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/cleancache.h>
  43 #include <linux/uaccess.h>
  44 #include <linux/iversion.h>
  45 #include <linux/unicode.h>
  46 #include <linux/part_stat.h>
  47 #include <linux/kthread.h>
  48 #include <linux/freezer.h>
  49
  50 #include "ext4.h"
  51 #include "ext4_extents.h"       /* Needed for trace points definition */
  52 #include "ext4_jbd2.h"
  53 #include "xattr.h"
  54 #include "acl.h"
  55 #include "mballoc.h"
  56 #include "fsmap.h"
  57
  58 #define CREATE_TRACE_POINTS
  59 #include <trace/events/ext4.h>
  60
  61 static struct ext4_lazy_init *ext4_li_info;
  62 static struct mutex ext4_li_mtx;
  63 static struct ratelimit_state ext4_mount_msg_ratelimit;
  64
  65 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  66                              unsigned long journal_devnum);
  67 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  68 static int ext4_commit_super(struct super_block *sb, int sync);
  69 static void ext4_mark_recovery_complete(struct super_block *sb,
  70                                         struct ext4_super_block *es);
  71 static void ext4_clear_journal_err(struct super_block *sb,
  72                                    struct ext4_super_block *es);
  73 static int ext4_sync_fs(struct super_block *sb, int wait);
  74 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  75 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  76 static int ext4_unfreeze(struct super_block *sb);
  77 static int ext4_freeze(struct super_block *sb);
  78 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  79                        const char *dev_name, void *data);
  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 int ext4_feature_set_ok(struct super_block *sb, int readonly);
  83 static void ext4_destroy_lazyinit_thread(void);
  84 static void ext4_unregister_li_request(struct super_block *sb);
  85 static void ext4_clear_request_list(void);
  86 static struct inode *ext4_get_journal_inode(struct super_block *sb,
  87                                             unsigned int journal_inum);
  88
  89 /*
  90  * Lock ordering
  91  *
  92  * Note the difference between i_mmap_sem (EXT4_I(inode)->i_mmap_sem) and
  93  * i_mmap_rwsem (inode->i_mmap_rwsem)!
  94  *
  95  * page fault path:
  96  * mmap_lock -> sb_start_pagefault -> i_mmap_sem (r) -> transaction start ->
  97  *   page lock -> i_data_sem (rw)
  98  *
  99  * buffered write path:
 100  * sb_start_write -> i_mutex -> mmap_lock
 101  * sb_start_write -> i_mutex -> transaction start -> page lock ->
 102  *   i_data_sem (rw)
 103  *
 104  * truncate:
 105  * sb_start_write -> i_mutex -> i_mmap_sem (w) -> i_mmap_rwsem (w) -> page lock
 106  * sb_start_write -> i_mutex -> i_mmap_sem (w) -> transaction start ->
 107  *   i_data_sem (rw)
 108  *
 109  * direct IO:
 110  * sb_start_write -> i_mutex -> mmap_lock
 111  * sb_start_write -> i_mutex -> transaction start -> i_data_sem (rw)
 112  *
 113  * writepages:
 114  * transaction start -> page lock(s) -> i_data_sem (rw)
 115  */
 116
 117 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
 118 static struct file_system_type ext2_fs_type = {
 119         .owner          = THIS_MODULE,
 120         .name           = "ext2",
 121         .mount          = ext4_mount,
 122         .kill_sb        = kill_block_super,
 123         .fs_flags       = FS_REQUIRES_DEV,
 124 };
 125 MODULE_ALIAS_FS("ext2");
 126 MODULE_ALIAS("ext2");
 127 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
 128 #else
 129 #define IS_EXT2_SB(sb) (0)
 130 #endif
 131
 132
 133 static struct file_system_type ext3_fs_type = {
 134         .owner          = THIS_MODULE,
 135         .name           = "ext3",
 136         .mount          = ext4_mount,
 137         .kill_sb        = kill_block_super,
 138         .fs_flags       = FS_REQUIRES_DEV,
 139 };
 140 MODULE_ALIAS_FS("ext3");
 141 MODULE_ALIAS("ext3");
 142 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 143
 144 /*
 145  * This works like sb_bread() except it uses ERR_PTR for error
 146  * returns.  Currently with sb_bread it's impossible to distinguish
 147  * between ENOMEM and EIO situations (since both result in a NULL
 148  * return.
 149  */
 150 struct buffer_head *
 151 ext4_sb_bread(struct super_block *sb, sector_t block, int op_flags)
 152 {
 153         struct buffer_head *bh = sb_getblk(sb, block);
 154
 155         if (bh == NULL)
 156                 return ERR_PTR(-ENOMEM);
 157         if (ext4_buffer_uptodate(bh))
 158                 return bh;
 159         ll_rw_block(REQ_OP_READ, REQ_META | op_flags, 1, &bh);
 160         wait_on_buffer(bh);
 161         if (buffer_uptodate(bh))
 162                 return bh;
 163         put_bh(bh);
 164         return ERR_PTR(-EIO);
 165 }
 166
 167 static int ext4_verify_csum_type(struct super_block *sb,
 168                                  struct ext4_super_block *es)
 169 {
 170         if (!ext4_has_feature_metadata_csum(sb))
 171                 return 1;
 172
 173         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 174 }
 175
 176 static __le32 ext4_superblock_csum(struct super_block *sb,
 177                                    struct ext4_super_block *es)
 178 {
 179         struct ext4_sb_info *sbi = EXT4_SB(sb);
 180         int offset = offsetof(struct ext4_super_block, s_checksum);
 181         __u32 csum;
 182
 183         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 184
 185         return cpu_to_le32(csum);
 186 }
 187
 188 static int ext4_superblock_csum_verify(struct super_block *sb,
 189                                        struct ext4_super_block *es)
 190 {
 191         if (!ext4_has_metadata_csum(sb))
 192                 return 1;
 193
 194         return es->s_checksum == ext4_superblock_csum(sb, es);
 195 }
 196
 197 void ext4_superblock_csum_set(struct super_block *sb)
 198 {
 199         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 200
 201         if (!ext4_has_metadata_csum(sb))
 202                 return;
 203
 204         es->s_checksum = ext4_superblock_csum(sb, es);
 205 }
 206
 207 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 208                                struct ext4_group_desc *bg)
 209 {
 210         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 211                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 212                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 213 }
 214
 215 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 216                                struct ext4_group_desc *bg)
 217 {
 218         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 219                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 220                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 221 }
 222
 223 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 224                               struct ext4_group_desc *bg)
 225 {
 226         return le32_to_cpu(bg->bg_inode_table_lo) |
 227                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 228                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 229 }
 230
 231 __u32 ext4_free_group_clusters(struct super_block *sb,
 232                                struct ext4_group_desc *bg)
 233 {
 234         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 235                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 236                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 237 }
 238
 239 __u32 ext4_free_inodes_count(struct super_block *sb,
 240                               struct ext4_group_desc *bg)
 241 {
 242         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 243                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 244                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 245 }
 246
 247 __u32 ext4_used_dirs_count(struct super_block *sb,
 248                               struct ext4_group_desc *bg)
 249 {
 250         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 251                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 252                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 253 }
 254
 255 __u32 ext4_itable_unused_count(struct super_block *sb,
 256                               struct ext4_group_desc *bg)
 257 {
 258         return le16_to_cpu(bg->bg_itable_unused_lo) |
 259                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 260                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 261 }
 262
 263 void ext4_block_bitmap_set(struct super_block *sb,
 264                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 265 {
 266         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 267         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 268                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 269 }
 270
 271 void ext4_inode_bitmap_set(struct super_block *sb,
 272                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 273 {
 274         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 275         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 276                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 277 }
 278
 279 void ext4_inode_table_set(struct super_block *sb,
 280                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 281 {
 282         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 283         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 284                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 285 }
 286
 287 void ext4_free_group_clusters_set(struct super_block *sb,
 288                                   struct ext4_group_desc *bg, __u32 count)
 289 {
 290         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 291         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 292                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 293 }
 294
 295 void ext4_free_inodes_set(struct super_block *sb,
 296                           struct ext4_group_desc *bg, __u32 count)
 297 {
 298         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 299         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 300                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 301 }
 302
 303 void ext4_used_dirs_set(struct super_block *sb,
 304                           struct ext4_group_desc *bg, __u32 count)
 305 {
 306         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 307         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 308                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 309 }
 310
 311 void ext4_itable_unused_set(struct super_block *sb,
 312                           struct ext4_group_desc *bg, __u32 count)
 313 {
 314         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 315         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 316                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 317 }
 318
 319 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi)
 320 {
 321         time64_t now = ktime_get_real_seconds();
 322
 323         now = clamp_val(now, 0, (1ull << 40) - 1);
 324
 325         *lo = cpu_to_le32(lower_32_bits(now));
 326         *hi = upper_32_bits(now);
 327 }
 328
 329 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi)
 330 {
 331         return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo);
 332 }
 333 #define ext4_update_tstamp(es, tstamp) \
 334         __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
 335 #define ext4_get_tstamp(es, tstamp) \
 336         __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
 337
 338 static void __save_error_info(struct super_block *sb, int error,
 339                               __u32 ino, __u64 block,
 340                               const char *func, unsigned int line)
 341 {
 342         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 343         int err;
 344
 345         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 346         if (bdev_read_only(sb->s_bdev))
 347                 return;
 348         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 349         ext4_update_tstamp(es, s_last_error_time);
 350         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 351         es->s_last_error_line = cpu_to_le32(line);
 352         es->s_last_error_ino = cpu_to_le32(ino);
 353         es->s_last_error_block = cpu_to_le64(block);
 354         switch (error) {
 355         case EIO:
 356                 err = EXT4_ERR_EIO;
 357                 break;
 358         case ENOMEM:
 359                 err = EXT4_ERR_ENOMEM;
 360                 break;
 361         case EFSBADCRC:
 362                 err = EXT4_ERR_EFSBADCRC;
 363                 break;
 364         case 0:
 365         case EFSCORRUPTED:
 366                 err = EXT4_ERR_EFSCORRUPTED;
 367                 break;
 368         case ENOSPC:
 369                 err = EXT4_ERR_ENOSPC;
 370                 break;
 371         case ENOKEY:
 372                 err = EXT4_ERR_ENOKEY;
 373                 break;
 374         case EROFS:
 375                 err = EXT4_ERR_EROFS;
 376                 break;
 377         case EFBIG:
 378                 err = EXT4_ERR_EFBIG;
 379                 break;
 380         case EEXIST:
 381                 err = EXT4_ERR_EEXIST;
 382                 break;
 383         case ERANGE:
 384                 err = EXT4_ERR_ERANGE;
 385                 break;
 386         case EOVERFLOW:
 387                 err = EXT4_ERR_EOVERFLOW;
 388                 break;
 389         case EBUSY:
 390                 err = EXT4_ERR_EBUSY;
 391                 break;
 392         case ENOTDIR:
 393                 err = EXT4_ERR_ENOTDIR;
 394                 break;
 395         case ENOTEMPTY:
 396                 err = EXT4_ERR_ENOTEMPTY;
 397                 break;
 398         case ESHUTDOWN:
 399                 err = EXT4_ERR_ESHUTDOWN;
 400                 break;
 401         case EFAULT:
 402                 err = EXT4_ERR_EFAULT;
 403                 break;
 404         default:
 405                 err = EXT4_ERR_UNKNOWN;
 406         }
 407         es->s_last_error_errcode = err;
 408         if (!es->s_first_error_time) {
 409                 es->s_first_error_time = es->s_last_error_time;
 410                 es->s_first_error_time_hi = es->s_last_error_time_hi;
 411                 strncpy(es->s_first_error_func, func,
 412                         sizeof(es->s_first_error_func));
 413                 es->s_first_error_line = cpu_to_le32(line);
 414                 es->s_first_error_ino = es->s_last_error_ino;
 415                 es->s_first_error_block = es->s_last_error_block;
 416                 es->s_first_error_errcode = es->s_last_error_errcode;
 417         }
 418         /*
 419          * Start the daily error reporting function if it hasn't been
 420          * started already
 421          */
 422         if (!es->s_error_count)
 423                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 424         le32_add_cpu(&es->s_error_count, 1);
 425 }
 426
 427 static void save_error_info(struct super_block *sb, int error,
 428                             __u32 ino, __u64 block,
 429                             const char *func, unsigned int line)
 430 {
 431         __save_error_info(sb, error, ino, block, func, line);
 432         if (!bdev_read_only(sb->s_bdev))
 433                 ext4_commit_super(sb, 1);
 434 }
 435
 436 /*
 437  * The del_gendisk() function uninitializes the disk-specific data
 438  * structures, including the bdi structure, without telling anyone
 439  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 440  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 441  * This is a kludge to prevent these oops until we can put in a proper
 442  * hook in del_gendisk() to inform the VFS and file system layers.
 443  */
 444 static int block_device_ejected(struct super_block *sb)
 445 {
 446         struct inode *bd_inode = sb->s_bdev->bd_inode;
 447         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 448
 449         return bdi->dev == NULL;
 450 }
 451
 452 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 453 {
 454         struct super_block              *sb = journal->j_private;
 455         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 456         int                             error = is_journal_aborted(journal);
 457         struct ext4_journal_cb_entry    *jce;
 458
 459         BUG_ON(txn->t_state == T_FINISHED);
 460
 461         ext4_process_freed_data(sb, txn->t_tid);
 462
 463         spin_lock(&sbi->s_md_lock);
 464         while (!list_empty(&txn->t_private_list)) {
 465                 jce = list_entry(txn->t_private_list.next,
 466                                  struct ext4_journal_cb_entry, jce_list);
 467                 list_del_init(&jce->jce_list);
 468                 spin_unlock(&sbi->s_md_lock);
 469                 jce->jce_func(sb, jce, error);
 470                 spin_lock(&sbi->s_md_lock);
 471         }
 472         spin_unlock(&sbi->s_md_lock);
 473 }
 474
 475 static bool system_going_down(void)
 476 {
 477         return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF
 478                 || system_state == SYSTEM_RESTART;
 479 }
 480
 481 /* Deal with the reporting of failure conditions on a filesystem such as
 482  * inconsistencies detected or read IO failures.
 483  *
 484  * On ext2, we can store the error state of the filesystem in the
 485  * superblock.  That is not possible on ext4, because we may have other
 486  * write ordering constraints on the superblock which prevent us from
 487  * writing it out straight away; and given that the journal is about to
 488  * be aborted, we can't rely on the current, or future, transactions to
 489  * write out the superblock safely.
 490  *
 491  * We'll just use the jbd2_journal_abort() error code to record an error in
 492  * the journal instead.  On recovery, the journal will complain about
 493  * that error until we've noted it down and cleared it.
 494  */
 495
 496 static void ext4_handle_error(struct super_block *sb)
 497 {
 498         if (test_opt(sb, WARN_ON_ERROR))
 499                 WARN_ON_ONCE(1);
 500
 501         if (sb_rdonly(sb))
 502                 return;
 503
 504         if (!test_opt(sb, ERRORS_CONT)) {
 505                 journal_t *journal = EXT4_SB(sb)->s_journal;
 506
 507                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 508                 if (journal)
 509                         jbd2_journal_abort(journal, -EIO);
 510         }
 511         /*
 512          * We force ERRORS_RO behavior when system is rebooting. Otherwise we
 513          * could panic during 'reboot -f' as the underlying device got already
 514          * disabled.
 515          */
 516         if (test_opt(sb, ERRORS_RO) || system_going_down()) {
 517                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 518                 /*
 519                  * Make sure updated value of ->s_mount_flags will be visible
 520                  * before ->s_flags update
 521                  */
 522                 smp_wmb();
 523                 sb->s_flags |= SB_RDONLY;
 524         } else if (test_opt(sb, ERRORS_PANIC)) {
 525                 panic("EXT4-fs (device %s): panic forced after error\n",
 526                         sb->s_id);
 527         }
 528 }
 529
 530 #define ext4_error_ratelimit(sb)                                        \
 531                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 532                              "EXT4-fs error")
 533
 534 void __ext4_error(struct super_block *sb, const char *function,
 535                   unsigned int line, int error, __u64 block,
 536                   const char *fmt, ...)
 537 {
 538         struct va_format vaf;
 539         va_list args;
 540
 541         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 542                 return;
 543
 544         trace_ext4_error(sb, function, line);
 545         if (ext4_error_ratelimit(sb)) {
 546                 va_start(args, fmt);
 547                 vaf.fmt = fmt;
 548                 vaf.va = &args;
 549                 printk(KERN_CRIT
 550                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 551                        sb->s_id, function, line, current->comm, &vaf);
 552                 va_end(args);
 553         }
 554         save_error_info(sb, error, 0, block, function, line);
 555         ext4_handle_error(sb);
 556 }
 557
 558 void __ext4_error_inode(struct inode *inode, const char *function,
 559                         unsigned int line, ext4_fsblk_t block, int error,
 560                         const char *fmt, ...)
 561 {
 562         va_list args;
 563         struct va_format vaf;
 564
 565         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 566                 return;
 567
 568         trace_ext4_error(inode->i_sb, function, line);
 569         if (ext4_error_ratelimit(inode->i_sb)) {
 570                 va_start(args, fmt);
 571                 vaf.fmt = fmt;
 572                 vaf.va = &args;
 573                 if (block)
 574                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 575                                "inode #%lu: block %llu: comm %s: %pV\n",
 576                                inode->i_sb->s_id, function, line, inode->i_ino,
 577                                block, current->comm, &vaf);
 578                 else
 579                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 580                                "inode #%lu: comm %s: %pV\n",
 581                                inode->i_sb->s_id, function, line, inode->i_ino,
 582                                current->comm, &vaf);
 583                 va_end(args);
 584         }
 585         save_error_info(inode->i_sb, error, inode->i_ino, block,
 586                         function, line);
 587         ext4_handle_error(inode->i_sb);
 588 }
 589
 590 void __ext4_error_file(struct file *file, const char *function,
 591                        unsigned int line, ext4_fsblk_t block,
 592                        const char *fmt, ...)
 593 {
 594         va_list args;
 595         struct va_format vaf;
 596         struct inode *inode = file_inode(file);
 597         char pathname[80], *path;
 598
 599         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 600                 return;
 601
 602         trace_ext4_error(inode->i_sb, function, line);
 603         if (ext4_error_ratelimit(inode->i_sb)) {
 604                 path = file_path(file, pathname, sizeof(pathname));
 605                 if (IS_ERR(path))
 606                         path = "(unknown)";
 607                 va_start(args, fmt);
 608                 vaf.fmt = fmt;
 609                 vaf.va = &args;
 610                 if (block)
 611                         printk(KERN_CRIT
 612                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 613                                "block %llu: comm %s: path %s: %pV\n",
 614                                inode->i_sb->s_id, function, line, inode->i_ino,
 615                                block, current->comm, path, &vaf);
 616                 else
 617                         printk(KERN_CRIT
 618                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 619                                "comm %s: path %s: %pV\n",
 620                                inode->i_sb->s_id, function, line, inode->i_ino,
 621                                current->comm, path, &vaf);
 622                 va_end(args);
 623         }
 624         save_error_info(inode->i_sb, EFSCORRUPTED, inode->i_ino, block,
 625                         function, line);
 626         ext4_handle_error(inode->i_sb);
 627 }
 628
 629 const char *ext4_decode_error(struct super_block *sb, int errno,
 630                               char nbuf[16])
 631 {
 632         char *errstr = NULL;
 633
 634         switch (errno) {
 635         case -EFSCORRUPTED:
 636                 errstr = "Corrupt filesystem";
 637                 break;
 638         case -EFSBADCRC:
 639                 errstr = "Filesystem failed CRC";
 640                 break;
 641         case -EIO:
 642                 errstr = "IO failure";
 643                 break;
 644         case -ENOMEM:
 645                 errstr = "Out of memory";
 646                 break;
 647         case -EROFS:
 648                 if (!sb || (EXT4_SB(sb)->s_journal &&
 649                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 650                         errstr = "Journal has aborted";
 651                 else
 652                         errstr = "Readonly filesystem";
 653                 break;
 654         default:
 655                 /* If the caller passed in an extra buffer for unknown
 656                  * errors, textualise them now.  Else we just return
 657                  * NULL. */
 658                 if (nbuf) {
 659                         /* Check for truncated error codes... */
 660                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 661                                 errstr = nbuf;
 662                 }
 663                 break;
 664         }
 665
 666         return errstr;
 667 }
 668
 669 /* __ext4_std_error decodes expected errors from journaling functions
 670  * automatically and invokes the appropriate error response.  */
 671
 672 void __ext4_std_error(struct super_block *sb, const char *function,
 673                       unsigned int line, int errno)
 674 {
 675         char nbuf[16];
 676         const char *errstr;
 677
 678         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 679                 return;
 680
 681         /* Special case: if the error is EROFS, and we're not already
 682          * inside a transaction, then there's really no point in logging
 683          * an error. */
 684         if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb))
 685                 return;
 686
 687         if (ext4_error_ratelimit(sb)) {
 688                 errstr = ext4_decode_error(sb, errno, nbuf);
 689                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 690                        sb->s_id, function, line, errstr);
 691         }
 692
 693         save_error_info(sb, -errno, 0, 0, function, line);
 694         ext4_handle_error(sb);
 695 }
 696
 697 /*
 698  * ext4_abort is a much stronger failure handler than ext4_error.  The
 699  * abort function may be used to deal with unrecoverable failures such
 700  * as journal IO errors or ENOMEM at a critical moment in log management.
