consistency(fc, sb);
6480         if (ret < 0)
6481                 return ret;
6482
6483         ret = __ext4_remount(fc, sb);
6484         if (ret < 0)
6485                 return ret;
6486
6487         ext4_msg(sb, KERN_INFO, "re-mounted. Quota mode: %s.",
6488                  ext4_quota_mode(sb));
6489
6490         return 0;
6491 }
6492
6493 #ifdef CONFIG_QUOTA
6494 static int ext4_statfs_project(struct super_block *sb,
6495                                kprojid_t projid, struct kstatfs *buf)
6496 {
6497         struct kqid qid;
6498         struct dquot *dquot;
6499         u64 limit;
6500         u64 curblock;
6501
6502         qid = make_kqid_projid(projid);
6503         dquot = dqget(sb, qid);
6504         if (IS_ERR(dquot))
6505                 return PTR_ERR(dquot);
6506         spin_lock(&dquot->dq_dqb_lock);
6507
6508         limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit,
6509                              dquot->dq_dqb.dqb_bhardlimit);
6510         limit >>= sb->s_blocksize_bits;
6511
6512         if (limit && buf->f_blocks > limit) {
6513                 curblock = (dquot->dq_dqb.dqb_curspace +
6514                             dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits;
6515                 buf->f_blocks = limit;
6516                 buf->f_bfree = buf->f_bavail =
6517                         (buf->f_blocks > curblock) ?
6518                          (buf->f_blocks - curblock) : 0;
6519         }
6520
6521         limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit,
6522                              dquot->dq_dqb.dqb_ihardlimit);
6523         if (limit && buf->f_files > limit) {
6524                 buf->f_files = limit;
6525                 buf->f_ffree =
6526                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
6527                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
6528         }
6529
6530         spin_unlock(&dquot->dq_dqb_lock);
6531         dqput(dquot);
6532         return 0;
6533 }
6534 #endif
6535
6536 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
6537 {
6538         struct super_block *sb = dentry->d_sb;
6539         struct ext4_sb_info *sbi = EXT4_SB(sb);
6540         struct ext4_super_block *es = sbi->s_es;
6541         ext4_fsblk_t overhead = 0, resv_blocks;
6542         s64 bfree;
6543         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
6544
6545         if (!test_opt(sb, MINIX_DF))
6546                 overhead = sbi->s_overhead;
6547
6548         buf->f_type = EXT4_SUPER_MAGIC;
6549         buf->f_bsize = sb->s_blocksize;
6550         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
6551         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
6552                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
6553         /* prevent underflow in case that few free space is available */
6554         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
6555         buf->f_bavail = buf->f_bfree -
6556                         (ext4_r_blocks_count(es) + resv_blocks);
6557         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
6558                 buf->f_bavail = 0;
6559         buf->f_files = le32_to_cpu(es->s_inodes_count);
6560         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
6561         buf->f_namelen = EXT4_NAME_LEN;
6562         buf->f_fsid = uuid_to_fsid(es->s_uuid);
6563
6564 #ifdef CONFIG_QUOTA
6565         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
6566             sb_has_quota_limits_enabled(sb, PRJQUOTA))
6567                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
6568 #endif
6569         return 0;
6570 }
6571
6572
6573 #ifdef CONFIG_QUOTA
6574
6575 /*
6576  * Helper functions so that transaction is started before we acquire dqio_sem
6577  * to keep correct lock ordering of transaction > dqio_sem
6578  */
6579 static inline struct inode *dquot_to_inode(struct dquot *dquot)
6580 {
6581         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
6582 }
6583
6584 static int ext4_write_dquot(struct dquot *dquot)
6585 {
6586         int ret, err;
6587         handle_t *handle;
6588         struct inode *inode;
6589
6590         inode = dquot_to_inode(dquot);
6591         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
6592                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
6593         if (IS_ERR(handle))
6594                 return PTR_ERR(handle);
6595         ret = dquot_commit(dquot);
6596         err = ext4_journal_stop(handle);
6597         if (!ret)
6598                 ret = err;
6599         return ret;
6600 }
6601
6602 static int ext4_acquire_dquot(struct dquot *dquot)
6603 {
6604         int ret, err;
6605         handle_t *handle;
6606
6607         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
6608                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
6609         if (IS_ERR(handle))
6610                 return PTR_ERR(handle);
6611         ret = dquot_acquire(dquot);