 701  *
 702  * We unconditionally force the filesystem into an ABORT|READONLY state,
 703  * unless the error response on the fs has been set to panic in which
 704  * case we take the easy way out and panic immediately.
 705  */
 706
 707 void __ext4_abort(struct super_block *sb, const char *function,
 708                   unsigned int line, int error, const char *fmt, ...)
 709 {
 710         struct va_format vaf;
 711         va_list args;
 712
 713         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 714                 return;
 715
 716         save_error_info(sb, error, 0, 0, function, line);
 717         va_start(args, fmt);
 718         vaf.fmt = fmt;
 719         vaf.va = &args;
 720         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: %pV\n",
 721                sb->s_id, function, line, &vaf);
 722         va_end(args);
 723
 724         if (sb_rdonly(sb) == 0) {
 725                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 726                 if (EXT4_SB(sb)->s_journal)
 727                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 728
 729                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 730                 /*
 731                  * Make sure updated value of ->s_mount_flags will be visible
 732                  * before ->s_flags update
 733                  */
 734                 smp_wmb();
 735                 sb->s_flags |= SB_RDONLY;
 736         }
 737         if (test_opt(sb, ERRORS_PANIC) && !system_going_down())
 738                 panic("EXT4-fs panic from previous error\n");
 739 }
 740
 741 void __ext4_msg(struct super_block *sb,
 742                 const char *prefix, const char *fmt, ...)
 743 {
 744         struct va_format vaf;
 745         va_list args;
 746
 747         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 748                 return;
 749
 750         va_start(args, fmt);
 751         vaf.fmt = fmt;
 752         vaf.va = &args;
 753         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 754         va_end(args);
 755 }
 756
 757 #define ext4_warning_ratelimit(sb)                                      \
 758                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
 759                              "EXT4-fs warning")
 760
 761 void __ext4_warning(struct super_block *sb, const char *function,
 762                     unsigned int line, const char *fmt, ...)
 763 {
 764         struct va_format vaf;
 765         va_list args;
 766
 767         if (!ext4_warning_ratelimit(sb))
 768                 return;
 769
 770         va_start(args, fmt);
 771         vaf.fmt = fmt;
 772         vaf.va = &args;
 773         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 774                sb->s_id, function, line, &vaf);
 775         va_end(args);
 776 }
 777
 778 void __ext4_warning_inode(const struct inode *inode, const char *function,
 779                           unsigned int line, const char *fmt, ...)
 780 {
 781         struct va_format vaf;
 782         va_list args;
 783
 784         if (!ext4_warning_ratelimit(inode->i_sb))
 785                 return;
 786
 787         va_start(args, fmt);
 788         vaf.fmt = fmt;
 789         vaf.va = &args;
 790         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 791                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 792                function, line, inode->i_ino, current->comm, &vaf);
 793         va_end(args);
 794 }
 795
 796 void __ext4_grp_locked_error(const char *function, unsigned int line,
 797                              struct super_block *sb, ext4_group_t grp,
 798                              unsigned long ino, ext4_fsblk_t block,
 799                              const char *fmt, ...)
 800 __releases(bitlock)
 801 __acquires(bitlock)
 802 {
 803         struct va_format vaf;
 804         va_list args;
 805
 806         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 807                 return;
 808
 809         trace_ext4_error(sb, function, line);
 810         __save_error_info(sb, EFSCORRUPTED, ino, block, function, line);
 811
 812         if (ext4_error_ratelimit(sb)) {
 813                 va_start(args, fmt);
 814                 vaf.fmt = fmt;
 815                 vaf.va = &args;
 816                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 817                        sb->s_id, function, line, grp);
 818                 if (ino)
 819                         printk(KERN_CONT "inode %lu: ", ino);
 820                 if (block)
 821                         printk(KERN_CONT "block %llu:",
 822                                (unsigned long long) block);
 823                 printk(KERN_CONT "%pV\n", &vaf);
 824                 va_end(args);
 825         }
 826
 827         if (test_opt(sb, WARN_ON_ERROR))
 828                 WARN_ON_ONCE(1);
 829
 830         if (test_opt(sb, ERRORS_CONT)) {
 831                 ext4_commit_super(sb, 0);
 832                 return;
 833         }
 834
 835         ext4_unlock_group(sb, grp);
 836         ext4_commit_super(sb, 1);
 837         ext4_handle_error(sb);
 838         /*
 839          * We only get here in the ERRORS_RO case; relocking the group
 840          * may be dangerous, but nothing bad will happen since the
 841          * filesystem will have already been marked read/only and the
 842          * journal has been aborted.  We return 1 as a hint to callers
 843          * who might what to use the return value from
 844          * ext4_grp_locked_error() to distinguish between the
 845          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 846          * aggressively from the ext4 function in question, with a
 847          * more appropriate error code.
 848          */
 849         ext4_lock_group(sb, grp);
 850         return;
 851 }
 852
 853 void ext4_mark_group_bitmap_corrupted(struct super_block *sb,
 854                                      ext4_group_t group,
 855                                      unsigned int flags)
 856 {
 857         struct ext4_sb_info *sbi = EXT4_SB(sb);
 858         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
 859         struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
 860         int ret;
 861
 862         if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) {
 863                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT,
 864                                             &grp->bb_state);
 865                 if (!ret)
 866                         percpu_counter_sub(&sbi->s_freeclusters_counter,
 867                                            grp->bb_free);
 868         }
 869
 870         if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) {
 871                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT,
 872                                             &grp->bb_state);
 873                 if (!ret && gdp) {
 874                         int count;
 875
 876                         count = ext4_free_inodes_count(sb, gdp);
 877                         percpu_counter_sub(&sbi->s_freeinodes_counter,
 878                                            count);
 879                 }
 880         }
 881 }
 882
 883 void ext4_update_dynamic_rev(struct super_block *sb)
 884 {
 885         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 886
 887         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 888                 return;
 889
 890         ext4_warning(sb,
 891                      "updating to rev %d because of new feature flag, "
 892                      "running e2fsck is recommended",
 893                      EXT4_DYNAMIC_REV);
 894
 895         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 896         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 897         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 898         /* leave es->s_feature_*compat flags alone */
 899         /* es->s_uuid will be set by e2fsck if empty */
 900
 901         /*
 902          * The rest of the superblock fields should be zero, and if not it
 903          * means they are likely already in use, so leave them alone.  We
 904          * can leave it up to e2fsck to clean up any inconsistencies there.
 905          */
 906 }
 907
 908 /*
 909  * Open the external journal device
 910  */
 911 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 912 {
 913         struct block_device *bdev;
 914
 915         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 916         if (IS_ERR(bdev))
 917                 goto fail;
 918         return bdev;
 919
 920 fail:
 921         ext4_msg(sb, KERN_ERR,
 922                  "failed to open journal device unknown-block(%u,%u) %ld",
 923                  MAJOR(dev), MINOR(dev), PTR_ERR(bdev));
 924         return NULL;
 925 }
 926
 927 /*
 928  * Release the journal device
 929  */
 930 static void ext4_blkdev_put(struct block_device *bdev)
 931 {
 932         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 933 }
 934
 935 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 936 {
 937         struct block_device *bdev;
 938         bdev = sbi->journal_bdev;
 939         if (bdev) {
 940                 ext4_blkdev_put(bdev);
 941                 sbi->journal_bdev = NULL;
 942         }
 943 }
 944
 945 static inline struct inode *orphan_list_entry(struct list_head *l)
 946 {
 947         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 948 }
 949
 950 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 951 {
 952         struct list_head *l;
 953
 954         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 955                  le32_to_cpu(sbi->s_es->s_last_orphan));
 956
 957         printk(KERN_ERR "sb_info orphan list:\n");
 958         list_for_each(l, &sbi->s_orphan) {
 959                 struct inode *inode = orphan_list_entry(l);
 960                 printk(KERN_ERR "  "
 961                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 962                        inode->i_sb->s_id, inode->i_ino, inode,
 963                        inode->i_mode, inode->i_nlink,
 964                        NEXT_ORPHAN(inode));
 965         }
 966 }
 967
 968 #ifdef CONFIG_QUOTA
 969 static int ext4_quota_off(struct super_block *sb, int type);
 970
 971 static inline void ext4_quota_off_umount(struct super_block *sb)
 972 {
 973         int type;
 974
 975         /* Use our quota_off function to clear inode flags etc. */
 976         for (type = 0; type < EXT4_MAXQUOTAS; type++)
 977                 ext4_quota_off(sb, type);
 978 }
 979
 980 /*
 981  * This is a helper function which is used in the mount/remount
 982  * codepaths (which holds s_umount) to fetch the quota file name.
 983  */
 984 static inline char *get_qf_name(struct super_block *sb,
 985                                 struct ext4_sb_info *sbi,
 986                                 int type)
 987 {
 988         return rcu_dereference_protected(sbi->s_qf_names[type],
 989                                          lockdep_is_held(&sb->s_umount));
 990 }
 991 #else
 992 static inline void ext4_quota_off_umount(struct super_block *sb)
 993 {
 994 }
 995 #endif
 996
 997 static void ext4_put_super(struct super_block *sb)
 998 {
 999         struct ext4_sb_info *sbi = EXT4_SB(sb);
1000         struct ext4_super_block *es = sbi->s_es;
1001         struct buffer_head **group_desc;
1002         struct flex_groups **flex_groups;
1003         int aborted = 0;
1004         int i, err;
1005
1006         ext4_unregister_li_request(sb);
1007         ext4_quota_off_umount(sb);
1008
1009         destroy_workqueue(sbi->rsv_conversion_wq);
1010
1011         /*
1012          * Unregister sysfs before destroying jbd2 journal.
1013          * Since we could still access attr_journal_task attribute via sysfs
1014          * path which could have sbi->s_journal->j_task as NULL
1015          */
1016         ext4_unregister_sysfs(sb);
1017
1018         if (sbi->s_journal) {
1019                 aborted = is_journal_aborted(sbi->s_journal);
1020                 err = jbd2_journal_destroy(sbi->s_journal);
1021                 sbi->s_journal = NULL;
1022                 if ((err < 0) && !aborted) {
1023                         ext4_abort(sb, -err, "Couldn't clean up the journal");
1024                 }
1025         }
1026
1027         ext4_es_unregister_shrinker(sbi);
1028         del_timer_sync(&sbi->s_err_report);
1029         ext4_release_system_zone(sb);
1030         ext4_mb_release(sb);
1031         ext4_ext_release(sb);
1032
1033         if (!sb_rdonly(sb) && !aborted) {
1034                 ext4_clear_feature_journal_needs_recovery(sb);
1035                 es->s_state = cpu_to_le16(sbi->s_mount_state);
1036         }
1037         if (!sb_rdonly(sb))
1038                 ext4_commit_super(sb, 1);
1039
1040         rcu_read_lock();
1041         group_desc = rcu_dereference(sbi->s_group_desc);
1042         for (i = 0; i < sbi->s_gdb_count; i++)
1043                 brelse(group_desc[i]);
1044         kvfree(group_desc);
1045         flex_groups = rcu_dereference(sbi->s_flex_groups);
1046         if (flex_groups) {
1047                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
1048                         kvfree(flex_groups[i]);
1049                 kvfree(flex_groups);
1050         }
1051         rcu_read_unlock();
1052         percpu_counter_destroy(&sbi->s_freeclusters_counter);
1053         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1054         percpu_counter_destroy(&sbi->s_dirs_counter);
1055         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
1056         percpu_free_rwsem(&sbi->s_writepages_rwsem);
1057 #ifdef CONFIG_QUOTA
1058         for (i = 0; i < EXT4_MAXQUOTAS; i++)
1059                 kfree(get_qf_name(sb, sbi, i));
1060 #endif
1061
1062         /* Debugging code just in case the in-memory inode orphan list
1063          * isn't empty.  The on-disk one can be non-empty if we've
1064          * detected an error and taken the fs readonly, but the
1065          * in-memory list had better be clean by this point. */
1066         if (!list_empty(&sbi->s_orphan))
1067                 dump_orphan_list(sb, sbi);
1068         J_ASSERT(list_empty(&sbi->s_orphan));
1069
1070         sync_blockdev(sb->s_bdev);
1071         invalidate_bdev(sb->s_bdev);
1072         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
1073                 /*
1074                  * Invalidate the journal device's buffers.  We don't want them
1075                  * floating about in memory - the physical journal device may
1076                  * hotswapped, and it breaks the `ro-after' testing code.
1077                  */
1078                 sync_blockdev(sbi->journal_bdev);
1079                 invalidate_bdev(sbi->journal_bdev);
1080                 ext4_blkdev_remove(sbi);
1081         }
1082
1083         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
1084         sbi->s_ea_inode_cache = NULL;
1085
1086         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
1087         sbi->s_ea_block_cache = NULL;
1088
1089         if (sbi->s_mmp_tsk)
1090                 kthread_stop(sbi->s_mmp_tsk);
1091         brelse(sbi->s_sbh);
1092         sb->s_fs_info = NULL;
1093         /*
1094          * Now that we are completely done shutting down the
1095          * superblock, we need to actually destroy the kobject.
1096          */
1097         kobject_put(&sbi->s_kobj);
1098         wait_for_completion(&sbi->s_kobj_unregister);
1099         if (sbi->s_chksum_driver)
1100                 crypto_free_shash(sbi->s_chksum_driver);
1101         kfree(sbi->s_blockgroup_lock);
1102         fs_put_dax(sbi->s_daxdev);
1103         fscrypt_free_dummy_context(&sbi->s_dummy_enc_ctx);
1104 #ifdef CONFIG_UNICODE
1105         utf8_unload(sbi->s_encoding);
1106 #endif
1107         kfree(sbi);
1108 }
1109
1110 static struct kmem_cache *ext4_inode_cachep;
1111
1112 /*
1113  * Called inside transaction, so use GFP_NOFS
1114  */
1115 static struct inode *ext4_alloc_inode(struct super_block *sb)
1116 {
1117         struct ext4_inode_info *ei;
1118
1119         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
1120         if (!ei)
1121                 return NULL;
1122
1123         inode_set_iversion(&ei->vfs_inode, 1);
1124         spin_lock_init(&ei->i_raw_lock);
1125         INIT_LIST_HEAD(&ei->i_prealloc_list);
1126         spin_lock_init(&ei->i_prealloc_lock);
1127         ext4_es_init_tree(&ei->i_es_tree);
1128         rwlock_init(&ei->i_es_lock);
1129         INIT_LIST_HEAD(&ei->i_es_list);
1130         ei->i_es_all_nr = 0;
1131         ei->i_es_shk_nr = 0;
1132         ei->i_es_shrink_lblk = 0;
1133         ei->i_reserved_data_blocks = 0;
1134         spin_lock_init(&(ei->i_block_reservation_lock));
1135         ext4_init_pending_tree(&ei->i_pending_tree);
1136 #ifdef CONFIG_QUOTA
1137         ei->i_reserved_quota = 0;
1138         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
1139 #endif
1140         ei->jinode = NULL;
1141         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
1142         spin_lock_init(&ei->i_completed_io_lock);
1143         ei->i_sync_tid = 0;
1144         ei->i_datasync_tid = 0;
1145         atomic_set(&ei->i_unwritten, 0);
1146         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
1147         return &ei->vfs_inode;
1148 }
1149
1150 static int ext4_drop_inode(struct inode *inode)
1151 {
1152         int drop = generic_drop_inode(inode);
1153
1154         if (!drop)
1155                 drop = fscrypt_drop_inode(inode);
1156
1157         trace_ext4_drop_inode(inode, drop);
1158         return drop;
1159 }
1160
1161 static void ext4_free_in_core_inode(struct inode *inode)
1162 {
1163         fscrypt_free_inode(inode);
1164         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1165 }
1166
1167 static void ext4_destroy_inode(struct inode *inode)
1168 {
1169         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1170                 ext4_msg(inode->i_sb, KERN_ERR,
1171                          "Inode %lu (%p): orphan list check failed!",
1172                          inode->i_ino, EXT4_I(inode));
1173                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1174                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1175                                 true);
1176                 dump_stack();
1177         }
1178 }
1179
1180 static void init_once(void *foo)
1181 {
1182         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1183
1184         INIT_LIST_HEAD(&ei->i_orphan);
1185         init_rwsem(&ei->xattr_sem);
1186         init_rwsem(&ei->i_data_sem);
1187         init_rwsem(&ei->i_mmap_sem);
1188         inode_init_once(&ei->vfs_inode);
1189 }
1190
1191 static int __init init_inodecache(void)
1192 {
1193         ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache",
1194                                 sizeof(struct ext4_inode_info), 0,
1195                                 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
1196                                         SLAB_ACCOUNT),
1197                                 offsetof(struct ext4_inode_info, i_data),
1198                                 sizeof_field(struct ext4_inode_info, i_data),
1199                                 init_once);
1200         if (ext4_inode_cachep == NULL)
1201                 return -ENOMEM;
1202         return 0;
1203 }
1204
1205 static void destroy_inodecache(void)
1206 {
1207         /*
1208          * Make sure all delayed rcu free inodes are flushed before we
1209          * destroy cache.
1210          */
1211         rcu_barrier();
1212         kmem_cache_destroy(ext4_inode_cachep);
1213 }
1214
1215 void ext4_clear_inode(struct inode *inode)
1216 {
1217         invalidate_inode_buffers(inode);
1218         clear_inode(inode);
1219         ext4_discard_preallocations(inode);
1220         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1221         dquot_drop(inode);
1222         if (EXT4_I(inode)->jinode) {
1223                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1224                                                EXT4_I(inode)->jinode);
1225                 jbd2_free_inode(EXT4_I(inode)->jinode);
1226                 EXT4_I(inode)->jinode = NULL;
1227         }
1228         fscrypt_put_encryption_info(inode);
1229         fsverity_cleanup_inode(inode);
1230 }
1231
1232 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1233                                         u64 ino, u32 generation)
1234 {
1235         struct inode *inode;
1236
1237         /*
1238          * Currently we don't know the generation for parent directory, so
1239          * a generation of 0 means "accept any"
1240          */
1241         inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE);
1242         if (IS_ERR(inode))
1243                 return ERR_CAST(inode);
1244         if (generation && inode->i_generation != generation) {
1245                 iput(inode);
1246                 return ERR_PTR(-ESTALE);
1247         }
1248
1249         return inode;
1250 }
1251
1252 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1253                                         int fh_len, int fh_type)
1254 {
1255         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1256                                     ext4_nfs_get_inode);
1257 }
1258
1259 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1260                                         int fh_len, int fh_type)
1261 {
1262         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1263                                     ext4_nfs_get_inode);
1264 }
1265
1266 static int ext4_nfs_commit_metadata(struct inode *inode)
1267 {
1268         struct writeback_control wbc = {
1269                 .sync_mode = WB_SYNC_ALL
1270         };
1271
1272         trace_ext4_nfs_commit_metadata(inode);
1273         return ext4_write_inode(inode, &wbc);
1274 }
1275
1276 /*
1277  * Try to release metadata pages (indirect blocks, directories) which are
1278  * mapped via the block device.  Since these pages could have journal heads
1279  * which would prevent try_to_free_buffers() from freeing them, we must use
1280  * jbd2 layer's try_to_free_buffers() function to release them.
1281  */
1282 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1283                                  gfp_t wait)
1284 {
1285         journal_t *journal = EXT4_SB(sb)->s_journal;
1286
1287         WARN_ON(PageChecked(page));
1288         if (!page_has_buffers(page))
1289                 return 0;
1290         if (journal)
1291                 return jbd2_journal_try_to_free_buffers(journal, page,
1292                                                 wait & ~__GFP_DIRECT_RECLAIM);
1293         return try_to_free_buffers(page);
1294 }
1295
1296 #ifdef CONFIG_FS_ENCRYPTION
1297 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1298 {
1299         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1300                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1301 }
1302
1303 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1304                                                         void *fs_data)
1305 {
1306         handle_t *handle = fs_data;
1307         int res, res2, credits, retries = 0;
1308
1309         /*
1310          * Encrypting the root directory is not allowed because e2fsck expects
1311          * lost+found to exist and be unencrypted, and encrypting the root
1312          * directory would imply encrypting the lost+found directory as well as
1313          * the filename "lost+found" itself.
1314          */
1315         if (inode->i_ino == EXT4_ROOT_INO)
1316                 return -EPERM;
1317
1318         if (WARN_ON_ONCE(IS_DAX(inode) && i_size_read(inode)))
1319                 return -EINVAL;
1320
1321         if (ext4_test_inode_flag(inode, EXT4_INODE_DAX))
1322                 return -EOPNOTSUPP;
1323
1324         res = ext4_convert_inline_data(inode);
1325         if (res)
1326                 return res;
1327
1328         /*
1329          * If a journal handle was specified, then the encryption context is
1330          * being set on a new inode via inheritance and is part of a larger
1331          * transaction to create the inode.  Otherwise the encryption context is
1332          * being set on an existing inode in its own transaction.  Only in the
1333          * latter case should the "retry on ENOSPC" logic be used.
1334          */
1335
1336         if (handle) {
1337                 res = ext4_xattr_set_handle(handle, inode,
1338                                             EXT4_XATTR_INDEX_ENCRYPTION,
1339                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1340                                             ctx, len, 0);
1341                 if (!res) {
1342                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1343                         ext4_clear_inode_state(inode,
1344                                         EXT4_STATE_MAY_INLINE_DATA);
1345                         /*
1346                          * Update inode->i_flags - S_ENCRYPTED will be enabled,
1347                          * S_DAX may be disabled
1348                          */
1349                         ext4_set_inode_flags(inode, false);
1350                 }
1351                 return res;
1352         }
1353
1354         res = dquot_initialize(inode);
1355         if (res)
1356                 return res;
1357 retry:
1358         res = ext4_xattr_set_credits(inode, len, false /* is_create */,
1359                                      &credits);
1360         if (res)
1361                 return res;
1362
1363         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
1364         if (IS_ERR(handle))
1365                 return PTR_ERR(handle);
1366
1367         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1368                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1369                                     ctx, len, 0);
1370         if (!res) {
1371                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1372                 /*
1373                  * Update inode->i_flags - S_ENCRYPTED will be enabled,
1374                  * S_DAX may be disabled
1375                  */
1376                 ext4_set_inode_flags(inode, false);
1377                 res = ext4_mark_inode_dirty(handle, inode);
1378                 if (res)
1379                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1380         }
1381         res2 = ext4_journal_stop(handle);
1382
1383         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1384                 goto retry;
1385         if (!res)
1386                 res = res2;
1387         return res;
1388 }
1389
1390 static const union fscrypt_context *
1391 ext4_get_dummy_context(struct super_block *sb)
1392 {
1393         return EXT4_SB(sb)->s_dummy_enc_ctx.ctx;
1394 }
1395
1396 static bool ext4_has_stable_inodes(struct super_block *sb)
1397 {
1398         return ext4_has_feature_stable_inodes(sb);
1399 }
1400
1401 static void ext4_get_ino_and_lblk_bits(struct super_block *sb,
1402                                        int *ino_bits_ret, int *lblk_bits_ret)
1403 {
1404         *ino_bits_ret = 8 * sizeof(EXT4_SB(sb)->s_es->s_inodes_count);
1405         *lblk_bits_ret = 8 * sizeof(ext4_lblk_t);
1406 }
1407
1408 static const struct fscrypt_operations ext4_cryptops = {
1409         .key_prefix             = "ext4:",
1410         .get_context            = ext4_get_context,
1411         .set_context            = ext4_set_context,
1412         .get_dummy_context      = ext4_get_dummy_context,
1413         .empty_dir              = ext4_empty_dir,
1414         .max_namelen            = EXT4_NAME_LEN,
1415         .has_stable_inodes      = ext4_has_stable_inodes,
1416         .get_ino_and_lblk_bits  = ext4_get_ino_and_lblk_bits,
1417 };
1418 #endif
1419
1420 #ifdef CONFIG_QUOTA
1421 static const char * const quotatypes[] = INITQFNAMES;
1422 #define QTYPE2NAME(t) (quotatypes[t])
1423
1424 static int ext4_write_dquot(struct dquot *dquot);
1425 static int ext4_acquire_dquot(struct dquot *dquot);
1426 static int ext4_release_dquot(struct dquot *dquot);
1427 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1428 static int ext4_write_info(struct super_block *sb, int type);
1429 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1430                          const struct path *path);
1431 static int ext4_quota_on_mount(struct super_block *sb, int type);
1432 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1433                                size_t len, loff_t off);
1434 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1435                                 const char *data, size_t len, loff_t off);
1436 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1437                              unsigned int flags);
1438 static int ext4_enable_quotas(struct super_block *sb);
1439
1440 static struct dquot **ext4_get_dquots(struct inode *inode)
1441 {
1442         return EXT4_I(inode)->i_dquot;
1443 }
1444
1445 static const struct dquot_operations ext4_quota_operations = {
1446         .get_reserved_space     = ext4_get_reserved_space,
1447         .write_dquot            = ext4_write_dquot,
1448         .acquire_dquot          = ext4_acquire_dquot,
1449         .release_dquot          = ext4_release_dquot,
1450         .mark_dirty             = ext4_mark_dquot_dirty,
1451         .write_info             = ext4_write_info,
1452         .alloc_dquot            = dquot_alloc,
1453         .destroy_dquot          = dquot_destroy,
1454         .get_projid             = ext4_get_projid,
1455         .get_inode_usage        = ext4_get_inode_usage,
1456         .get_next_id            = dquot_get_next_id,
1457 };
1458
1459 static const struct quotactl_ops ext4_qctl_operations = {
1460         .quota_on       = ext4_quota_on,
1461         .quota_off      = ext4_quota_off,
1462         .quota_sync     = dquot_quota_sync,
1463         .get_state      = dquot_get_state,
1464         .set_info       = dquot_set_dqinfo,
1465         .get_dqblk      = dquot_get_dqblk,
1466         .set_dqblk      = dquot_set_dqblk,
1467         .get_nextdqblk  = dquot_get_next_dqblk,
1468 };
1469 #endif
1470
1471 static const struct super_operations ext4_sops = {
1472         .alloc_inode    = ext4_alloc_inode,
1473         .free_inode     = ext4_free_in_core_inode,
1474         .destroy_inode  = ext4_destroy_inode,
1475         .write_inode    = ext4_write_inode,
1476         .dirty_inode    = ext4_dirty_inode,
1477         .drop_inode     = ext4_drop_inode,
1478         .evict_inode    = ext4_evict_inode,
1479         .put_super      = ext4_put_super,
1480         .sync_fs        = ext4_sync_fs,
1481         .freeze_fs      = ext4_freeze,
1482         .unfreeze_fs    = ext4_unfreeze,
1483         .statfs         = ext4_statfs,
1484         .remount_fs     = ext4_remount,
1485         .show_options   = ext4_show_options,
1486 #ifdef CONFIG_QUOTA
1487         .quota_read     = ext4_quota_read,
1488         .quota_write    = ext4_quota_write,
1489         .get_dquots     = ext4_get_dquots,
1490 #endif
1491         .bdev_try_to_free_page = bdev_try_to_free_page,
1492 };
1493
1494 static const struct export_operations ext4_export_ops = {
1495         .fh_to_dentry = ext4_fh_to_dentry,
1496         .fh_to_parent = ext4_fh_to_parent,
1497         .get_parent = ext4_get_parent,
1498         .commit_metadata = ext4_nfs_commit_metadata,
1499 };
1500
1501 enum {
1502         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1503         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1504         Opt_nouid32, Opt_debug, Opt_removed,
1505         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1506         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1507         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1508         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1509         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1510         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1511         Opt_inlinecrypt,
1512         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1513         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1514         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1515         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version,
1516         Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never,
1517         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
1518         Opt_nowarn_on_error, Opt_mblk_io_submit,
1519         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1520         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1521         Opt_inode_readahead_blks, Opt_journal_ioprio,
1522         Opt_dioread_nolock, Opt_dioread_lock,
1523         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1524         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1525 };
1526
1527 static const match_table_t tokens = {
1528         {Opt_bsd_df, "bsddf"},
1529         {Opt_minix_df, "minixdf"},
1530         {Opt_grpid, "grpid"},
1531         {Opt_grpid, "bsdgroups"},
1532         {Opt_nogrpid, "nogrpid"},
1533         {Opt_nogrpid, "sysvgroups"},
1534         {Opt_resgid, "resgid=%u"},
1535         {Opt_resuid, "resuid=%u"},
1536         {Opt_sb, "sb=%u"},
1537         {Opt_err_cont, "errors=continue"},
1538         {Opt_err_panic, "errors=panic"},
1539         {Opt_err_ro, "errors=remount-ro"},
1540         {Opt_nouid32, "nouid32"},
1541         {Opt_debug, "debug"},
1542         {Opt_removed, "oldalloc"},
1543         {Opt_removed, "orlov"},
1544         {Opt_user_xattr, "user_xattr"},
1545         {Opt_nouser_xattr, "nouser_xattr"},
1546         {Opt_acl, "acl"},
1547         {Opt_noacl, "noacl"},
1548         {Opt_noload, "norecovery"},
1549         {Opt_noload, "noload"},
1550         {Opt_removed, "nobh"},
1551         {Opt_removed, "bh"},
1552         {Opt_commit, "commit=%u"},
1553         {Opt_min_batch_time, "min_batch_time=%u"},
1554         {Opt_max_batch_time, "max_batch_time=%u"},
1555         {Opt_journal_dev, "journal_dev=%u"},
1556         {Opt_journal_path, "journal_path=%s"},
1557         {Opt_journal_checksum, "journal_checksum"},
1558         {Opt_nojournal_checksum, "nojournal_checksum"},
1559         {Opt_journal_async_commit, "journal_async_commit"},
1560         {Opt_abort, "abort"},
1561         {Opt_data_journal, "data=journal"},
1562         {Opt_data_ordered, "data=ordered"},
1563         {Opt_data_writeback, "data=writeback"},
1564         {Opt_data_err_abort, "data_err=abort"},
1565         {Opt_data_err_ignore, "data_err=ignore"},
1566         {Opt_offusrjquota, "usrjquota="},
1567         {Opt_usrjquota, "usrjquota=%s"},
1568         {Opt_offgrpjquota, "grpjquota="},
1569         {Opt_grpjquota, "grpjquota=%s"},
1570         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1571         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1572         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1573         {Opt_grpquota, "grpquota"},
1574         {Opt_noquota, "noquota"},
1575         {Opt_quota, "quota"},
1576         {Opt_usrquota, "usrquota"},
1577         {Opt_prjquota, "prjquota"},
1578         {Opt_barrier, "barrier=%u"},
1579         {Opt_barrier, "barrier"},
1580         {Opt_nobarrier, "nobarrier"},
1581         {Opt_i_version, "i_version"},
1582         {Opt_dax, "dax"},
1583         {Opt_dax_always, "dax=always"},
1584         {Opt_dax_inode, "dax=inode"},
1585         {Opt_dax_never, "dax=never"},
1586         {Opt_stripe, "stripe=%u"},
1587         {Opt_delalloc, "delalloc"},
1588         {Opt_warn_on_error, "warn_on_error"},
1589         {Opt_nowarn_on_error, "nowarn_on_error"},
1590         {Opt_lazytime, "lazytime"},
1591         {Opt_nolazytime, "nolazytime"},
1592         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1593         {Opt_nodelalloc, "nodelalloc"},
1594         {Opt_removed, "mblk_io_submit"},
1595         {Opt_removed, "nomblk_io_submit"},
1596         {Opt_block_validity, "block_validity"},
1597         {Opt_noblock_validity, "noblock_validity"},
1598         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1599         {Opt_journal_ioprio, "journal_ioprio=%u"},
1600         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1601         {Opt_auto_da_alloc, "auto_da_alloc"},
1602         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1603         {Opt_dioread_nolock, "dioread_nolock"},
1604         {Opt_dioread_lock, "nodioread_nolock"},
1605         {Opt_dioread_lock, "dioread_lock"},
1606         {Opt_discard, "discard"},
1607         {Opt_nodiscard, "nodiscard"},
1608         {Opt_init_itable, "init_itable=%u"},
1609         {Opt_init_itable, "init_itable"},
1610         {Opt_noinit_itable, "noinit_itable"},
1611         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1612         {Opt_test_dummy_encryption, "test_dummy_encryption=%s"},
1613         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1614         {Opt_inlinecrypt, "inlinecrypt"},
1615         {Opt_nombcache, "nombcache"},
1616         {Opt_nombcache, "no_mbcache"},  /* for backward compatibility */
1617         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1618         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1619         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1620         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1621         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1622         {Opt_err, NULL},
1623 };
1624
1625 static ext4_fsblk_t get_sb_block(void **data)
1626 {
1627         ext4_fsblk_t    sb_block;
1628         char            *options = (char *) *data;
1629
1630         if (!options || strncmp(options, "sb=", 3) != 0)
1631                 return 1;       /* Default location */
1632
1633         options += 3;
1634         /* TODO: use simple_strtoll with >32bit ext4 */
1635         sb_block = simple_strtoul(options, &options, 0);
1636         if (*options && *options != ',') {
1637                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1638                        (char *) *data);
1639                 return 1;
1640         }
1641         if (*options == ',')
1642                 options++;
1643         *data = (void *) options;
1644
1645         return sb_block;
1646 }
1647
1648 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1649 static const char deprecated_msg[] =
1650         "Mount option \"%s\" will be removed by %s\n"
1651         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1652
1653 #ifdef CONFIG_QUOTA
1654 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1655 {
1656         struct ext4_sb_info *sbi = EXT4_SB(sb);
1657         char *qname, *old_qname = get_qf_name(sb, sbi, qtype);
1658         int ret = -1;
1659
1660         if (sb_any_quota_loaded(sb) && !old_qname) {
1661                 ext4_msg(sb, KERN_ERR,
1662                         "Cannot change journaled "
1663                         "quota options when quota turned on");
1664                 return -1;
1665         }
1666         if (ext4_has_feature_quota(sb)) {
1667                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1668                          "ignored when QUOTA feature is enabled");
1669                 return 1;
1670         }
1671         qname = match_strdup(args);
1672         if (!qname) {
1673                 ext4_msg(sb, KERN_ERR,
1674                         "Not enough memory for storing quotafile name");
1675                 return -1;
1676         }
1677         if (old_qname) {
1678                 if (strcmp(old_qname, qname) == 0)
1679                         ret = 1;
1680                 else
1681                         ext4_msg(sb, KERN_ERR,
1682                                  "%s quota file already specified",
1683                                  QTYPE2NAME(qtype));
1684                 goto errout;
1685         }
1686         if (strchr(qname, '/')) {
1687                 ext4_msg(sb, KERN_ERR,
1688                         "quotafile must be on filesystem root");
1689                 goto errout;
1690         }
1691         rcu_assign_pointer(sbi->s_qf_names[qtype], qname);
1692         set_opt(sb, QUOTA);
1693         return 1;
1694 errout:
1695         kfree(qname);
1696         return ret;
1697 }
1698
1699 static int clear_qf_name(struct super_block *sb, int qtype)
1700 {
1701
1702         struct ext4_sb_info *sbi = EXT4_SB(sb);
1703         char *old_qname = get_qf_name(sb, sbi, qtype);
1704
1705         if (sb_any_quota_loaded(sb) && old_qname) {
1706                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1707                         " when quota turned on");
1708                 return -1;
1709         }
1710         rcu_assign_pointer(sbi->s_qf_names[qtype], NULL);
1711         synchronize_rcu();
1712         kfree(old_qname);
1713         return 1;
1714 }
1715 #endif
1716
1717 #define MOPT_SET        0x0001
1718 #define MOPT_CLEAR      0x0002
1719 #define MOPT_NOSUPPORT  0x0004
1720 #define MOPT_EXPLICIT   0x0008
1721 #define MOPT_CLEAR_ERR  0x0010
1722 #define MOPT_GTE0       0x0020
1723 #ifdef CONFIG_QUOTA
1724 #define MOPT_Q          0
1725 #define MOPT_QFMT       0x0040
1726 #else
1727 #define MOPT_Q          MOPT_NOSUPPORT
1728 #define MOPT_QFMT       MOPT_NOSUPPORT
1729 #endif
1730 #define MOPT_DATAJ      0x0080
1731 #define MOPT_NO_EXT2    0x0100
1732 #define MOPT_NO_EXT3    0x0200
1733 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1734 #define MOPT_STRING     0x0400
1735 #define MOPT_SKIP       0x0800
1736
1737 static const struct mount_opts {
1738         int     token;
1739         int     mount_opt;
1740         int     flags;
1741 } ext4_mount_opts[] = {
1742         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1743         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1744         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1745         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1746         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1747         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1748         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1749          MOPT_EXT4_ONLY | MOPT_SET},
1750         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1751          MOPT_EXT4_ONLY | MOPT_CLEAR},
1752         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1753         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1754         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1755          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1756         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1757          MOPT_EXT4_ONLY | MOPT_CLEAR},
1758         {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET},
1759         {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR},
1760         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1761          MOPT_EXT4_ONLY | MOPT_CLEAR},
1762         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1763          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1764         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1765                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1766          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1767         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1768         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1769         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1770         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1771         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1772          MOPT_NO_EXT2},
1773         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1774          MOPT_NO_EXT2},
1775         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1776         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1777         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1778         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1779         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1780         {Opt_commit, 0, MOPT_GTE0},
1781         {Opt_max_batch_time, 0, MOPT_GTE0},
1782         {Opt_min_batch_time, 0, MOPT_GTE0},
1783         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1784         {Opt_init_itable, 0, MOPT_GTE0},
1785         {Opt_dax, EXT4_MOUNT_DAX_ALWAYS, MOPT_SET | MOPT_SKIP},
1786         {Opt_dax_always, EXT4_MOUNT_DAX_ALWAYS,
1787                 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP},
1788         {Opt_dax_inode, EXT4_MOUNT2_DAX_INODE,
1789                 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP},
1790         {Opt_dax_never, EXT4_MOUNT2_DAX_NEVER,
1791                 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP},
1792         {Opt_stripe, 0, MOPT_GTE0},
1793         {Opt_resuid, 0, MOPT_GTE0},
1794         {Opt_resgid, 0, MOPT_GTE0},
1795         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1796         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1797         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1798         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1799         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1800         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1801          MOPT_NO_EXT2 | MOPT_DATAJ},
1802         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1803         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1804 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1805         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1806         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1807 #else
1808         {Opt_acl, 0, MOPT_NOSUPPORT},
1809         {Opt_noacl, 0, MOPT_NOSUPPORT},
1810 #endif
1811         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1812         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1813         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1814         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1815         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1816                                                         MOPT_SET | MOPT_Q},
1817         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1818                                                         MOPT_SET | MOPT_Q},
1819         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1820                                                         MOPT_SET | MOPT_Q},
1821         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1822                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1823                                                         MOPT_CLEAR | MOPT_Q},
1824         {Opt_usrjquota, 0, MOPT_Q},
1825         {Opt_grpjquota, 0, MOPT_Q},
1826         {Opt_offusrjquota, 0, MOPT_Q},
1827         {Opt_offgrpjquota, 0, MOPT_Q},
1828         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1829         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1830         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1831         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1832         {Opt_test_dummy_encryption, 0, MOPT_STRING},
1833         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1834         {Opt_err, 0, 0}
1835 };
1836
1837 #ifdef CONFIG_UNICODE
1838 static const struct ext4_sb_encodings {
1839         __u16 magic;
1840         char *name;
1841         char *version;
1842 } ext4_sb_encoding_map[] = {
1843         {EXT4_ENC_UTF8_12_1, "utf8", "12.1.0"},
1844 };
1845
1846 static int ext4_sb_read_encoding(const struct ext4_super_block *es,
1847                                  const struct ext4_sb_encodings **encoding,
1848                                  __u16 *flags)
1849 {
1850         __u16 magic = le16_to_cpu(es->s_encoding);
1851         int i;
1852
1853         for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++)
1854                 if (magic == ext4_sb_encoding_map[i].magic)
1855                         break;
1856
1857         if (i >= ARRAY_SIZE(ext4_sb_encoding_map))
1858                 return -EINVAL;
1859
1860         *encoding = &ext4_sb_encoding_map[i];
1861         *flags = le16_to_cpu(es->s_encoding_flags);
1862
1863         return 0;
1864 }
1865 #endif
1866
1867 static int ext4_set_test_dummy_encryption(struct super_block *sb,
1868                                           const char *opt,
1869                                           const substring_t *arg,
1870                                           bool is_remount)
1871 {
1872 #ifdef CONFIG_FS_ENCRYPTION
1873         struct ext4_sb_info *sbi = EXT4_SB(sb);
1874         int err;
1875
1876         /*
1877          * This mount option is just for testing, and it's not worthwhile to
1878          * implement the extra complexity (e.g. RCU protection) that would be
1879          * needed to allow it to be set or changed during remount.  We do allow
1880          * it to be specified during remount, but only if there is no change.