6612         err = ext4_journal_stop(handle);
6613         if (!ret)
6614                 ret = err;
6615         return ret;
6616 }
6617
6618 static int ext4_release_dquot(struct dquot *dquot)
6619 {
6620         int ret, err;
6621         handle_t *handle;
6622
6623         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
6624                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
6625         if (IS_ERR(handle)) {
6626                 /* Release dquot anyway to avoid endless cycle in dqput() */
6627                 dquot_release(dquot);
6628                 return PTR_ERR(handle);
6629         }
6630         ret = dquot_release(dquot);
6631         err = ext4_journal_stop(handle);
6632         if (!ret)
6633                 ret = err;
6634         return ret;
6635 }
6636
6637 static int ext4_mark_dquot_dirty(struct dquot *dquot)
6638 {
6639         struct super_block *sb = dquot->dq_sb;
6640
6641         if (ext4_is_quota_journalled(sb)) {
6642                 dquot_mark_dquot_dirty(dquot);
6643                 return ext4_write_dquot(dquot);
6644         } else {
6645                 return dquot_mark_dquot_dirty(dquot);
6646         }
6647 }
6648
6649 static int ext4_write_info(struct super_block *sb, int type)
6650 {
6651         int ret, err;
6652         handle_t *handle;
6653
6654         /* Data block + inode block */
6655         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
6656         if (IS_ERR(handle))
6657                 return PTR_ERR(handle);
6658         ret = dquot_commit_info(sb, type);
6659         err = ext4_journal_stop(handle);
6660         if (!ret)
6661                 ret = err;
6662         return ret;
6663 }
6664
6665 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
6666 {
6667         struct ext4_inode_info *ei = EXT4_I(inode);
6668
6669         /* The first argument of lockdep_set_subclass has to be
6670          * *exactly* the same as the argument to init_rwsem() --- in
6671          * this case, in init_once() --- or lockdep gets unhappy
6672          * because the name of the lock is set using the
6673          * stringification of the argument to init_rwsem().
6674          */
6675         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
6676         lockdep_set_subclass(&ei->i_data_sem, subclass);
6677 }
6678
6679 /*
6680  * Standard function to be called on quota_on
6681  */
6682 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
6683                          const struct path *path)
6684 {
6685         int err;
6686
6687         if (!test_opt(sb, QUOTA))
6688                 return -EINVAL;
6689
6690         /* Quotafile not on the same filesystem? */
6691         if (path->dentry->d_sb != sb)
6692                 return -EXDEV;
6693
6694         /* Quota already enabled for this file? */
6695         if (IS_NOQUOTA(d_inode(path->dentry)))
6696                 return -EBUSY;
6697
6698         /* Journaling quota? */
6699         if (EXT4_SB(sb)->s_qf_names[type]) {
6700                 /* Quotafile not in fs root? */
6701                 if (path->dentry->d_parent != sb->s_root)
6702                         ext4_msg(sb, KERN_WARNING,
6703                                 "Quota file not on filesystem root. "
6704                                 "Journaled quota will not work");
6705                 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
6706         } else {
6707                 /*
6708                  * Clear the flag just in case mount options changed since
6709                  * last time.
6710                  */
6711                 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
6712         }
6713
6714         /*
6715          * When we journal data on quota file, we have to flush journal to see
6716          * all updates to the file when we bypass pagecache...
6717          */
6718         if (EXT4_SB(sb)->s_journal &&
6719             ext4_should_journal_data(d_inode(path->dentry))) {
6720                 /*
6721                  * We don't need to lock updates but journal_flush() could
6722                  * otherwise be livelocked...
6723                  */
6724                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
6725                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal, 0);
6726                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
6727                 if (err)
6728                         return err;
6729         }
6730
6731         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
6732         err = dquot_quota_on(sb, type, format_id, path);
6733         if (!err) {
6734                 struct inode *inode = d_inode(path->dentry);
6735                 handle_t *handle;
6736
6737                 /*
6738                  * Set inode flags to prevent userspace from messing with quota
6739                  * files. If this fails, we return success anyway since quotas
6740                  * are already enabled and this is not a hard failure.