1881          */
1882         if (is_remount && !sbi->s_dummy_enc_ctx.ctx) {
1883                 ext4_msg(sb, KERN_WARNING,
1884                          "Can't set test_dummy_encryption on remount");
1885                 return -1;
1886         }
1887         err = fscrypt_set_test_dummy_encryption(sb, arg, &sbi->s_dummy_enc_ctx);
1888         if (err) {
1889                 if (err == -EEXIST)
1890                         ext4_msg(sb, KERN_WARNING,
1891                                  "Can't change test_dummy_encryption on remount");
1892                 else if (err == -EINVAL)
1893                         ext4_msg(sb, KERN_WARNING,
1894                                  "Value of option \"%s\" is unrecognized", opt);
1895                 else
1896                         ext4_msg(sb, KERN_WARNING,
1897                                  "Error processing option \"%s\" [%d]",
1898                                  opt, err);
1899                 return -1;
1900         }
1901         ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled");
1902 #else
1903         ext4_msg(sb, KERN_WARNING,
1904                  "Test dummy encryption mount option ignored");
1905 #endif
1906         return 1;
1907 }
1908
1909 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1910                             substring_t *args, unsigned long *journal_devnum,
1911                             unsigned int *journal_ioprio, int is_remount)
1912 {
1913         struct ext4_sb_info *sbi = EXT4_SB(sb);
1914         const struct mount_opts *m;
1915         kuid_t uid;
1916         kgid_t gid;
1917         int arg = 0;
1918
1919 #ifdef CONFIG_QUOTA
1920         if (token == Opt_usrjquota)
1921                 return set_qf_name(sb, USRQUOTA, &args[0]);
1922         else if (token == Opt_grpjquota)
1923                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1924         else if (token == Opt_offusrjquota)
1925                 return clear_qf_name(sb, USRQUOTA);
1926         else if (token == Opt_offgrpjquota)
1927                 return clear_qf_name(sb, GRPQUOTA);
1928 #endif
1929         switch (token) {
1930         case Opt_noacl:
1931         case Opt_nouser_xattr:
1932                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1933                 break;
1934         case Opt_sb:
1935                 return 1;       /* handled by get_sb_block() */
1936         case Opt_removed:
1937                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1938                 return 1;
1939         case Opt_abort:
1940                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1941                 return 1;
1942         case Opt_i_version:
1943                 sb->s_flags |= SB_I_VERSION;
1944                 return 1;
1945         case Opt_lazytime:
1946                 sb->s_flags |= SB_LAZYTIME;
1947                 return 1;
1948         case Opt_nolazytime:
1949                 sb->s_flags &= ~SB_LAZYTIME;
1950                 return 1;
1951         case Opt_inlinecrypt:
1952 #ifdef CONFIG_FS_ENCRYPTION_INLINE_CRYPT
1953                 sb->s_flags |= SB_INLINECRYPT;
1954 #else
1955                 ext4_msg(sb, KERN_ERR, "inline encryption not supported");
1956 #endif
1957                 return 1;
1958         }
1959
1960         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1961                 if (token == m->token)
1962                         break;
1963
1964         if (m->token == Opt_err) {
1965                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1966                          "or missing value", opt);
1967                 return -1;
1968         }
1969
1970         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1971                 ext4_msg(sb, KERN_ERR,
1972                          "Mount option \"%s\" incompatible with ext2", opt);
1973                 return -1;
1974         }
1975         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1976                 ext4_msg(sb, KERN_ERR,
1977                          "Mount option \"%s\" incompatible with ext3", opt);
1978                 return -1;
1979         }
1980
1981         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1982                 return -1;
1983         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1984                 return -1;
1985         if (m->flags & MOPT_EXPLICIT) {
1986                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1987                         set_opt2(sb, EXPLICIT_DELALLOC);
1988                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1989                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1990                 } else
1991                         return -1;
1992         }
1993         if (m->flags & MOPT_CLEAR_ERR)
1994                 clear_opt(sb, ERRORS_MASK);
1995         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1996                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1997                          "options when quota turned on");
1998                 return -1;
1999         }
2000
2001         if (m->flags & MOPT_NOSUPPORT) {
2002                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
2003         } else if (token == Opt_commit) {
2004                 if (arg == 0)
2005                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
2006                 else if (arg > INT_MAX / HZ) {
2007                         ext4_msg(sb, KERN_ERR,
2008                                  "Invalid commit interval %d, "
2009                                  "must be smaller than %d",
2010                                  arg, INT_MAX / HZ);
2011                         return -1;
2012                 }
2013                 sbi->s_commit_interval = HZ * arg;
2014         } else if (token == Opt_debug_want_extra_isize) {
2015                 if ((arg & 1) ||
2016                     (arg < 4) ||
2017                     (arg > (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE))) {
2018                         ext4_msg(sb, KERN_ERR,
2019                                  "Invalid want_extra_isize %d", arg);
2020                         return -1;
2021                 }
2022                 sbi->s_want_extra_isize = arg;
2023         } else if (token == Opt_max_batch_time) {
2024                 sbi->s_max_batch_time = arg;
2025         } else if (token == Opt_min_batch_time) {
2026                 sbi->s_min_batch_time = arg;
2027         } else if (token == Opt_inode_readahead_blks) {
2028                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
2029                         ext4_msg(sb, KERN_ERR,
2030                                  "EXT4-fs: inode_readahead_blks must be "
2031                                  "0 or a power of 2 smaller than 2^31");
2032                         return -1;
2033                 }
2034                 sbi->s_inode_readahead_blks = arg;
2035         } else if (token == Opt_init_itable) {
2036                 set_opt(sb, INIT_INODE_TABLE);
2037                 if (!args->from)
2038                         arg = EXT4_DEF_LI_WAIT_MULT;
2039                 sbi->s_li_wait_mult = arg;
2040         } else if (token == Opt_max_dir_size_kb) {
2041                 sbi->s_max_dir_size_kb = arg;
2042         } else if (token == Opt_stripe) {
2043                 sbi->s_stripe = arg;
2044         } else if (token == Opt_resuid) {
2045                 uid = make_kuid(current_user_ns(), arg);
2046                 if (!uid_valid(uid)) {
2047                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
2048                         return -1;
2049                 }
2050                 sbi->s_resuid = uid;
2051         } else if (token == Opt_resgid) {
2052                 gid = make_kgid(current_user_ns(), arg);
2053                 if (!gid_valid(gid)) {
2054                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
2055                         return -1;
2056                 }
2057                 sbi->s_resgid = gid;
2058         } else if (token == Opt_journal_dev) {
2059                 if (is_remount) {
2060                         ext4_msg(sb, KERN_ERR,
2061                                  "Cannot specify journal on remount");
2062                         return -1;
2063                 }
2064                 *journal_devnum = arg;
2065         } else if (token == Opt_journal_path) {
2066                 char *journal_path;
2067                 struct inode *journal_inode;
2068                 struct path path;
2069                 int error;
2070
2071                 if (is_remount) {
2072                         ext4_msg(sb, KERN_ERR,
2073                                  "Cannot specify journal on remount");
2074                         return -1;
2075                 }
2076                 journal_path = match_strdup(&args[0]);
2077                 if (!journal_path) {
2078                         ext4_msg(sb, KERN_ERR, "error: could not dup "
2079                                 "journal device string");
2080                         return -1;
2081                 }
2082
2083                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
2084                 if (error) {
2085                         ext4_msg(sb, KERN_ERR, "error: could not find "
2086                                 "journal device path: error %d", error);
2087                         kfree(journal_path);
2088                         return -1;
2089                 }
2090
2091                 journal_inode = d_inode(path.dentry);
2092                 if (!S_ISBLK(journal_inode->i_mode)) {
2093                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
2094                                 "is not a block device", journal_path);
2095                         path_put(&path);
2096                         kfree(journal_path);
2097                         return -1;
2098                 }
2099
2100                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
2101                 path_put(&path);
2102                 kfree(journal_path);
2103         } else if (token == Opt_journal_ioprio) {
2104                 if (arg > 7) {
2105                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
2106                                  " (must be 0-7)");
2107                         return -1;
2108                 }
2109                 *journal_ioprio =
2110                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
2111         } else if (token == Opt_test_dummy_encryption) {
2112                 return ext4_set_test_dummy_encryption(sb, opt, &args[0],
2113                                                       is_remount);
2114         } else if (m->flags & MOPT_DATAJ) {
2115                 if (is_remount) {
2116                         if (!sbi->s_journal)
2117                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
2118                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
2119                                 ext4_msg(sb, KERN_ERR,
2120                                          "Cannot change data mode on remount");
2121                                 return -1;
2122                         }
2123                 } else {
2124                         clear_opt(sb, DATA_FLAGS);
2125                         sbi->s_mount_opt |= m->mount_opt;
2126                 }
2127 #ifdef CONFIG_QUOTA
2128         } else if (m->flags & MOPT_QFMT) {
2129                 if (sb_any_quota_loaded(sb) &&
2130                     sbi->s_jquota_fmt != m->mount_opt) {
2131                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
2132                                  "quota options when quota turned on");
2133                         return -1;
2134                 }
2135                 if (ext4_has_feature_quota(sb)) {
2136                         ext4_msg(sb, KERN_INFO,
2137                                  "Quota format mount options ignored "
2138                                  "when QUOTA feature is enabled");
2139                         return 1;
2140                 }
2141                 sbi->s_jquota_fmt = m->mount_opt;
2142 #endif
2143         } else if (token == Opt_dax || token == Opt_dax_always ||
2144                    token == Opt_dax_inode || token == Opt_dax_never) {
2145 #ifdef CONFIG_FS_DAX
2146                 switch (token) {
2147                 case Opt_dax:
2148                 case Opt_dax_always:
2149                         if (is_remount &&
2150                             (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2151                              (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) {
2152                         fail_dax_change_remount:
2153                                 ext4_msg(sb, KERN_ERR, "can't change "
2154                                          "dax mount option while remounting");
2155                                 return -1;
2156                         }
2157                         if (is_remount &&
2158                             (test_opt(sb, DATA_FLAGS) ==
2159                              EXT4_MOUNT_JOURNAL_DATA)) {
2160                                     ext4_msg(sb, KERN_ERR, "can't mount with "
2161                                              "both data=journal and dax");
2162                                     return -1;
2163                         }
2164                         ext4_msg(sb, KERN_WARNING,
2165                                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
2166                         sbi->s_mount_opt |= EXT4_MOUNT_DAX_ALWAYS;
2167                         sbi->s_mount_opt2 &= ~EXT4_MOUNT2_DAX_NEVER;
2168                         break;
2169                 case Opt_dax_never:
2170                         if (is_remount &&
2171                             (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2172                              (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS)))
2173                                 goto fail_dax_change_remount;
2174                         sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_NEVER;
2175                         sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS;
2176                         break;
2177                 case Opt_dax_inode:
2178                         if (is_remount &&
2179                             ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2180                              (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2181                              !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE)))
2182                                 goto fail_dax_change_remount;
2183                         sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS;
2184                         sbi->s_mount_opt2 &= ~EXT4_MOUNT2_DAX_NEVER;
2185                         /* Strictly for printing options */
2186                         sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_INODE;
2187                         break;
2188                 }
2189 #else
2190                 ext4_msg(sb, KERN_INFO, "dax option not supported");
2191                 sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_NEVER;
2192                 sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS;
2193                 return -1;
2194 #endif
2195         } else if (token == Opt_data_err_abort) {
2196                 sbi->s_mount_opt |= m->mount_opt;
2197         } else if (token == Opt_data_err_ignore) {
2198                 sbi->s_mount_opt &= ~m->mount_opt;
2199         } else {
2200                 if (!args->from)
2201                         arg = 1;
2202                 if (m->flags & MOPT_CLEAR)
2203                         arg = !arg;
2204                 else if (unlikely(!(m->flags & MOPT_SET))) {
2205                         ext4_msg(sb, KERN_WARNING,
2206                                  "buggy handling of option %s", opt);
2207                         WARN_ON(1);
2208                         return -1;
2209                 }
2210                 if (arg != 0)
2211                         sbi->s_mount_opt |= m->mount_opt;
2212                 else
2213                         sbi->s_mount_opt &= ~m->mount_opt;
2214         }
2215         return 1;
2216 }
2217
2218 static int parse_options(char *options, struct super_block *sb,
2219                          unsigned long *journal_devnum,
2220                          unsigned int *journal_ioprio,
2221                          int is_remount)
2222 {
2223         struct ext4_sb_info __maybe_unused *sbi = EXT4_SB(sb);
2224         char *p, __maybe_unused *usr_qf_name, __maybe_unused *grp_qf_name;
2225         substring_t args[MAX_OPT_ARGS];
2226         int token;
2227
2228         if (!options)
2229                 return 1;
2230
2231         while ((p = strsep(&options, ",")) != NULL) {
2232                 if (!*p)
2233                         continue;
2234                 /*
2235                  * Initialize args struct so we know whether arg was
2236                  * found; some options take optional arguments.
2237                  */
2238                 args[0].to = args[0].from = NULL;
2239                 token = match_token(p, tokens, args);
2240                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
2241                                      journal_ioprio, is_remount) < 0)
2242                         return 0;
2243         }
2244 #ifdef CONFIG_QUOTA
2245         /*
2246          * We do the test below only for project quotas. 'usrquota' and
2247          * 'grpquota' mount options are allowed even without quota feature
2248          * to support legacy quotas in quota files.
2249          */
2250         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
2251                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
2252                          "Cannot enable project quota enforcement.");
2253                 return 0;
2254         }
2255         usr_qf_name = get_qf_name(sb, sbi, USRQUOTA);
2256         grp_qf_name = get_qf_name(sb, sbi, GRPQUOTA);
2257         if (usr_qf_name || grp_qf_name) {
2258                 if (test_opt(sb, USRQUOTA) && usr_qf_name)
2259                         clear_opt(sb, USRQUOTA);
2260
2261                 if (test_opt(sb, GRPQUOTA) && grp_qf_name)
2262                         clear_opt(sb, GRPQUOTA);
2263
2264                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
2265                         ext4_msg(sb, KERN_ERR, "old and new quota "
2266                                         "format mixing");
2267                         return 0;
2268                 }
2269
2270                 if (!sbi->s_jquota_fmt) {
2271                         ext4_msg(sb, KERN_ERR, "journaled quota format "
2272                                         "not specified");
2273                         return 0;
2274                 }
2275         }
2276 #endif
2277         if (test_opt(sb, DIOREAD_NOLOCK)) {
2278                 int blocksize =
2279                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
2280                 if (blocksize < PAGE_SIZE)
2281                         ext4_msg(sb, KERN_WARNING, "Warning: mounting with an "
2282                                  "experimental mount option 'dioread_nolock' "
2283                                  "for blocksize < PAGE_SIZE");
2284         }
2285         return 1;
2286 }
2287
2288 static inline void ext4_show_quota_options(struct seq_file *seq,
2289                                            struct super_block *sb)
2290 {
2291 #if defined(CONFIG_QUOTA)
2292         struct ext4_sb_info *sbi = EXT4_SB(sb);
2293         char *usr_qf_name, *grp_qf_name;
2294
2295         if (sbi->s_jquota_fmt) {
2296                 char *fmtname = "";
2297
2298                 switch (sbi->s_jquota_fmt) {
2299                 case QFMT_VFS_OLD:
2300                         fmtname = "vfsold";
2301                         break;
2302                 case QFMT_VFS_V0:
2303                         fmtname = "vfsv0";
2304                         break;
2305                 case QFMT_VFS_V1:
2306                         fmtname = "vfsv1";
2307                         break;
2308                 }
2309                 seq_printf(seq, ",jqfmt=%s", fmtname);
2310         }
2311
2312         rcu_read_lock();
2313         usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]);
2314         grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]);
2315         if (usr_qf_name)
2316                 seq_show_option(seq, "usrjquota", usr_qf_name);
2317         if (grp_qf_name)
2318                 seq_show_option(seq, "grpjquota", grp_qf_name);
2319         rcu_read_unlock();
2320 #endif
2321 }
2322
2323 static const char *token2str(int token)
2324 {
2325         const struct match_token *t;
2326
2327         for (t = tokens; t->token != Opt_err; t++)
2328                 if (t->token == token && !strchr(t->pattern, '='))
2329                         break;
2330         return t->pattern;
2331 }
2332
2333 /*
2334  * Show an option if
2335  *  - it's set to a non-default value OR
2336  *  - if the per-sb default is different from the global default
2337  */
2338 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2339                               int nodefs)
2340 {
2341         struct ext4_sb_info *sbi = EXT4_SB(sb);
2342         struct ext4_super_block *es = sbi->s_es;
2343         int def_errors, def_mount_opt = sbi->s_def_mount_opt;
2344         const struct mount_opts *m;
2345         char sep = nodefs ? '\n' : ',';
2346
2347 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2348 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2349
2350         if (sbi->s_sb_block != 1)
2351                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2352
2353         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2354                 int want_set = m->flags & MOPT_SET;
2355                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2356                     (m->flags & MOPT_CLEAR_ERR) || m->flags & MOPT_SKIP)
2357                         continue;
2358                 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2359                         continue; /* skip if same as the default */
2360                 if ((want_set &&
2361                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2362                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2363                         continue; /* select Opt_noFoo vs Opt_Foo */
2364                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2365         }
2366
2367         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2368             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2369                 SEQ_OPTS_PRINT("resuid=%u",
2370                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2371         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2372             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2373                 SEQ_OPTS_PRINT("resgid=%u",
2374                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2375         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2376         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2377                 SEQ_OPTS_PUTS("errors=remount-ro");
2378         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2379                 SEQ_OPTS_PUTS("errors=continue");
2380         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2381                 SEQ_OPTS_PUTS("errors=panic");
2382         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2383                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2384         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2385                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2386         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2387                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2388         if (sb->s_flags & SB_I_VERSION)
2389                 SEQ_OPTS_PUTS("i_version");
2390         if (nodefs || sbi->s_stripe)
2391                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2392         if (nodefs || EXT4_MOUNT_DATA_FLAGS &
2393                         (sbi->s_mount_opt ^ def_mount_opt)) {
2394                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2395                         SEQ_OPTS_PUTS("data=journal");
2396                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2397                         SEQ_OPTS_PUTS("data=ordered");
2398                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2399                         SEQ_OPTS_PUTS("data=writeback");
2400         }
2401         if (nodefs ||
2402             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2403                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2404                                sbi->s_inode_readahead_blks);
2405
2406         if (test_opt(sb, INIT_INODE_TABLE) && (nodefs ||
2407                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2408                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2409         if (nodefs || sbi->s_max_dir_size_kb)
2410                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2411         if (test_opt(sb, DATA_ERR_ABORT))
2412                 SEQ_OPTS_PUTS("data_err=abort");
2413
2414         fscrypt_show_test_dummy_encryption(seq, sep, sb);
2415
2416         if (sb->s_flags & SB_INLINECRYPT)
2417                 SEQ_OPTS_PUTS("inlinecrypt");
2418
2419         if (test_opt(sb, DAX_ALWAYS)) {
2420                 if (IS_EXT2_SB(sb))
2421                         SEQ_OPTS_PUTS("dax");
2422                 else
2423                         SEQ_OPTS_PUTS("dax=always");
2424         } else if (test_opt2(sb, DAX_NEVER)) {
2425                 SEQ_OPTS_PUTS("dax=never");
2426         } else if (test_opt2(sb, DAX_INODE)) {
2427                 SEQ_OPTS_PUTS("dax=inode");
2428         }
2429
2430         ext4_show_quota_options(seq, sb);
2431         return 0;
2432 }
2433
2434 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2435 {
2436         return _ext4_show_options(seq, root->d_sb, 0);
2437 }
2438
2439 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2440 {
2441         struct super_block *sb = seq->private;
2442         int rc;
2443
2444         seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
2445         rc = _ext4_show_options(seq, sb, 1);
2446         seq_puts(seq, "\n");
2447         return rc;
2448 }
2449
2450 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2451                             int read_only)
2452 {
2453         struct ext4_sb_info *sbi = EXT4_SB(sb);
2454         int err = 0;
2455
2456         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2457                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2458                          "forcing read-only mode");
2459                 err = -EROFS;
2460                 goto done;
2461         }
2462         if (read_only)
2463                 goto done;