6741                  */
6742                 inode_lock(inode);
6743                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
6744                 if (IS_ERR(handle))
6745                         goto unlock_inode;
6746                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
6747                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
6748                                 S_NOATIME | S_IMMUTABLE);
6749                 err = ext4_mark_inode_dirty(handle, inode);
6750                 ext4_journal_stop(handle);
6751         unlock_inode:
6752                 inode_unlock(inode);
6753                 if (err)
6754                         dquot_quota_off(sb, type);
6755         }
6756         if (err)
6757                 lockdep_set_quota_inode(path->dentry->d_inode,
6758                                              I_DATA_SEM_NORMAL);
6759         return err;
6760 }
6761
6762 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
6763                              unsigned int flags)
6764 {
6765         int err;
6766         struct inode *qf_inode;
6767         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
6768                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
6769                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
6770                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
6771         };
6772
6773         BUG_ON(!ext4_has_feature_quota(sb));
6774
6775         if (!qf_inums[type])
6776                 return -EPERM;
6777
6778         qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL);
6779         if (IS_ERR(qf_inode)) {
6780                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
6781                 return PTR_ERR(qf_inode);
6782         }
6783
6784         /* Don't account quota for quota files to avoid recursion */
6785         qf_inode->i_flags |= S_NOQUOTA;
6786         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
6787         err = dquot_load_quota_inode(qf_inode, type, format_id, flags);
6788         if (err)
6789                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
6790         iput(qf_inode);
6791
6792         return err;
6793 }
6794
6795 /* Enable usage tracking for all quota types. */
6796 int ext4_enable_quotas(struct super_block *sb)
6797 {
6798         int type, err = 0;
6799         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
6800                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
6801                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
6802                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
6803         };
6804         bool quota_mopt[EXT4_MAXQUOTAS] = {
6805                 test_opt(sb, USRQUOTA),
6806                 test_opt(sb, GRPQUOTA),
6807                 test_opt(sb, PRJQUOTA),
6808         };
6809
6810         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
6811         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
6812                 if (qf_inums[type]) {
6813                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
6814                                 DQUOT_USAGE_ENABLED |
6815                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
6816                         if (err) {
6817                                 ext4_warning(sb,
6818                                         "Failed to enable quota tracking "
6819                                         "(type=%d, err=%d). Please run "
6820                                         "e2fsck to fix.", type, err);
6821                                 for (type--; type >= 0; type--) {
6822                                         struct inode *inode;
6823
6824                                         inode = sb_dqopt(sb)->files[type];
6825                                         if (inode)
6826                                                 inode = igrab(inode);
6827                                         dquot_quota_off(sb, type);
6828                                         if (inode) {
6829                                                 lockdep_set_quota_inode(inode,
6830                                                         I_DATA_SEM_NORMAL);
6831                                                 iput(inode);
6832                                         }
6833                                 }
6834
6835                                 return err;
6836                         }
6837                 }
6838         }
6839         return 0;
6840 }
6841
6842 static int ext4_quota_off(struct super_block *sb, int type)
6843 {
6844         struct inode *inode = sb_dqopt(sb)->files[type];
6845         handle_t *handle;
6846         int err;
6847
6848         /* Force all delayed allocation blocks to be allocated.
6849          * Caller already holds s_umount sem */
6850         if (test_opt(sb, DELALLOC))
6851                 sync_filesystem(sb);
6852
6853         if (!inode || !igrab(inode))
6854                 goto out;
6855
6856         err = dquot_quota_off(sb, type);
6857         if (err || ext4_has_feature_quota(sb))
6858                 goto out_put;
6859
6860         inode_lock(inode);
6861         /*
6862          * Update modification times of quota files when userspace can
6863          * start looking at them. If we fail, we return success anyway since
6864          * this is not a hard failure and quotas are already disabled.
6865          */
6866         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
6867         if (IS_ERR(handle)) {
6868                 err = PTR_ERR(handle);
6869                 goto out_unlock;
6870         }
6871         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
6872         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
6873         inode->i_mtime = inode->i_ctime = current_time(inode);
6874         err = ext4_mark_inode_dirty(handle, inode);
6875         ext4_journal_stop(handle);
6876 out_unlock:
6877         inode_unlock(inode);
6878 out_put:
6879         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
6880         iput(inode);
6881         return err;
6882 out:
6883         return dquot_quota_off(sb, type);
6884 }
6885
6886 /* Read data from quotafile - avoid pagecache and such because we cannot afford
6887  * acquiring the locks... As quota files are never truncated and quota code
6888  * itself serializes the operations (and no one else should touch the files)
6889  * we don't have to be afraid of races */
6890 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
6891                                size_t len, loff_t off)
6892 {
6893         struct inode *inode = sb_dqopt(sb)->files[type];
6894         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
6895         int offset = off & (sb->s_blocksize - 1);
6896         int tocopy;
6897         size_t toread;
6898         struct buffer_head *bh;
6899         loff_t i_size = i_size_read(inode);
6900
6901         if (off > i_size)
6902                 return 0;
6903         if (off+len > i_size)
6904                 len = i_size-off;
6905         toread = len;
6906         while (toread > 0) {
6907                 tocopy = sb->s_blocksize - offset < toread ?
6908                                 sb->s_blocksize - offset : toread;
6909                 bh = ext4_bread(NULL, inode, blk, 0);
6910                 if (IS_ERR(bh))
6911                         return PTR_ERR(bh);
6912                 if (!bh)        /* A hole? */
6913                         memset(data, 0, tocopy);
6914                 else
6915                         memcpy(data, bh->b_data+offset, tocopy);
6916                 brelse(bh);
6917                 offset = 0;
6918                 toread -= tocopy;
6919                 data += tocopy;
6920                 blk++;
6921         }
6922         return len;
6923 }
6924
6925 /* Write to quotafile (we know the transaction is already started and has
6926  * enough credits) */
6927 static ssize_t ext4_quota_write(struct super_block *sb, int type,
6928                                 const char *data, size_t len, loff_t off)
6929 {
6930         struct inode *inode = sb_dqopt(sb)->files[type];
6931         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
6932         int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1);
6933         int retries = 0;
6934         struct buffer_head *bh;
6935         handle_t *handle = journal_current_handle();
6936
6937         if (!handle) {
6938                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
6939                         " cancelled because transaction is not started",
6940                         (unsigned long long)off, (unsigned long long)len);
6941                 return -EIO;
6942         }
6943         /*
6944          * Since we account only one data block in transaction credits,
6945          * then it is impossible to cross a block boundary.
6946          */
6947         if (sb->s_blocksize - offset < len) {
6948                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
6949                         " cancelled because not block aligned",
6950                         (unsigned long long)off, (unsigned long long)len);
6951                 return -EIO;
6952         }
6953
6954         do {
6955                 bh = ext4_bread(handle, inode, blk,
6956                                 EXT4_GET_BLOCKS_CREATE |
6957                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
6958         } while (PTR_ERR(bh) == -ENOSPC &&
6959                  ext4_should_retry_alloc(inode->i_sb, &retries));
6960         if (IS_ERR(bh))
6961                 return PTR_ERR(bh);
6962         if (!bh)
6963                 goto out;
6964         BUFFER_TRACE(bh, "get write access");
6965         err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
6966         if (err) {
6967                 brelse(bh);
6968                 return err;
6969         }
6970         lock_buffer(bh);
6971         memcpy(bh->b_data+offset, data, len);
6972         flush_dcache_page(bh->b_page);
6973         unlock_buffer(bh);
6974         err = ext4_handle_dirty_metadata(handle, NULL, bh);
6975         brelse(bh);
6976 out:
6977         if (inode->i_size < off + len) {
6978                 i_size_write(inode, off + len);
6979                 EXT4_I(inode)->i_disksize = inode->i_size;
6980                 err2 = ext4_mark_inode_dirty(handle, inode);
6981                 if (unlikely(err2 && !err))
6982                         err = err2;
6983         }
6984         return err ? err : len;
6985 }
6986 #endif
6987
6988 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
6989 static inline void register_as_ext2(void)
6990 {
6991         int err = register_filesystem(&ext2_fs_type);
6992         if (err)
6993                 printk(KERN_WARNING
6994                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
6995 }
6996
6997 static inline void unregister_as_ext2(void)
6998 {
6999         unregister_filesystem(&ext2_fs_type);
7000 }
7001
7002 static inline int ext2_feature_set_ok(struct super_block *sb)
7003 {
7004         if (ext4_has_unknown_ext2_incompat_features(sb))
7005                 return 0;
7006         if (sb_rdonly(sb))
7007                 return 1;
7008         if (ext4_has_unknown_ext2_ro_compat_features(sb))
7009                 return 0;
7010         return 1;
7011 }
7012 #else
7013 static inline void register_as_ext2(void) { }
7014 static inline void unregister_as_ext2(void) { }