2464         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2465                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2466                          "running e2fsck is recommended");
2467         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2468                 ext4_msg(sb, KERN_WARNING,
2469                          "warning: mounting fs with errors, "
2470                          "running e2fsck is recommended");
2471         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2472                  le16_to_cpu(es->s_mnt_count) >=
2473                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2474                 ext4_msg(sb, KERN_WARNING,
2475                          "warning: maximal mount count reached, "
2476                          "running e2fsck is recommended");
2477         else if (le32_to_cpu(es->s_checkinterval) &&
2478                  (ext4_get_tstamp(es, s_lastcheck) +
2479                   le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds()))
2480                 ext4_msg(sb, KERN_WARNING,
2481                          "warning: checktime reached, "
2482                          "running e2fsck is recommended");
2483         if (!sbi->s_journal)
2484                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2485         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2486                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2487         le16_add_cpu(&es->s_mnt_count, 1);
2488         ext4_update_tstamp(es, s_mtime);
2489         if (sbi->s_journal)
2490                 ext4_set_feature_journal_needs_recovery(sb);
2491
2492         err = ext4_commit_super(sb, 1);
2493 done:
2494         if (test_opt(sb, DEBUG))
2495                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2496                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2497                         sb->s_blocksize,
2498                         sbi->s_groups_count,
2499                         EXT4_BLOCKS_PER_GROUP(sb),
2500                         EXT4_INODES_PER_GROUP(sb),
2501                         sbi->s_mount_opt, sbi->s_mount_opt2);
2502
2503         cleancache_init_fs(sb);
2504         return err;
2505 }
2506
2507 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2508 {
2509         struct ext4_sb_info *sbi = EXT4_SB(sb);
2510         struct flex_groups **old_groups, **new_groups;
2511         int size, i, j;
2512
2513         if (!sbi->s_log_groups_per_flex)
2514                 return 0;
2515
2516         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2517         if (size <= sbi->s_flex_groups_allocated)
2518                 return 0;
2519
2520         new_groups = kvzalloc(roundup_pow_of_two(size *
2521                               sizeof(*sbi->s_flex_groups)), GFP_KERNEL);
2522         if (!new_groups) {
2523                 ext4_msg(sb, KERN_ERR,
2524                          "not enough memory for %d flex group pointers", size);
2525                 return -ENOMEM;
2526         }
2527         for (i = sbi->s_flex_groups_allocated; i < size; i++) {
2528                 new_groups[i] = kvzalloc(roundup_pow_of_two(
2529                                          sizeof(struct flex_groups)),
2530                                          GFP_KERNEL);
2531                 if (!new_groups[i]) {
2532                         for (j = sbi->s_flex_groups_allocated; j < i; j++)
2533                                 kvfree(new_groups[j]);
2534                         kvfree(new_groups);
2535                         ext4_msg(sb, KERN_ERR,
2536                                  "not enough memory for %d flex groups", size);
2537                         return -ENOMEM;
2538                 }
2539         }
2540         rcu_read_lock();
2541         old_groups = rcu_dereference(sbi->s_flex_groups);
2542         if (old_groups)
2543                 memcpy(new_groups, old_groups,
2544                        (sbi->s_flex_groups_allocated *
2545                         sizeof(struct flex_groups *)));
2546         rcu_read_unlock();
2547         rcu_assign_pointer(sbi->s_flex_groups, new_groups);
2548         sbi->s_flex_groups_allocated = size;
2549         if (old_groups)
2550                 ext4_kvfree_array_rcu(old_groups);
2551         return 0;
2552 }
2553
2554 static int ext4_fill_flex_info(struct super_block *sb)
2555 {
2556         struct ext4_sb_info *sbi = EXT4_SB(sb);
2557         struct ext4_group_desc *gdp = NULL;
2558         struct flex_groups *fg;
2559         ext4_group_t flex_group;
2560         int i, err;
2561
2562         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2563         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2564                 sbi->s_log_groups_per_flex = 0;
2565                 return 1;
2566         }
2567
2568         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2569         if (err)
2570                 goto failed;
2571
2572         for (i = 0; i < sbi->s_groups_count; i++) {
2573                 gdp = ext4_get_group_desc(sb, i, NULL);
2574
2575                 flex_group = ext4_flex_group(sbi, i);
2576                 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
2577                 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes);
2578                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2579                              &fg->free_clusters);
2580                 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs);
2581         }
2582
2583         return 1;
2584 failed:
2585         return 0;
2586 }
2587
2588 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2589                                    struct ext4_group_desc *gdp)
2590 {
2591         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2592         __u16 crc = 0;
2593         __le32 le_group = cpu_to_le32(block_group);
2594         struct ext4_sb_info *sbi = EXT4_SB(sb);
2595
2596         if (ext4_has_metadata_csum(sbi->s_sb)) {
2597                 /* Use new metadata_csum algorithm */
2598                 __u32 csum32;
2599                 __u16 dummy_csum = 0;
2600
2601                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2602                                      sizeof(le_group));
2603                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2604                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2605                                      sizeof(dummy_csum));
2606                 offset += sizeof(dummy_csum);
2607                 if (offset < sbi->s_desc_size)
2608                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2609                                              sbi->s_desc_size - offset);
2610
2611                 crc = csum32 & 0xFFFF;
2612                 goto out;
2613         }
2614
2615         /* old crc16 code */
2616         if (!ext4_has_feature_gdt_csum(sb))
2617                 return 0;
2618
2619         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2620         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2621         crc = crc16(crc, (__u8 *)gdp, offset);
2622         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2623         /* for checksum of struct ext4_group_desc do the rest...*/
2624         if (ext4_has_feature_64bit(sb) &&
2625             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2626                 crc = crc16(crc, (__u8 *)gdp + offset,
2627                             le16_to_cpu(sbi->s_es->s_desc_size) -
2628                                 offset);
2629
2630 out:
2631         return cpu_to_le16(crc);
2632 }
2633
2634 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2635                                 struct ext4_group_desc *gdp)
2636 {
2637         if (ext4_has_group_desc_csum(sb) &&
2638             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2639                 return 0;
2640
2641         return 1;
2642 }
2643
2644 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2645                               struct ext4_group_desc *gdp)
2646 {
2647         if (!ext4_has_group_desc_csum(sb))
2648                 return;
2649         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2650 }
2651
2652 /* Called at mount-time, super-block is locked */
2653 static int ext4_check_descriptors(struct super_block *sb,
2654                                   ext4_fsblk_t sb_block,
2655                                   ext4_group_t *first_not_zeroed)
2656 {
2657         struct ext4_sb_info *sbi = EXT4_SB(sb);
2658         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2659         ext4_fsblk_t last_block;
2660         ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0);
2661         ext4_fsblk_t block_bitmap;
2662         ext4_fsblk_t inode_bitmap;
2663         ext4_fsblk_t inode_table;
2664         int flexbg_flag = 0;
2665         ext4_group_t i, grp = sbi->s_groups_count;
2666
2667         if (ext4_has_feature_flex_bg(sb))
2668                 flexbg_flag = 1;
2669
2670         ext4_debug("Checking group descriptors");
2671
2672         for (i = 0; i < sbi->s_groups_count; i++) {
2673                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2674
2675                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2676                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2677                 else
2678                         last_block = first_block +
2679                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2680
2681                 if ((grp == sbi->s_groups_count) &&
2682                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2683                         grp = i;
2684
2685                 block_bitmap = ext4_block_bitmap(sb, gdp);
2686                 if (block_bitmap == sb_block) {
2687                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2688                                  "Block bitmap for group %u overlaps "
2689                                  "superblock", i);
2690                         if (!sb_rdonly(sb))
2691                                 return 0;
2692                 }
2693                 if (block_bitmap >= sb_block + 1 &&
2694                     block_bitmap <= last_bg_block) {
2695                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2696                                  "Block bitmap for group %u overlaps "
2697                                  "block group descriptors", i);
2698                         if (!sb_rdonly(sb))
2699                                 return 0;
2700                 }
2701                 if (block_bitmap < first_block || block_bitmap > last_block) {
2702                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2703                                "Block bitmap for group %u not in group "
2704                                "(block %llu)!", i, block_bitmap);
2705                         return 0;
2706                 }
2707                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2708                 if (inode_bitmap == sb_block) {
2709                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2710                                  "Inode bitmap for group %u overlaps "
2711                                  "superblock", i);
2712                         if (!sb_rdonly(sb))
2713                                 return 0;
2714                 }
2715                 if (inode_bitmap >= sb_block + 1 &&
2716                     inode_bitmap <= last_bg_block) {
2717                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2718                                  "Inode bitmap for group %u overlaps "
2719                                  "block group descriptors", i);
2720                         if (!sb_rdonly(sb))
2721                                 return 0;
2722                 }
2723                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2724                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2725                                "Inode bitmap for group %u not in group "
2726                                "(block %llu)!", i, inode_bitmap);
2727                         return 0;
2728                 }
2729                 inode_table = ext4_inode_table(sb, gdp);
2730                 if (inode_table == sb_block) {
2731                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2732                                  "Inode table for group %u overlaps "
2733                                  "superblock", i);
2734                         if (!sb_rdonly(sb))
2735                                 return 0;
2736                 }
2737                 if (inode_table >= sb_block + 1 &&
2738                     inode_table <= last_bg_block) {
2739                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2740                                  "Inode table for group %u overlaps "
2741                                  "block group descriptors", i);
2742                         if (!sb_rdonly(sb))
2743                                 return 0;
2744                 }
2745                 if (inode_table < first_block ||
2746                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2747                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2748                                "Inode table for group %u not in group "
2749                                "(block %llu)!", i, inode_table);
2750                         return 0;
2751                 }
2752                 ext4_lock_group(sb, i);
2753                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2754                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2755                                  "Checksum for group %u failed (%u!=%u)",
2756                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2757                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2758                         if (!sb_rdonly(sb)) {
2759                                 ext4_unlock_group(sb, i);
2760                                 return 0;
2761                         }
2762                 }
2763                 ext4_unlock_group(sb, i);
2764                 if (!flexbg_flag)
2765                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2766         }
2767         if (NULL != first_not_zeroed)
2768                 *first_not_zeroed = grp;
2769         return 1;
2770 }
2771
2772 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2773  * the superblock) which were deleted from all directories, but held open by
2774  * a process at the time of a crash.  We walk the list and try to delete these
2775  * inodes at recovery time (only with a read-write filesystem).
2776  *
2777  * In order to keep the orphan inode chain consistent during traversal (in
2778  * case of crash during recovery), we link each inode into the superblock
2779  * orphan list_head and handle it the same way as an inode deletion during
2780  * normal operation (which journals the operations for us).
2781  *
2782  * We only do an iget() and an iput() on each inode, which is very safe if we
2783  * accidentally point at an in-use or already deleted inode.  The worst that
2784  * can happen in this case is that we get a "bit already cleared" message from
2785  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2786  * e2fsck was run on this filesystem, and it must have already done the orphan
2787  * inode cleanup for us, so we can safely abort without any further action.
2788  */
2789 static void ext4_orphan_cleanup(struct super_block *sb,
2790                                 struct ext4_super_block *es)
2791 {
2792         unsigned int s_flags = sb->s_flags;
2793         int ret, nr_orphans = 0, nr_truncates = 0;
2794 #ifdef CONFIG_QUOTA
2795         int quota_update = 0;
2796         int i;
2797 #endif
2798         if (!es->s_last_orphan) {
2799                 jbd_debug(4, "no orphan inodes to clean up\n");
2800                 return;
2801         }
2802
2803         if (bdev_read_only(sb->s_bdev)) {
2804                 ext4_msg(sb, KERN_ERR, "write access "
2805                         "unavailable, skipping orphan cleanup");
2806                 return;
2807         }
2808
2809         /* Check if feature set would not allow a r/w mount */
2810         if (!ext4_feature_set_ok(sb, 0)) {
2811                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2812                          "unknown ROCOMPAT features");
2813                 return;
2814         }
2815
2816         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2817                 /* don't clear list on RO mount w/ errors */
2818                 if (es->s_last_orphan && !(s_flags & SB_RDONLY)) {
2819                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2820                                   "clearing orphan list.\n");
2821                         es->s_last_orphan = 0;
2822                 }
2823                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2824                 return;
2825         }
2826
2827         if (s_flags & SB_RDONLY) {
2828                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2829                 sb->s_flags &= ~SB_RDONLY;
2830         }
2831 #ifdef CONFIG_QUOTA
2832         /* Needed for iput() to work correctly and not trash data */
2833         sb->s_flags |= SB_ACTIVE;
2834
2835         /*
2836          * Turn on quotas which were not enabled for read-only mounts if
2837          * filesystem has quota feature, so that they are updated correctly.
2838          */
2839         if (ext4_has_feature_quota(sb) && (s_flags & SB_RDONLY)) {
2840                 int ret = ext4_enable_quotas(sb);
2841
2842                 if (!ret)
2843                         quota_update = 1;
2844                 else
2845                         ext4_msg(sb, KERN_ERR,
2846                                 "Cannot turn on quotas: error %d", ret);
2847         }
2848
2849         /* Turn on journaled quotas used for old sytle */
2850         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2851                 if (EXT4_SB(sb)->s_qf_names[i]) {
2852                         int ret = ext4_quota_on_mount(sb, i);
2853
2854                         if (!ret)
2855                                 quota_update = 1;
2856                         else
2857                                 ext4_msg(sb, KERN_ERR,
2858                                         "Cannot turn on journaled "
2859                                         "quota: type %d: error %d", i, ret);
2860                 }
2861         }
2862 #endif
2863
2864         while (es->s_last_orphan) {
2865                 struct inode *inode;
2866
2867                 /*
2868                  * We may have encountered an error during cleanup; if
2869                  * so, skip the rest.
2870                  */
2871                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2872                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2873                         es->s_last_orphan = 0;
2874                         break;
2875                 }
2876
2877                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2878                 if (IS_ERR(inode)) {
2879                         es->s_last_orphan = 0;
2880                         break;
2881                 }
2882
2883                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2884                 dquot_initialize(inode);
2885                 if (inode->i_nlink) {
2886                         if (test_opt(sb, DEBUG))
2887                                 ext4_msg(sb, KERN_DEBUG,
2888                                         "%s: truncating inode %lu to %lld bytes",
2889                                         __func__, inode->i_ino, inode->i_size);
2890                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2891                                   inode->i_ino, inode->i_size);
2892                         inode_lock(inode);
2893                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2894                         ret = ext4_truncate(inode);
2895                         if (ret)
2896                                 ext4_std_error(inode->i_sb, ret);
2897                         inode_unlock(inode);
2898                         nr_truncates++;
2899                 } else {
2900                         if (test_opt(sb, DEBUG))
2901                                 ext4_msg(sb, KERN_DEBUG,
2902                                         "%s: deleting unreferenced inode %lu",
2903                                         __func__, inode->i_ino);
2904                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2905                                   inode->i_ino);
2906                         nr_orphans++;
2907                 }
2908                 iput(inode);  /* The delete magic happens here! */
2909         }
2910
2911 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2912
2913         if (nr_orphans)
2914                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2915                        PLURAL(nr_orphans));
2916         if (nr_truncates)
2917                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2918                        PLURAL(nr_truncates));
2919 #ifdef CONFIG_QUOTA
2920         /* Turn off quotas if they were enabled for orphan cleanup */
2921         if (quota_update) {
2922                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2923                         if (sb_dqopt(sb)->files[i])
2924                                 dquot_quota_off(sb, i);
2925                 }
2926         }
2927 #endif
2928         sb->s_flags = s_flags; /* Restore SB_RDONLY status */
2929 }
2930
2931 /*
2932  * Maximal extent format file size.
2933  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2934  * extent format containers, within a sector_t, and within i_blocks
2935  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2936  * so that won't be a limiting factor.
2937  *
2938  * However there is other limiting factor. We do store extents in the form
2939  * of starting block and length, hence the resulting length of the extent
2940  * covering maximum file size must fit into on-disk format containers as
2941  * well. Given that length is always by 1 unit bigger than max unit (because
2942  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2943  *
2944  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2945  */
2946 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2947 {
2948         loff_t res;
2949         loff_t upper_limit = MAX_LFS_FILESIZE;
2950
2951         BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64));
2952
2953         if (!has_huge_files) {
2954                 upper_limit = (1LL << 32) - 1;
2955
2956                 /* total blocks in file system block size */
2957                 upper_limit >>= (blkbits - 9);
2958                 upper_limit <<= blkbits;
2959         }
2960
2961         /*
2962          * 32-bit extent-start container, ee_block. We lower the maxbytes
2963          * by one fs block, so ee_len can cover the extent of maximum file
2964          * size
2965          */
2966         res = (1LL << 32) - 1;
2967         res <<= blkbits;
2968
2969         /* Sanity check against vm- & vfs- imposed limits */
2970         if (res > upper_limit)
2971                 res = upper_limit;
2972
2973         return res;
2974 }
2975
2976 /*
2977  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2978  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2979  * We need to be 1 filesystem block less than the 2^48 sector limit.
2980  */
2981 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2982 {
2983         loff_t res = EXT4_NDIR_BLOCKS;
2984         int meta_blocks;
2985         loff_t upper_limit;
2986         /* This is calculated to be the largest file size for a dense, block
2987          * mapped file such that the file's total number of 512-byte sectors,
2988          * including data and all indirect blocks, does not exceed (2^48 - 1).
2989          *
2990          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2991          * number of 512-byte sectors of the file.
2992          */
2993
2994         if (!has_huge_files) {
2995                 /*
2996                  * !has_huge_files or implies that the inode i_block field
2997                  * represents total file blocks in 2^32 512-byte sectors ==
2998                  * size of vfs inode i_blocks * 8
2999                  */
3000                 upper_limit = (1LL << 32) - 1;
3001
3002                 /* total blocks in file system block size */
3003                 upper_limit >>= (bits - 9);
3004
3005         } else {
3006                 /*
3007                  * We use 48 bit ext4_inode i_blocks
3008                  * With EXT4_HUGE_FILE_FL set the i_blocks
3009                  * represent total number of blocks in
3010                  * file system block size
3011                  */
3012                 upper_limit = (1LL << 48) - 1;
3013
3014         }
3015
3016         /* indirect blocks */
3017         meta_blocks = 1;
3018         /* double indirect blocks */
3019         meta_blocks += 1 + (1LL << (bits-2));
3020         /* tripple indirect blocks */
3021         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
3022
3023         upper_limit -= meta_blocks;
3024         upper_limit <<= bits;
3025
3026         res += 1LL << (bits-2);
3027         res += 1LL << (2*(bits-2));
3028         res += 1LL << (3*(bits-2));
3029         res <<= bits;
3030         if (res > upper_limit)
3031                 res = upper_limit;
3032
3033         if (res > MAX_LFS_FILESIZE)
3034                 res = MAX_LFS_FILESIZE;
3035
3036         return res;
3037 }
3038
3039 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
3040                                    ext4_fsblk_t logical_sb_block, int nr)
3041 {
3042         struct ext4_sb_info *sbi = EXT4_SB(sb);
3043         ext4_group_t bg, first_meta_bg;
3044         int has_super = 0;
3045
3046         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
3047
3048         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
3049                 return logical_sb_block + nr + 1;
3050         bg = sbi->s_desc_per_block * nr;
3051         if (ext4_bg_has_super(sb, bg))
3052                 has_super = 1;
3053
3054         /*
3055          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
3056          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
3057          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
3058          * compensate.
3059          */
3060         if (sb->s_blocksize == 1024 && nr == 0 &&
3061             le32_to_cpu(sbi->s_es->s_first_data_block) == 0)
3062                 has_super++;
3063
3064         return (has_super + ext4_group_first_block_no(sb, bg));
3065 }
3066
3067 /**
3068  * ext4_get_stripe_size: Get the stripe size.
3069  * @sbi: In memory super block info
3070  *
3071  * If we have specified it via mount option, then
3072  * use the mount option value. If the value specified at mount time is
3073  * greater than the blocks per group use the super block value.
3074  * If the super block value is greater than blocks per group return 0.
3075  * Allocator needs it be less than blocks per group.
3076  *
3077  */
3078 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
3079 {
3080         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
3081         unsigned long stripe_width =
3082                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
3083         int ret;
3084
3085         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
3086                 ret = sbi->s_stripe;
3087         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
3088                 ret = stripe_width;
3089         else if (stride && stride <= sbi->s_blocks_per_group)
3090                 ret = stride;
3091         else
3092                 ret = 0;
3093
3094         /*
3095          * If the stripe width is 1, this makes no sense and
3096          * we set it to 0 to turn off stripe handling code.
3097          */
3098         if (ret <= 1)
3099                 ret = 0;
3100
3101         return ret;
3102 }
3103
3104 /*
3105  * Check whether this filesystem can be mounted based on
3106  * the features present and the RDONLY/RDWR mount requested.
3107  * Returns 1 if this filesystem can be mounted as requested,
3108  * 0 if it cannot be.
3109  */
3110 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
3111 {
3112         if (ext4_has_unknown_ext4_incompat_features(sb)) {
3113                 ext4_msg(sb, KERN_ERR,
3114                         "Couldn't mount because of "
3115                         "unsupported optional features (%x)",
3116                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
3117                         ~EXT4_FEATURE_INCOMPAT_SUPP));
3118                 return 0;
3119         }
3120
3121 #ifndef CONFIG_UNICODE
3122         if (ext4_has_feature_casefold(sb)) {
3123                 ext4_msg(sb, KERN_ERR,
3124                          "Filesystem with casefold feature cannot be "
3125                          "mounted without CONFIG_UNICODE");
3126                 return 0;
3127         }
3128 #endif
3129
3130         if (readonly)
3131                 return 1;
3132
3133         if (ext4_has_feature_readonly(sb)) {
3134                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
3135                 sb->s_flags |= SB_RDONLY;
3136                 return 1;
3137         }
3138
3139         /* Check that feature set is OK for a read-write mount */
3140         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
3141                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
3142                          "unsupported optional features (%x)",
3143                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
3144                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
3145                 return 0;
3146         }
3147         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
3148                 ext4_msg(sb, KERN_ERR,
3149                          "Can't support bigalloc feature without "
3150                          "extents feature\n");
3151                 return 0;
3152         }
3153
3154 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2)
3155         if (!readonly && (ext4_has_feature_quota(sb) ||
3156                           ext4_has_feature_project(sb))) {
3157                 ext4_msg(sb, KERN_ERR,
3158                          "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2");
3159                 return 0;
3160         }
3161 #endif  /* CONFIG_QUOTA */
3162         return 1;
3163 }
3164
3165 /*
3166  * This function is called once a day if we have errors logged
3167  * on the file system
3168  */
3169 static void print_daily_error_info(struct timer_list *t)
3170 {
3171         struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report);
3172         struct super_block *sb = sbi->s_sb;
3173         struct ext4_super_block *es = sbi->s_es;
3174
3175         if (es->s_error_count)
3176                 /* fsck newer than v1.41.13 is needed to clean this condition. */
3177                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
3178                          le32_to_cpu(es->s_error_count));
3179         if (es->s_first_error_time) {
3180                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d",
3181                        sb->s_id,
3182                        ext4_get_tstamp(es, s_first_error_time),
3183                        (int) sizeof(es->s_first_error_func),
3184                        es->s_first_error_func,
3185                        le32_to_cpu(es->s_first_error_line));
3186                 if (es->s_first_error_ino)
3187                         printk(KERN_CONT ": inode %u",
3188                                le32_to_cpu(es->s_first_error_ino));
3189                 if (es->s_first_error_block)
3190                         printk(KERN_CONT ": block %llu", (unsigned long long)
3191                                le64_to_cpu(es->s_first_error_block));
3192                 printk(KERN_CONT "\n");
3193         }
3194         if (es->s_last_error_time) {
3195                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d",
3196                        sb->s_id,
3197                        ext4_get_tstamp(es, s_last_error_time),
3198                        (int) sizeof(es->s_last_error_func),
3199                        es->s_last_error_func,
3200                        le32_to_cpu(es->s_last_error_line));
3201                 if (es->s_last_error_ino)
3202                         printk(KERN_CONT ": inode %u",
3203                                le32_to_cpu(es->s_last_error_ino));
3204                 if (es->s_last_error_block)
3205                         printk(KERN_CONT ": block %llu", (unsigned long long)
3206                                le64_to_cpu(es->s_last_error_block));
3207                 printk(KERN_CONT "\n");
3208         }
3209         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
3210 }
3211
3212 /* Find next suitable group and run ext4_init_inode_table */
3213 static int ext4_run_li_request(struct ext4_li_request *elr)
3214 {
3215         struct ext4_group_desc *gdp = NULL;
3216         ext4_group_t group, ngroups;
3217         struct super_block *sb;
3218         unsigned long timeout = 0;
3219         int ret = 0;
3220
3221         sb = elr->lr_super;
3222         ngroups = EXT4_SB(sb)->s_groups_count;
3223
3224         for (group = elr->lr_next_group; group < ngroups; group++) {
3225                 gdp = ext4_get_group_desc(sb, group, NULL);
3226                 if (!gdp) {
3227                         ret = 1;
3228                         break;
3229                 }
3230
3231                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3232                         break;
3233         }
3234
3235         if (group >= ngroups)
3236                 ret = 1;
3237
3238         if (!ret) {
3239                 timeout = jiffies;
3240                 ret = ext4_init_inode_table(sb, group,
3241                                             elr->lr_timeout ? 0 : 1);
3242                 if (elr->lr_timeout == 0) {
3243                         timeout = (jiffies - timeout) *
3244                                   elr->lr_sbi->s_li_wait_mult;
3245                         elr->lr_timeout = timeout;
3246                 }
3247                 elr->lr_next_sched = jiffies + elr->lr_timeout;
3248                 elr->lr_next_group = group + 1;
3249         }
3250         return ret;
3251 }
3252
3253 /*
3254  * Remove lr_request from the list_request and free the
3255  * request structure. Should be called with li_list_mtx held
3256  */
3257 static void ext4_remove_li_request(struct ext4_li_request *elr)
3258 {
3259         struct ext4_sb_info *sbi;
3260
3261         if (!elr)
3262                 return;
3263
3264         sbi = elr->lr_sbi;
3265
3266         list_del(&elr->lr_request);
3267         sbi->s_li_request = NULL;
3268         kfree(elr);
3269 }
3270
3271 static void ext4_unregister_li_request(struct super_block *sb)
3272 {
3273         mutex_lock(&ext4_li_mtx);
3274         if (!ext4_li_info) {
3275                 mutex_unlock(&ext4_li_mtx);
3276                 return;
3277         }
3278
3279         mutex_lock(&ext4_li_info->li_list_mtx);
3280         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
3281         mutex_unlock(&ext4_li_info->li_list_mtx);
3282         mutex_unlock(&ext4_li_mtx);
3283 }
3284
3285 static struct task_struct *ext4_lazyinit_task;
3286
3287 /*
3288  * This is the function where ext4lazyinit thread lives. It walks
3289  * through the request list searching for next scheduled filesystem.
3290  * When such a fs is found, run the lazy initialization request
3291  * (ext4_rn_li_request) and keep track of the time spend in this
3292  * function. Based on that time we compute next schedule time of
3293  * the request. When walking through the list is complete, compute
3294  * next waking time and put itself into sleep.
3295  */
3296 static int ext4_lazyinit_thread(void *arg)
3297 {
3298         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
3299         struct list_head *pos, *n;
3300         struct ext4_li_request *elr;
3301         unsigned long next_wakeup, cur;
3302
3303         BUG_ON(NULL == eli);
3304
3305 cont_thread:
3306         while (true) {
3307                 next_wakeup = MAX_JIFFY_OFFSET;
3308
3309                 mutex_lock(&eli->li_list_mtx);
3310                 if (list_empty(&eli->li_request_list)) {
3311                         mutex_unlock(&eli->li_list_mtx);
3312                         goto exit_thread;
3313                 }
3314                 list_for_each_safe(pos, n, &eli->li_request_list) {
3315                         int err = 0;
3316                         int progress = 0;
3317                         elr = list_entry(pos, struct ext4_li_request,
3318                                          lr_request);
3319
3320                         if (time_before(jiffies, elr->lr_next_sched)) {
3321                                 if (time_before(elr->lr_next_sched, next_wakeup))
3322                                         next_wakeup = elr->lr_next_sched;
3323                                 continue;
3324                         }
3325                         if (down_read_trylock(&elr->lr_super->s_umount)) {
3326                                 if (sb_start_write_trylock(elr->lr_super)) {
3327                                         progress = 1;
3328                                         /*
3329                                          * We hold sb->s_umount, sb can not
3330                                          * be removed from the list, it is
3331                                          * now safe to drop li_list_mtx
3332                                          */
3333                                         mutex_unlock(&eli->li_list_mtx);
3334                                         err = ext4_run_li_request(elr);
3335                                         sb_end_write(elr->lr_super);
3336                                         mutex_lock(&eli->li_list_mtx);
3337                                         n = pos->next;
3338                                 }
3339                                 up_read((&elr->lr_super->s_umount));
3340                         }
3341                         /* error, remove the lazy_init job */
3342                         if (err) {
3343                                 ext4_remove_li_request(elr);
3344                                 continue;
3345                         }
3346                         if (!progress) {
3347                                 elr->lr_next_sched = jiffies +
3348                                         (prandom_u32()
3349                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3350                         }
3351                         if (time_before(elr->lr_next_sched, next_wakeup))
3352                                 next_wakeup = elr->lr_next_sched;
3353                 }
3354                 mutex_unlock(&eli->li_list_mtx);
3355
3356                 try_to_freeze();
3357
3358                 cur = jiffies;
3359                 if ((time_after_eq(cur, next_wakeup)) ||
3360                     (MAX_JIFFY_OFFSET == next_wakeup)) {
3361                         cond_resched();
3362                         continue;
3363                 }
3364
3365                 schedule_timeout_interruptible(next_wakeup - cur);
3366
3367                 if (kthread_should_stop()) {
3368                         ext4_clear_request_list();
3369                         goto exit_thread;
3370                 }
3371         }
3372
3373 exit_thread:
3374         /*
3375          * It looks like the request list is empty, but we need
3376          * to check it under the li_list_mtx lock, to prevent any
3377          * additions into it, and of course we should lock ext4_li_mtx
3378          * to atomically free the list and ext4_li_info, because at
3379          * this point another ext4 filesystem could be registering
3380          * new one.
3381          */
3382         mutex_lock(&ext4_li_mtx);
3383         mutex_lock(&eli->li_list_mtx);
3384         if (!list_empty(&eli->li_request_list)) {
3385                 mutex_unlock(&eli->li_list_mtx);
3386                 mutex_unlock(&ext4_li_mtx);
3387                 goto cont_thread;
3388         }
3389         mutex_unlock(&eli->li_list_mtx);
3390         kfree(ext4_li_info);
3391         ext4_li_info = NULL;
3392         mutex_unlock(&ext4_li_mtx);
3393
3394         return 0;
3395 }
3396
3397 static void ext4_clear_request_list(void)
3398 {
3399         struct list_head *pos, *n;
3400         struct ext4_li_request *elr;
3401
3402         mutex_lock(&ext4_li_info->li_list_mtx);
3403         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3404                 elr = list_entry(pos, struct ext4_li_request,
3405                                  lr_request);
3406                 ext4_remove_li_request(elr);
3407         }
3408         mutex_unlock(&ext4_li_info->li_list_mtx);
3409 }
3410
3411 static int ext4_run_lazyinit_thread(void)
3412 {
3413         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3414                                          ext4_li_info, "ext4lazyinit");
3415         if (IS_ERR(ext4_lazyinit_task)) {
3416                 int err = PTR_ERR(ext4_lazyinit_task);
3417                 ext4_clear_request_list();
3418                 kfree(ext4_li_info);
3419                 ext4_li_info = NULL;
3420                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3421                                  "initialization thread\n",
3422                                  err);
3423                 return err;
3424         }
3425         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3426         return 0;
3427 }
3428
3429 /*
3430  * Check whether it make sense to run itable init. thread or not.
3431  * If there is at least one uninitialized inode table, return
3432  * corresponding group number, else the loop goes through all
3433  * groups and return total number of groups.
3434  */
3435 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3436 {
3437         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3438         struct ext4_group_desc *gdp = NULL;
3439
3440         if (!ext4_has_group_desc_csum(sb))
3441                 return ngroups;
3442
3443         for (group = 0; group < ngroups; group++) {
3444                 gdp = ext4_get_group_desc(sb, group, NULL);
3445                 if (!gdp)
3446                         continue;
3447
3448                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3449                         break;
3450         }
3451
3452         return group;
3453 }
3454
3455 static int ext4_li_info_new(void)
3456 {
3457         struct ext4_lazy_init *eli = NULL;
3458
3459         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3460         if (!eli)
3461                 return -ENOMEM;
3462
3463         INIT_LIST_HEAD(&eli->li_request_list);
3464         mutex_init(&eli->li_list_mtx);
3465
3466         eli->li_state |= EXT4_LAZYINIT_QUIT;
3467
3468         ext4_li_info = eli;
3469
3470         return 0;
3471 }
3472
3473 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3474                                             ext4_group_t start)
3475 {
3476         struct ext4_sb_info *sbi = EXT4_SB(sb);
3477         struct ext4_li_request *elr;
3478
3479         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3480         if (!elr)
3481                 return NULL;
3482
3483         elr->lr_super = sb;
3484         elr->lr_sbi = sbi;
3485         elr->lr_next_group = start;
3486
3487         /*
3488          * Randomize first schedule time of the request to
3489          * spread the inode table initialization requests
3490          * better.
3491          */
3492         elr->lr_next_sched = jiffies + (prandom_u32() %
3493                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3494         return elr;
3495 }
3496
3497 int ext4_register_li_request(struct super_block *sb,
3498                              ext4_group_t first_not_zeroed)
3499 {
3500         struct ext4_sb_info *sbi = EXT4_SB(sb);
3501         struct ext4_li_request *elr = NULL;
3502         ext4_group_t ngroups = sbi->s_groups_count;
3503         int ret = 0;
3504
3505         mutex_lock(&ext4_li_mtx);
3506         if (sbi->s_li_request != NULL) {
3507                 /*
3508                  * Reset timeout so it can be computed again, because
3509                  * s_li_wait_mult might have changed.
3510                  */
3511                 sbi->s_li_request->lr_timeout = 0;
3512                 goto out;
3513         }
3514
3515         if (first_not_zeroed == ngroups || sb_rdonly(sb) ||
3516             !test_opt(sb, INIT_INODE_TABLE))
3517                 goto out;
3518
3519         elr = ext4_li_request_new(sb, first_not_zeroed);
3520         if (!elr) {
3521                 ret = -ENOMEM;
3522                 goto out;
3523         }
3524
3525         if (NULL == ext4_li_info) {
3526                 ret = ext4_li_info_new();
3527                 if (ret)
3528                         goto out;
3529         }
3530
3531         mutex_lock(&ext4_li_info->li_list_mtx);
3532         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3533         mutex_unlock(&ext4_li_info->li_list_mtx);
3534
3535         sbi->s_li_request = elr;
3536         /*
3537          * set elr to NULL here since it has been inserted to
3538          * the request_list and the removal and free of it is
3539          * handled by ext4_clear_request_list from now on.
3540          */
3541         elr = NULL;
3542
3543         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3544                 ret = ext4_run_lazyinit_thread();
3545                 if (ret)
3546                         goto out;
3547         }
3548 out:
3549         mutex_unlock(&ext4_li_mtx);
3550         if (ret)
3551                 kfree(elr);
3552         return ret;
3553 }
3554
3555 /*
3556  * We do not need to lock anything since this is called on
3557  * module unload.
3558  */
3559 static void ext4_destroy_lazyinit_thread(void)
3560 {
3561         /*
3562          * If thread exited earlier
3563          * there's nothing to be done.
3564          */
3565         if (!ext4_li_info || !ext4_lazyinit_task)
3566                 return;
3567
3568         kthread_stop(ext4_lazyinit_task);
3569 }
3570
3571 static int set_journal_csum_feature_set(struct super_block *sb)
3572 {
3573         int ret = 1;
3574         int compat, incompat;
3575         struct ext4_sb_info *sbi = EXT4_SB(sb);
3576
3577         if (ext4_has_metadata_csum(sb)) {
3578                 /* journal checksum v3 */
3579                 compat = 0;
3580                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3581         } else {
3582                 /* journal checksum v1 */
3583                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3584                 incompat = 0;
3585         }
3586
3587         jbd2_journal_clear_features(sbi->s_journal,
3588                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3589                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3590                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3591         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3592                 ret = jbd2_journal_set_features(sbi->s_journal,
3593                                 compat, 0,
3594                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3595                                 incompat);
3596         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3597                 ret = jbd2_journal_set_features(sbi->s_journal,
3598                                 compat, 0,
3599                                 incompat);
3600                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3601                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3602         } else {
3603                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3604                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3605         }
3606
3607         return ret;
3608 }
3609
3610 /*
3611  * Note: calculating the overhead so we can be compatible with
3612  * historical BSD practice is quite difficult in the face of
3613  * clusters/bigalloc.  This is because multiple metadata blocks from
3614  * different block group can end up in the same allocation cluster.
3615  * Calculating the exact overhead in the face of clustered allocation
3616  * requires either O(all block bitmaps) in memory or O(number of block
3617  * groups**2) in time.  We will still calculate the superblock for
3618  * older file systems --- and if we come across with a bigalloc file
3619  * system with zero in s_overhead_clusters the estimate will be close to
3620  * correct especially for very large cluster sizes --- but for newer
3621  * file systems, it's better to calculate this figure once at mkfs
3622  * time, and store it in the superblock.  If the superblock value is
3623  * present (even for non-bigalloc file systems), we will use it.
3624  */
3625 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3626                           char *buf)
3627 {
3628         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3629         struct ext4_group_desc  *gdp;
3630         ext4_fsblk_t            first_block, last_block, b;
3631         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3632         int                     s, j, count = 0;
3633
3634         if (!ext4_has_feature_bigalloc(sb))
3635                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3636                         sbi->s_itb_per_group + 2);
3637
3638         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3639                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3640         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3641         for (i = 0; i < ngroups; i++) {
3642                 gdp = ext4_get_group_desc(sb, i, NULL);
3643                 b = ext4_block_bitmap(sb, gdp);
3644                 if (b >= first_block && b <= last_block) {
3645                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3646                         count++;
3647                 }
3648                 b = ext4_inode_bitmap(sb, gdp);
3649                 if (b >= first_block && b <= last_block) {
3650                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3651                         count++;
3652                 }
3653                 b = ext4_inode_table(sb, gdp);
3654                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3655                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3656                                 int c = EXT4_B2C(sbi, b - first_block);
3657                                 ext4_set_bit(c, buf);
3658                                 count++;
3659                         }
3660                 if (i != grp)
3661                         continue;
3662                 s = 0;
3663                 if (ext4_bg_has_super(sb, grp)) {
3664                         ext4_set_bit(s++, buf);
3665                         count++;
3666                 }
3667                 j = ext4_bg_num_gdb(sb, grp);
3668                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3669                         ext4_error(sb, "Invalid number of block group "
3670                                    "descriptor blocks: %d", j);
3671                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3672                 }
3673                 count += j;
3674                 for (; j > 0; j--)
3675                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3676         }
3677         if (!count)
3678                 return 0;
3679         return EXT4_CLUSTERS_PER_GROUP(sb) -
3680                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3681 }
3682
3683 /*
3684  * Compute the overhead and stash it in sbi->s_overhead
3685  */
3686 int ext4_calculate_overhead(struct super_block *sb)
3687 {
3688         struct ext4_sb_info *sbi = EXT4_SB(sb);
3689         struct ext4_super_block *es = sbi->s_es;
3690         struct inode *j_inode;
3691         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3692         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3693         ext4_fsblk_t overhead = 0;
3694         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3695
3696         if (!buf)
3697                 return -ENOMEM;
3698
3699         /*
3700          * Compute the overhead (FS structures).  This is constant
3701          * for a given filesystem unless the number of block groups
3702          * changes so we cache the previous value until it does.