7015 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
7016 #endif
7017
7018 static inline void register_as_ext3(void)
7019 {
7020         int err = register_filesystem(&ext3_fs_type);
7021         if (err)
7022                 printk(KERN_WARNING
7023                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
7024 }
7025
7026 static inline void unregister_as_ext3(void)
7027 {
7028         unregister_filesystem(&ext3_fs_type);
7029 }
7030
7031 static inline int ext3_feature_set_ok(struct super_block *sb)
7032 {
7033         if (ext4_has_unknown_ext3_incompat_features(sb))
7034                 return 0;
7035         if (!ext4_has_feature_journal(sb))
7036                 return 0;
7037         if (sb_rdonly(sb))
7038                 return 1;
7039         if (ext4_has_unknown_ext3_ro_compat_features(sb))
7040                 return 0;
7041         return 1;
7042 }
7043
7044 static struct file_system_type ext4_fs_type = {
7045         .owner                  = THIS_MODULE,
7046         .name                   = "ext4",
7047         .init_fs_context        = ext4_init_fs_context,
7048         .parameters             = ext4_param_specs,
7049         .kill_sb                = kill_block_super,
7050         .fs_flags               = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
7051 };
7052 MODULE_ALIAS_FS("ext4");
7053
7054 /* Shared across all ext4 file systems */
7055 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
7056
7057 static int __init ext4_init_fs(void)
7058 {
7059         int i, err;
7060
7061         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
7062         ext4_li_info = NULL;
7063
7064         /* Build-time check for flags consistency */
7065         ext4_check_flag_values();
7066
7067         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
7068                 init_waitqueue_head(&ext4__ioend_wq[i]);
7069
7070         err = ext4_init_es();
7071         if (err)
7072                 return err;
7073
7074         err = ext4_init_pending();
7075         if (err)
7076                 goto out7;
7077
7078         err = ext4_init_post_read_processing();
7079         if (err)
7080                 goto out6;
7081
7082         err = ext4_init_pageio();
7083         if (err)
7084                 goto out5;
7085
7086         err = ext4_init_system_zone();
7087         if (err)
7088                 goto out4;
7089
7090         err = ext4_init_sysfs();
7091         if (err)
7092                 goto out3;
7093
7094         err = ext4_init_mballoc();
7095         if (err)
7096                 goto out2;
7097         err = init_inodecache();
7098         if (err)
7099                 goto out1;
7100
7101         err = ext4_fc_init_dentry_cache();
7102         if (err)
7103                 goto out05;
7104
7105         register_as_ext3();
7106         register_as_ext2();
7107         err = register_filesystem(&ext4_fs_type);
7108         if (err)
7109                 goto out;
7110
7111         return 0;
7112 out:
7113         unregister_as_ext2();
7114         unregister_as_ext3();
7115         ext4_fc_destroy_dentry_cache();
7116 out05:
7117         destroy_inodecache();
7118 out1:
7119         ext4_exit_mballoc();
7120 out2:
7121         ext4_exit_sysfs();
7122 out3:
7123         ext4_exit_system_zone();
7124 out4:
7125         ext4_exit_pageio();
7126 out5:
7127         ext4_exit_post_read_processing();
7128 out6:
7129         ext4_exit_pending();
7130 out7:
7131         ext4_exit_es();
7132
7133         return err;
7134 }
7135
7136 static void __exit ext4_exit_fs(void)
7137 {
7138         ext4_destroy_lazyinit_thread();
7139         unregister_as_ext2();
7140         unregister_as_ext3();
7141         unregister_filesystem(&ext4_fs_type);
7142         ext4_fc_destroy_dentry_cache();
7143         destroy_inodecache();
7144         ext4_exit_mballoc();
7145         ext4_exit_sysfs();
7146         ext4_exit_system_zone();
7147         ext4_exit_pageio();
7148         ext4_exit_post_read_processing();
7149         ext4_exit_es();
7150         ext4_exit_pending();
7151 }
7152
7153 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
7154 MODULE_DESCRIPTION("Fourth Extended Filesystem");
7155 MODULE_LICENSE("GPL");
7156 MODULE_SOFTDEP("pre: crc32c");
7157 module_init(ext4_init_fs)
7158 module_exit(ext4_exit_fs)