3703          */
3704
3705         /*
3706          * All of the blocks before first_data_block are overhead
3707          */
3708         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3709
3710         /*
3711          * Add the overhead found in each block group
3712          */
3713         for (i = 0; i < ngroups; i++) {
3714                 int blks;
3715
3716                 blks = count_overhead(sb, i, buf);
3717                 overhead += blks;
3718                 if (blks)
3719                         memset(buf, 0, PAGE_SIZE);
3720                 cond_resched();
3721         }
3722
3723         /*
3724          * Add the internal journal blocks whether the journal has been
3725          * loaded or not
3726          */
3727         if (sbi->s_journal && !sbi->journal_bdev)
3728                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3729         else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) {
3730                 /* j_inum for internal journal is non-zero */
3731                 j_inode = ext4_get_journal_inode(sb, j_inum);
3732                 if (j_inode) {
3733                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3734                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
3735                         iput(j_inode);
3736                 } else {
3737                         ext4_msg(sb, KERN_ERR, "can't get journal size");
3738                 }
3739         }
3740         sbi->s_overhead = overhead;
3741         smp_wmb();
3742         free_page((unsigned long) buf);
3743         return 0;
3744 }
3745
3746 static void ext4_set_resv_clusters(struct super_block *sb)
3747 {
3748         ext4_fsblk_t resv_clusters;
3749         struct ext4_sb_info *sbi = EXT4_SB(sb);
3750
3751         /*
3752          * There's no need to reserve anything when we aren't using extents.
3753          * The space estimates are exact, there are no unwritten extents,
3754          * hole punching doesn't need new metadata... This is needed especially
3755          * to keep ext2/3 backward compatibility.
3756          */
3757         if (!ext4_has_feature_extents(sb))
3758                 return;
3759         /*
3760          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3761          * This should cover the situations where we can not afford to run
3762          * out of space like for example punch hole, or converting
3763          * unwritten extents in delalloc path. In most cases such
3764          * allocation would require 1, or 2 blocks, higher numbers are
3765          * very rare.
3766          */
3767         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3768                          sbi->s_cluster_bits);
3769
3770         do_div(resv_clusters, 50);
3771         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3772
3773         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3774 }
3775
3776 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3777 {
3778         struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev);
3779         char *orig_data = kstrdup(data, GFP_KERNEL);
3780         struct buffer_head *bh, **group_desc;
3781         struct ext4_super_block *es = NULL;
3782         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3783         struct flex_groups **flex_groups;
3784         ext4_fsblk_t block;
3785         ext4_fsblk_t sb_block = get_sb_block(&data);
3786         ext4_fsblk_t logical_sb_block;
3787         unsigned long offset = 0;
3788         unsigned long journal_devnum = 0;
3789         unsigned long def_mount_opts;
3790         struct inode *root;
3791         const char *descr;
3792         int ret = -ENOMEM;
3793         int blocksize, clustersize;
3794         unsigned int db_count;
3795         unsigned int i;
3796         int needs_recovery, has_huge_files;
3797         __u64 blocks_count;
3798         int err = 0;
3799         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3800         ext4_group_t first_not_zeroed;
3801
3802         if ((data && !orig_data) || !sbi)
3803                 goto out_free_base;
3804
3805         sbi->s_daxdev = dax_dev;
3806         sbi->s_blockgroup_lock =
3807                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3808         if (!sbi->s_blockgroup_lock)
3809                 goto out_free_base;
3810
3811         sb->s_fs_info = sbi;
3812         sbi->s_sb = sb;
3813         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3814         sbi->s_sb_block = sb_block;
3815         if (sb->s_bdev->bd_part)
3816                 sbi->s_sectors_written_start =
3817                         part_stat_read(sb->s_bdev->bd_part, sectors[STAT_WRITE]);
3818
3819         /* Cleanup superblock name */
3820         strreplace(sb->s_id, '/', '!');
3821
3822         /* -EINVAL is default */
3823         ret = -EINVAL;
3824         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3825         if (!blocksize) {
3826                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3827                 goto out_fail;
3828         }
3829
3830         /*
3831          * The ext4 superblock will not be buffer aligned for other than 1kB
3832          * block sizes.  We need to calculate the offset from buffer start.
3833          */
3834         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3835                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3836                 offset = do_div(logical_sb_block, blocksize);
3837         } else {
3838                 logical_sb_block = sb_block;
3839         }
3840
3841         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3842                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3843                 goto out_fail;
3844         }
3845         /*
3846          * Note: s_es must be initialized as soon as possible because
3847          *       some ext4 macro-instructions depend on its value
3848          */
3849         es = (struct ext4_super_block *) (bh->b_data + offset);
3850         sbi->s_es = es;
3851         sb->s_magic = le16_to_cpu(es->s_magic);
3852         if (sb->s_magic != EXT4_SUPER_MAGIC)
3853                 goto cantfind_ext4;
3854         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3855
3856         /* Warn if metadata_csum and gdt_csum are both set. */
3857         if (ext4_has_feature_metadata_csum(sb) &&
3858             ext4_has_feature_gdt_csum(sb))
3859                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3860                              "redundant flags; please run fsck.");
3861
3862         /* Check for a known checksum algorithm */
3863         if (!ext4_verify_csum_type(sb, es)) {
3864                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3865                          "unknown checksum algorithm.");
3866                 silent = 1;
3867                 goto cantfind_ext4;
3868         }
3869
3870         /* Load the checksum driver */
3871         sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3872         if (IS_ERR(sbi->s_chksum_driver)) {
3873                 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3874                 ret = PTR_ERR(sbi->s_chksum_driver);
3875                 sbi->s_chksum_driver = NULL;
3876                 goto failed_mount;
3877         }
3878
3879         /* Check superblock checksum */
3880         if (!ext4_superblock_csum_verify(sb, es)) {
3881                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3882                          "invalid superblock checksum.  Run e2fsck?");
3883                 silent = 1;
3884                 ret = -EFSBADCRC;
3885                 goto cantfind_ext4;
3886         }
3887
3888         /* Precompute checksum seed for all metadata */
3889         if (ext4_has_feature_csum_seed(sb))
3890                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3891         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
3892                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3893                                                sizeof(es->s_uuid));
3894
3895         /* Set defaults before we parse the mount options */
3896         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3897         set_opt(sb, INIT_INODE_TABLE);
3898         if (def_mount_opts & EXT4_DEFM_DEBUG)
3899                 set_opt(sb, DEBUG);
3900         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3901                 set_opt(sb, GRPID);
3902         if (def_mount_opts & EXT4_DEFM_UID16)
3903                 set_opt(sb, NO_UID32);
3904         /* xattr user namespace & acls are now defaulted on */
3905         set_opt(sb, XATTR_USER);
3906 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3907         set_opt(sb, POSIX_ACL);
3908 #endif
3909         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3910         if (ext4_has_metadata_csum(sb))
3911                 set_opt(sb, JOURNAL_CHECKSUM);
3912
3913         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3914                 set_opt(sb, JOURNAL_DATA);
3915         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3916                 set_opt(sb, ORDERED_DATA);
3917         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3918                 set_opt(sb, WRITEBACK_DATA);
3919
3920         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3921                 set_opt(sb, ERRORS_PANIC);
3922         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3923                 set_opt(sb, ERRORS_CONT);
3924         else
3925                 set_opt(sb, ERRORS_RO);
3926         /* block_validity enabled by default; disable with noblock_validity */
3927         set_opt(sb, BLOCK_VALIDITY);
3928         if (def_mount_opts & EXT4_DEFM_DISCARD)
3929                 set_opt(sb, DISCARD);
3930
3931         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3932         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3933         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3934         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3935         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3936
3937         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3938                 set_opt(sb, BARRIER);
3939
3940         /*
3941          * enable delayed allocation by default
3942          * Use -o nodelalloc to turn it off
3943          */
3944         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3945             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3946                 set_opt(sb, DELALLOC);
3947
3948         /*
3949          * set default s_li_wait_mult for lazyinit, for the case there is
3950          * no mount option specified.
3951          */
3952         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3953
3954         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3955
3956         if (blocksize == PAGE_SIZE)
3957                 set_opt(sb, DIOREAD_NOLOCK);
3958
3959         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3960             blocksize > EXT4_MAX_BLOCK_SIZE) {
3961                 ext4_msg(sb, KERN_ERR,
3962                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3963                          blocksize, le32_to_cpu(es->s_log_block_size));
3964                 goto failed_mount;
3965         }
3966
3967         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3968                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3969                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3970         } else {
3971                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3972                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3973                 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) {
3974                         ext4_msg(sb, KERN_ERR, "invalid first ino: %u",
3975                                  sbi->s_first_ino);
3976                         goto failed_mount;
3977                 }
3978                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3979                     (!is_power_of_2(sbi->s_inode_size)) ||
3980                     (sbi->s_inode_size > blocksize)) {
3981                         ext4_msg(sb, KERN_ERR,
3982                                "unsupported inode size: %d",
3983                                sbi->s_inode_size);
3984                         ext4_msg(sb, KERN_ERR, "blocksize: %d", blocksize);
3985                         goto failed_mount;
3986                 }
3987                 /*
3988                  * i_atime_extra is the last extra field available for
3989                  * [acm]times in struct ext4_inode. Checking for that
3990                  * field should suffice to ensure we have extra space
3991                  * for all three.
3992                  */
3993                 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) +
3994                         sizeof(((struct ext4_inode *)0)->i_atime_extra)) {
3995                         sb->s_time_gran = 1;
3996                         sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX;
3997                 } else {
3998                         sb->s_time_gran = NSEC_PER_SEC;
3999                         sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX;
4000                 }
4001                 sb->s_time_min = EXT4_TIMESTAMP_MIN;
4002         }
4003         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
4004                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4005                         EXT4_GOOD_OLD_INODE_SIZE;
4006                 if (ext4_has_feature_extra_isize(sb)) {
4007                         unsigned v, max = (sbi->s_inode_size -
4008                                            EXT4_GOOD_OLD_INODE_SIZE);
4009
4010                         v = le16_to_cpu(es->s_want_extra_isize);
4011                         if (v > max) {
4012                                 ext4_msg(sb, KERN_ERR,
4013                                          "bad s_want_extra_isize: %d", v);
4014                                 goto failed_mount;
4015                         }
4016                         if (sbi->s_want_extra_isize < v)
4017                                 sbi->s_want_extra_isize = v;
4018
4019                         v = le16_to_cpu(es->s_min_extra_isize);
4020                         if (v > max) {
4021                                 ext4_msg(sb, KERN_ERR,
4022                                          "bad s_min_extra_isize: %d", v);
4023                                 goto failed_mount;
4024                         }
4025                         if (sbi->s_want_extra_isize < v)
4026                                 sbi->s_want_extra_isize = v;
4027                 }
4028         }
4029
4030         if (sbi->s_es->s_mount_opts[0]) {
4031                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
4032                                               sizeof(sbi->s_es->s_mount_opts),
4033                                               GFP_KERNEL);
4034                 if (!s_mount_opts)
4035                         goto failed_mount;
4036                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
4037                                    &journal_ioprio, 0)) {
4038                         ext4_msg(sb, KERN_WARNING,
4039                                  "failed to parse options in superblock: %s",
4040                                  s_mount_opts);
4041                 }
4042                 kfree(s_mount_opts);
4043         }
4044         sbi->s_def_mount_opt = sbi->s_mount_opt;
4045         if (!parse_options((char *) data, sb, &journal_devnum,
4046                            &journal_ioprio, 0))
4047                 goto failed_mount;
4048
4049 #ifdef CONFIG_UNICODE
4050         if (ext4_has_feature_casefold(sb) && !sbi->s_encoding) {
4051                 const struct ext4_sb_encodings *encoding_info;
4052                 struct unicode_map *encoding;
4053                 __u16 encoding_flags;
4054
4055                 if (ext4_has_feature_encrypt(sb)) {
4056                         ext4_msg(sb, KERN_ERR,
4057                                  "Can't mount with encoding and encryption");
4058                         goto failed_mount;
4059                 }
4060
4061                 if (ext4_sb_read_encoding(es, &encoding_info,
4062                                           &encoding_flags)) {
4063                         ext4_msg(sb, KERN_ERR,
4064                                  "Encoding requested by superblock is unknown");
4065                         goto failed_mount;
4066                 }
4067
4068                 encoding = utf8_load(encoding_info->version);
4069                 if (IS_ERR(encoding)) {
4070                         ext4_msg(sb, KERN_ERR,
4071                                  "can't mount with superblock charset: %s-%s "
4072                                  "not supported by the kernel. flags: 0x%x.",
4073                                  encoding_info->name, encoding_info->version,
4074                                  encoding_flags);
4075                         goto failed_mount;
4076                 }
4077                 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: "
4078                          "%s-%s with flags 0x%hx", encoding_info->name,
4079                          encoding_info->version?:"\b", encoding_flags);
4080
4081                 sbi->s_encoding = encoding;
4082                 sbi->s_encoding_flags = encoding_flags;
4083         }
4084 #endif
4085
4086         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4087                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with data=journal disables delayed allocation, dioread_nolock, and O_DIRECT support!\n");
4088                 /* can't mount with both data=journal and dioread_nolock. */
4089                 clear_opt(sb, DIOREAD_NOLOCK);
4090                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4091                         ext4_msg(sb, KERN_ERR, "can't mount with "
4092                                  "both data=journal and delalloc");
4093                         goto failed_mount;
4094                 }
4095                 if (test_opt(sb, DAX_ALWAYS)) {
4096                         ext4_msg(sb, KERN_ERR, "can't mount with "
4097                                  "both data=journal and dax");
4098                         goto failed_mount;
4099                 }
4100                 if (ext4_has_feature_encrypt(sb)) {
4101                         ext4_msg(sb, KERN_WARNING,
4102                                  "encrypted files will use data=ordered "
4103                                  "instead of data journaling mode");
4104                 }
4105                 if (test_opt(sb, DELALLOC))
4106                         clear_opt(sb, DELALLOC);
4107         } else {
4108                 sb->s_iflags |= SB_I_CGROUPWB;
4109         }
4110
4111         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
4112                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
4113
4114         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
4115             (ext4_has_compat_features(sb) ||
4116              ext4_has_ro_compat_features(sb) ||
4117              ext4_has_incompat_features(sb)))
4118                 ext4_msg(sb, KERN_WARNING,
4119                        "feature flags set on rev 0 fs, "
4120                        "running e2fsck is recommended");
4121
4122         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
4123                 set_opt2(sb, HURD_COMPAT);
4124                 if (ext4_has_feature_64bit(sb)) {
4125                         ext4_msg(sb, KERN_ERR,
4126                                  "The Hurd can't support 64-bit file systems");
4127                         goto failed_mount;
4128                 }
4129
4130                 /*
4131                  * ea_inode feature uses l_i_version field which is not
4132                  * available in HURD_COMPAT mode.
4133                  */
4134                 if (ext4_has_feature_ea_inode(sb)) {
4135                         ext4_msg(sb, KERN_ERR,
4136                                  "ea_inode feature is not supported for Hurd");
4137                         goto failed_mount;
4138                 }
4139         }
4140
4141         if (IS_EXT2_SB(sb)) {
4142                 if (ext2_feature_set_ok(sb))
4143                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
4144                                  "using the ext4 subsystem");
4145                 else {
4146                         /*
4147                          * If we're probing be silent, if this looks like
4148                          * it's actually an ext[34] filesystem.
4149                          */
4150                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4151                                 goto failed_mount;
4152                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
4153                                  "to feature incompatibilities");
4154                         goto failed_mount;
4155                 }
4156         }
4157
4158         if (IS_EXT3_SB(sb)) {
4159                 if (ext3_feature_set_ok(sb))
4160                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
4161                                  "using the ext4 subsystem");
4162                 else {
4163                         /*
4164                          * If we're probing be silent, if this looks like
4165                          * it's actually an ext4 filesystem.
4166                          */
4167                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4168                                 goto failed_mount;
4169                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
4170                                  "to feature incompatibilities");
4171                         goto failed_mount;
4172                 }
4173         }
4174
4175         /*
4176          * Check feature flags regardless of the revision level, since we
4177          * previously didn't change the revision level when setting the flags,
4178          * so there is a chance incompat flags are set on a rev 0 filesystem.
4179          */
4180         if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
4181                 goto failed_mount;
4182
4183         if (le32_to_cpu(es->s_log_block_size) >
4184             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4185                 ext4_msg(sb, KERN_ERR,
4186                          "Invalid log block size: %u",
4187                          le32_to_cpu(es->s_log_block_size));
4188                 goto failed_mount;
4189         }
4190         if (le32_to_cpu(es->s_log_cluster_size) >
4191             (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4192                 ext4_msg(sb, KERN_ERR,
4193                          "Invalid log cluster size: %u",
4194                          le32_to_cpu(es->s_log_cluster_size));
4195                 goto failed_mount;
4196         }
4197
4198         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
4199                 ext4_msg(sb, KERN_ERR,
4200                          "Number of reserved GDT blocks insanely large: %d",
4201                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
4202                 goto failed_mount;
4203         }
4204
4205         if (bdev_dax_supported(sb->s_bdev, blocksize))
4206                 set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags);
4207
4208         if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) {
4209                 if (ext4_has_feature_inline_data(sb)) {
4210                         ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
4211                                         " that may contain inline data");
4212                         goto failed_mount;
4213                 }
4214                 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) {
4215                         ext4_msg(sb, KERN_ERR,
4216                                 "DAX unsupported by block device.");
4217                         goto failed_mount;
4218                 }
4219         }
4220
4221         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
4222                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
4223                          es->s_encryption_level);
4224                 goto failed_mount;
4225         }
4226
4227         if (sb->s_blocksize != blocksize) {
4228                 /* Validate the filesystem blocksize */
4229                 if (!sb_set_blocksize(sb, blocksize)) {
4230                         ext4_msg(sb, KERN_ERR, "bad block size %d",
4231                                         blocksize);
4232                         goto failed_mount;
4233                 }
4234
4235                 brelse(bh);
4236                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
4237                 offset = do_div(logical_sb_block, blocksize);
4238                 bh = sb_bread_unmovable(sb, logical_sb_block);
4239                 if (!bh) {
4240                         ext4_msg(sb, KERN_ERR,
4241                                "Can't read superblock on 2nd try");
4242                         goto failed_mount;
4243                 }
4244                 es = (struct ext4_super_block *)(bh->b_data + offset);
4245                 sbi->s_es = es;
4246                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
4247                         ext4_msg(sb, KERN_ERR,
4248                                "Magic mismatch, very weird!");
4249                         goto failed_mount;
4250                 }
4251         }
4252
4253         has_huge_files = ext4_has_feature_huge_file(sb);
4254         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
4255                                                       has_huge_files);
4256         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
4257
4258         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
4259         if (ext4_has_feature_64bit(sb)) {
4260                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
4261                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
4262                     !is_power_of_2(sbi->s_desc_size)) {
4263                         ext4_msg(sb, KERN_ERR,
4264                                "unsupported descriptor size %lu",
4265                                sbi->s_desc_size);
4266                         goto failed_mount;
4267                 }
4268         } else
4269                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
4270
4271         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
4272         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
4273
4274         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
4275         if (sbi->s_inodes_per_block == 0)
4276                 goto cantfind_ext4;
4277         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
4278             sbi->s_inodes_per_group > blocksize * 8) {
4279                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
4280                          sbi->s_inodes_per_group);
4281                 goto failed_mount;
4282         }
4283         sbi->s_itb_per_group = sbi->s_inodes_per_group /
4284                                         sbi->s_inodes_per_block;
4285         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
4286         sbi->s_sbh = bh;
4287         sbi->s_mount_state = le16_to_cpu(es->s_state);
4288         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
4289         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
4290
4291         for (i = 0; i < 4; i++)
4292                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
4293         sbi->s_def_hash_version = es->s_def_hash_version;
4294         if (ext4_has_feature_dir_index(sb)) {
4295                 i = le32_to_cpu(es->s_flags);
4296                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
4297                         sbi->s_hash_unsigned = 3;
4298                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
4299 #ifdef __CHAR_UNSIGNED__
4300                         if (!sb_rdonly(sb))
4301                                 es->s_flags |=
4302                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
4303                         sbi->s_hash_unsigned = 3;
4304 #else
4305                         if (!sb_rdonly(sb))
4306                                 es->s_flags |=
4307                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
4308 #endif
4309                 }
4310         }
4311
4312         /* Handle clustersize */
4313         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
4314         if (ext4_has_feature_bigalloc(sb)) {
4315                 if (clustersize < blocksize) {
4316                         ext4_msg(sb, KERN_ERR,
4317                                  "cluster size (%d) smaller than "
4318                                  "block size (%d)", clustersize, blocksize);
4319                         goto failed_mount;
4320                 }
4321                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
4322                         le32_to_cpu(es->s_log_block_size);
4323                 sbi->s_clusters_per_group =
4324                         le32_to_cpu(es->s_clusters_per_group);
4325                 if (sbi->s_clusters_per_group > blocksize * 8) {
4326                         ext4_msg(sb, KERN_ERR,
4327                                  "#clusters per group too big: %lu",
4328                                  sbi->s_clusters_per_group);
4329                         goto failed_mount;
4330                 }
4331                 if (sbi->s_blocks_per_group !=
4332                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
4333                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
4334                                  "clusters per group (%lu) inconsistent",
4335                                  sbi->s_blocks_per_group,
4336                                  sbi->s_clusters_per_group);
4337                         goto failed_mount;
4338                 }
4339         } else {
4340                 if (clustersize != blocksize) {
4341                         ext4_msg(sb, KERN_ERR,
4342                                  "fragment/cluster size (%d) != "
4343                                  "block size (%d)", clustersize, blocksize);
4344                         goto failed_mount;
4345                 }
4346                 if (sbi->s_blocks_per_group > blocksize * 8) {
4347                         ext4_msg(sb, KERN_ERR,
4348                                  "#blocks per group too big: %lu",
4349                                  sbi->s_blocks_per_group);
4350                         goto failed_mount;
4351                 }
4352                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
4353                 sbi->s_cluster_bits = 0;
4354         }
4355         sbi->s_cluster_ratio = clustersize / blocksize;
4356
4357         /* Do we have standard group size of clustersize * 8 blocks ? */
4358         if (sbi->s_blocks_per_group == clustersize << 3)
4359                 set_opt2(sb, STD_GROUP_SIZE);
4360
4361         /*
4362          * Test whether we have more sectors than will fit in sector_t,
4363          * and whether the max offset is addressable by the page cache.
4364          */
4365         err = generic_check_addressable(sb->s_blocksize_bits,
4366                                         ext4_blocks_count(es));
4367         if (err) {
4368                 ext4_msg(sb, KERN_ERR, "filesystem"
4369                          " too large to mount safely on this system");
4370                 goto failed_mount;
4371         }
4372
4373         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
4374                 goto cantfind_ext4;
4375
4376         /* check blocks count against device size */
4377         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
4378         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
4379                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
4380                        "exceeds size of device (%llu blocks)",
4381                        ext4_blocks_count(es), blocks_count);
4382                 goto failed_mount;
4383         }
4384
4385         /*
4386          * It makes no sense for the first data block to be beyond the end
4387          * of the filesystem.
4388          */
4389         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
4390                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4391                          "block %u is beyond end of filesystem (%llu)",
4392                          le32_to_cpu(es->s_first_data_block),
4393                          ext4_blocks_count(es));
4394                 goto failed_mount;
4395         }
4396         if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) &&
4397             (sbi->s_cluster_ratio == 1)) {
4398                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4399                          "block is 0 with a 1k block and cluster size");
4400                 goto failed_mount;
4401         }
4402
4403         blocks_count = (ext4_blocks_count(es) -
4404                         le32_to_cpu(es->s_first_data_block) +
4405                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
4406         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
4407         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
4408                 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu "
4409                        "(block count %llu, first data block %u, "
4410                        "blocks per group %lu)", blocks_count,
4411                        ext4_blocks_count(es),
4412                        le32_to_cpu(es->s_first_data_block),
4413                        EXT4_BLOCKS_PER_GROUP(sb));
4414                 goto failed_mount;
4415         }
4416         sbi->s_groups_count = blocks_count;
4417         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
4418                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
4419         if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) !=
4420             le32_to_cpu(es->s_inodes_count)) {
4421                 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu",
4422                          le32_to_cpu(es->s_inodes_count),
4423                          ((u64)sbi->s_groups_count * sbi->s_inodes_per_group));
4424                 ret = -EINVAL;
4425                 goto failed_mount;
4426         }
4427         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
4428                    EXT4_DESC_PER_BLOCK(sb);
4429         if (ext4_has_feature_meta_bg(sb)) {
4430                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
4431                         ext4_msg(sb, KERN_WARNING,
4432                                  "first meta block group too large: %u "
4433                                  "(group descriptor block count %u)",
4434                                  le32_to_cpu(es->s_first_meta_bg), db_count);
4435                         goto failed_mount;
4436                 }
4437         }
4438         rcu_assign_pointer(sbi->s_group_desc,
4439                            kvmalloc_array(db_count,
4440                                           sizeof(struct buffer_head *),
4441                                           GFP_KERNEL));
4442         if (sbi->s_group_desc == NULL) {
4443                 ext4_msg(sb, KERN_ERR, "not enough memory");
4444                 ret = -ENOMEM;
4445                 goto failed_mount;
4446         }
4447
4448         bgl_lock_init(sbi->s_blockgroup_lock);
4449
4450         /* Pre-read the descriptors into the buffer cache */
4451         for (i = 0; i < db_count; i++) {
4452                 block = descriptor_loc(sb, logical_sb_block, i);
4453                 sb_breadahead_unmovable(sb, block);
4454         }
4455
4456         for (i = 0; i < db_count; i++) {
4457                 struct buffer_head *bh;
4458
4459                 block = descriptor_loc(sb, logical_sb_block, i);
4460                 bh = sb_bread_unmovable(sb, block);
4461                 if (!bh) {
4462                         ext4_msg(sb, KERN_ERR,
4463                                "can't read group descriptor %d", i);
4464                         db_count = i;
4465                         goto failed_mount2;
4466                 }
4467                 rcu_read_lock();
4468                 rcu_dereference(sbi->s_group_desc)[i] = bh;
4469                 rcu_read_unlock();
4470         }
4471         sbi->s_gdb_count = db_count;
4472         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
4473                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
4474                 ret = -EFSCORRUPTED;
4475                 goto failed_mount2;
4476         }
4477
4478         timer_setup(&sbi->s_err_report, print_daily_error_info, 0);
4479
4480         /* Register extent status tree shrinker */
4481         if (ext4_es_register_shrinker(sbi))
4482                 goto failed_mount3;
4483
4484         sbi->s_stripe = ext4_get_stripe_size(sbi);
4485         sbi->s_extent_max_zeroout_kb = 32;
4486
4487         /*
4488          * set up enough so that it can read an inode
4489          */
4490         sb->s_op = &ext4_sops;
4491         sb->s_export_op = &ext4_export_ops;
4492         sb->s_xattr = ext4_xattr_handlers;
4493 #ifdef CONFIG_FS_ENCRYPTION
4494         sb->s_cop = &ext4_cryptops;
4495 #endif
4496 #ifdef CONFIG_FS_VERITY
4497         sb->s_vop = &ext4_verityops;
4498 #endif
4499 #ifdef CONFIG_QUOTA
4500         sb->dq_op = &ext4_quota_operations;
4501         if (ext4_has_feature_quota(sb))
4502                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4503         else
4504                 sb->s_qcop = &ext4_qctl_operations;
4505         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
4506 #endif
4507         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
4508
4509         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
4510         mutex_init(&sbi->s_orphan_lock);
4511
4512         sb->s_root = NULL;
4513
4514         needs_recovery = (es->s_last_orphan != 0 ||
4515                           ext4_has_feature_journal_needs_recovery(sb));
4516
4517         if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb))
4518                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
4519                         goto failed_mount3a;
4520
4521         /*
4522          * The first inode we look at is the journal inode.  Don't try
4523          * root first: it may be modified in the journal!
4524          */
4525         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
4526                 err = ext4_load_journal(sb, es, journal_devnum);
4527                 if (err)
4528                         goto failed_mount3a;
4529         } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
4530                    ext4_has_feature_journal_needs_recovery(sb)) {
4531                 ext4_msg(sb, KERN_ERR, "required journal recovery "
4532                        "suppressed and not mounted read-only");
4533                 goto failed_mount_wq;
4534         } else {
4535                 /* Nojournal mode, all journal mount options are illegal */
4536                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
4537                         ext4_msg(sb, KERN_ERR, "can't mount with "
4538                                  "journal_checksum, fs mounted w/o journal");
4539                         goto failed_mount_wq;
4540                 }
4541                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4542                         ext4_msg(sb, KERN_ERR, "can't mount with "
4543                                  "journal_async_commit, fs mounted w/o journal");
4544                         goto failed_mount_wq;
4545                 }
4546                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
4547                         ext4_msg(sb, KERN_ERR, "can't mount with "
4548                                  "commit=%lu, fs mounted w/o journal",
4549                                  sbi->s_commit_interval / HZ);
4550                         goto failed_mount_wq;
4551                 }
4552                 if (EXT4_MOUNT_DATA_FLAGS &
4553                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
4554                         ext4_msg(sb, KERN_ERR, "can't mount with "
4555                                  "data=, fs mounted w/o journal");
4556                         goto failed_mount_wq;
4557                 }
4558                 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM;
4559                 clear_opt(sb, JOURNAL_CHECKSUM);
4560                 clear_opt(sb, DATA_FLAGS);
4561                 sbi->s_journal = NULL;
4562                 needs_recovery = 0;
4563                 goto no_journal;
4564         }
4565
4566         if (ext4_has_feature_64bit(sb) &&
4567             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4568                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4569                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4570                 goto failed_mount_wq;
4571         }
4572
4573         if (!set_journal_csum_feature_set(sb)) {
4574                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4575                          "feature set");
4576                 goto failed_mount_wq;
4577         }
4578
4579         /* We have now updated the journal if required, so we can
4580          * validate the data journaling mode. */
4581         switch (test_opt(sb, DATA_FLAGS)) {
4582         case 0:
4583                 /* No mode set, assume a default based on the journal
4584                  * capabilities: ORDERED_DATA if the journal can
4585                  * cope, else JOURNAL_DATA
4586                  */
4587                 if (jbd2_journal_check_available_features
4588                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4589                         set_opt(sb, ORDERED_DATA);
4590                         sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
4591                 } else {
4592                         set_opt(sb, JOURNAL_DATA);
4593                         sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
4594                 }
4595                 break;
4596
4597         case EXT4_MOUNT_ORDERED_DATA:
4598         case EXT4_MOUNT_WRITEBACK_DATA:
4599                 if (!jbd2_journal_check_available_features
4600                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4601                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4602                                "requested data journaling mode");
4603                         goto failed_mount_wq;
4604                 }
4605         default:
4606                 break;
4607         }
4608
4609         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4610             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4611                 ext4_msg(sb, KERN_ERR, "can't mount with "
4612                         "journal_async_commit in data=ordered mode");
4613                 goto failed_mount_wq;
4614         }
4615
4616         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4617
4618         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4619
4620 no_journal:
4621         if (!test_opt(sb, NO_MBCACHE)) {
4622                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
4623                 if (!sbi->s_ea_block_cache) {
4624                         ext4_msg(sb, KERN_ERR,
4625                                  "Failed to create ea_block_cache");
4626                         goto failed_mount_wq;
4627                 }
4628
4629                 if (ext4_has_feature_ea_inode(sb)) {
4630                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
4631                         if (!sbi->s_ea_inode_cache) {
4632                                 ext4_msg(sb, KERN_ERR,
4633                                          "Failed to create ea_inode_cache");
4634                                 goto failed_mount_wq;
4635                         }
4636                 }
4637         }
4638
4639         if (ext4_has_feature_verity(sb) && blocksize != PAGE_SIZE) {
4640                 ext4_msg(sb, KERN_ERR, "Unsupported blocksize for fs-verity");
4641                 goto failed_mount_wq;
4642         }
4643
4644         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) &&
4645             !ext4_has_feature_encrypt(sb)) {
4646                 ext4_set_feature_encrypt(sb);
4647                 ext4_commit_super(sb, 1);
4648         }
4649
4650         /*
4651          * Get the # of file system overhead blocks from the
4652          * superblock if present.
4653          */
4654         if (es->s_overhead_clusters)
4655                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4656         else {
4657                 err = ext4_calculate_overhead(sb);
4658                 if (err)
4659                         goto failed_mount_wq;
4660         }
4661
4662         /*
4663          * The maximum number of concurrent works can be high and
4664          * concurrency isn't really necessary.  Limit it to 1.
4665          */
4666         EXT4_SB(sb)->rsv_conversion_wq =
4667                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4668         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4669                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4670                 ret = -ENOMEM;
4671                 goto failed_mount4;
4672         }
4673
4674         /*
4675          * The jbd2_journal_load will have done any necessary log recovery,
4676          * so we can safely mount the rest of the filesystem now.
4677          */
4678
4679         root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL);
4680         if (IS_ERR(root)) {
4681                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4682                 ret = PTR_ERR(root);
4683                 root = NULL;
4684                 goto failed_mount4;
4685         }
4686         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4687                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4688                 iput(root);
4689                 goto failed_mount4;
4690         }
4691
4692 #ifdef CONFIG_UNICODE
4693         if (sbi->s_encoding)
4694                 sb->s_d_op = &ext4_dentry_ops;
4695 #endif
4696
4697         sb->s_root = d_make_root(root);
4698         if (!sb->s_root) {
4699                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4700                 ret = -ENOMEM;
4701                 goto failed_mount4;
4702         }
4703
4704         ret = ext4_setup_super(sb, es, sb_rdonly(sb));
4705         if (ret == -EROFS) {
4706                 sb->s_flags |= SB_RDONLY;
4707                 ret = 0;
4708         } else if (ret)
4709                 goto failed_mount4a;
4710
4711         ext4_set_resv_clusters(sb);
4712
4713         err = ext4_setup_system_zone(sb);
4714         if (err) {
4715                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4716                          "zone (%d)", err);
4717                 goto failed_mount4a;
4718         }
4719
4720         ext4_ext_init(sb);
4721         err = ext4_mb_init(sb);
4722         if (err) {
4723                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4724                          err);
4725                 goto failed_mount5;
4726         }
4727
4728         block = ext4_count_free_clusters(sb);
4729         ext4_free_blocks_count_set(sbi->s_es,
4730                                    EXT4_C2B(sbi, block));
4731         ext4_superblock_csum_set(sb);
4732         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4733                                   GFP_KERNEL);
4734         if (!err) {
4735                 unsigned long freei = ext4_count_free_inodes(sb);
4736                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4737                 ext4_superblock_csum_set(sb);
4738                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4739                                           GFP_KERNEL);
4740         }
4741         if (!err)
4742                 err = percpu_counter_init(&sbi->s_dirs_counter,
4743                                           ext4_count_dirs(sb), GFP_KERNEL);
4744         if (!err)
4745                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4746                                           GFP_KERNEL);
4747         if (!err)
4748                 err = percpu_init_rwsem(&sbi->s_writepages_rwsem);
4749
4750         if (err) {
4751                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4752                 goto failed_mount6;
4753         }
4754
4755         if (ext4_has_feature_flex_bg(sb))
4756                 if (!ext4_fill_flex_info(sb)) {
4757                         ext4_msg(sb, KERN_ERR,
4758                                "unable to initialize "
4759                                "flex_bg meta info!");
4760                         goto failed_mount6;
4761                 }
4762
4763         err = ext4_register_li_request(sb, first_not_zeroed);
4764         if (err)
4765                 goto failed_mount6;
4766
4767         err = ext4_register_sysfs(sb);
4768         if (err)
4769                 goto failed_mount7;
4770
4771 #ifdef CONFIG_QUOTA
4772         /* Enable quota usage during mount. */
4773         if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
4774                 err = ext4_enable_quotas(sb);
4775                 if (err)
4776                         goto failed_mount8;
4777         }
4778 #endif  /* CONFIG_QUOTA */
4779
4780         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4781         ext4_orphan_cleanup(sb, es);
4782         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4783         if (needs_recovery) {
4784                 ext4_msg(sb, KERN_INFO, "recovery complete");
4785                 ext4_mark_recovery_complete(sb, es);
4786         }
4787         if (EXT4_SB(sb)->s_journal) {
4788                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4789                         descr = " journalled data mode";
4790                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4791                         descr = " ordered data mode";
4792                 else
4793                         descr = " writeback data mode";
4794         } else
4795                 descr = "out journal";
4796
4797         if (test_opt(sb, DISCARD)) {
4798                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4799                 if (!blk_queue_discard(q))
4800                         ext4_msg(sb, KERN_WARNING,
4801                                  "mounting with \"discard\" option, but "
4802                                  "the device does not support discard");
4803         }
4804
4805         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4806                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4807                          "Opts: %.*s%s%s", descr,
4808                          (int) sizeof(sbi->s_es->s_mount_opts),
4809                          sbi->s_es->s_mount_opts,
4810                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4811
4812         if (es->s_error_count)
4813                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4814
4815         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4816         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4817         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4818         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4819
4820         kfree(orig_data);
4821         return 0;
4822
4823 cantfind_ext4:
4824         if (!silent)
4825                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4826         goto failed_mount;
4827
4828 #ifdef CONFIG_QUOTA
4829 failed_mount8:
4830         ext4_unregister_sysfs(sb);
4831 #endif
4832 failed_mount7:
4833         ext4_unregister_li_request(sb);
4834 failed_mount6:
4835         ext4_mb_release(sb);
4836         rcu_read_lock();
4837         flex_groups = rcu_dereference(sbi->s_flex_groups);
4838         if (flex_groups) {
4839                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
4840                         kvfree(flex_groups[i]);
4841                 kvfree(flex_groups);
4842         }
4843         rcu_read_unlock();
4844         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4845         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4846         percpu_counter_destroy(&sbi->s_dirs_counter);
4847         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4848         percpu_free_rwsem(&sbi->s_writepages_rwsem);
4849 failed_mount5:
4850         ext4_ext_release(sb);
4851         ext4_release_system_zone(sb);
4852 failed_mount4a:
4853         dput(sb->s_root);
4854         sb->s_root = NULL;
4855 failed_mount4:
4856         ext4_msg(sb, KERN_ERR, "mount failed");
4857         if (EXT4_SB(sb)->rsv_conversion_wq)
4858                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4859 failed_mount_wq:
4860         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
4861         sbi->s_ea_inode_cache = NULL;
4862
4863         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
4864         sbi->s_ea_block_cache = NULL;
4865
4866         if (sbi->s_journal) {
4867                 jbd2_journal_destroy(sbi->s_journal);
4868                 sbi->s_journal = NULL;
4869         }
4870 failed_mount3a:
4871         ext4_es_unregister_shrinker(sbi);
4872 failed_mount3:
4873         del_timer_sync(&sbi->s_err_report);
4874         if (sbi->s_mmp_tsk)
4875                 kthread_stop(sbi->s_mmp_tsk);
4876 failed_mount2:
4877         rcu_read_lock();
4878         group_desc = rcu_dereference(sbi->s_group_desc);
4879         for (i = 0; i < db_count; i++)
4880                 brelse(group_desc[i]);
4881         kvfree(group_desc);
4882         rcu_read_unlock();
4883 failed_mount:
4884         if (sbi->s_chksum_driver)
4885                 crypto_free_shash(sbi->s_chksum_driver);
4886
4887 #ifdef CONFIG_UNICODE
4888         utf8_unload(sbi->s_encoding);
4889 #endif
4890
4891 #ifdef CONFIG_QUOTA
4892         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4893                 kfree(get_qf_name(sb, sbi, i));
4894 #endif
4895         fscrypt_free_dummy_context(&sbi->s_dummy_enc_ctx);
4896         ext4_blkdev_remove(sbi);
4897         brelse(bh);
4898 out_fail:
4899         sb->s_fs_info = NULL;
4900         kfree(sbi->s_blockgroup_lock);
4901 out_free_base:
4902         kfree(sbi);
4903         kfree(orig_data);
4904         fs_put_dax(dax_dev);
4905         return err ? err : ret;
4906 }
4907
4908 /*
4909  * Setup any per-fs journal parameters now.  We'll do this both on
4910  * initial mount, once the journal has been initialised but before we've
4911  * done any recovery; and again on any subsequent remount.
4912  */
4913 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4914 {
4915         struct ext4_sb_info *sbi = EXT4_SB(sb);
4916
4917         journal->j_commit_interval = sbi->s_commit_interval;
4918         journal->j_min_batch_time = sbi->s_min_batch_time;
4919         journal->j_max_batch_time = sbi->s_max_batch_time;
4920
4921         write_lock(&journal->j_state_lock);
4922         if (test_opt(sb, BARRIER))
4923                 journal->j_flags |= JBD2_BARRIER;
4924         else
4925                 journal->j_flags &= ~JBD2_BARRIER;
4926         if (test_opt(sb, DATA_ERR_ABORT))
4927                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4928         else
4929                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4930         write_unlock(&journal->j_state_lock);
4931 }
4932
4933 static struct inode *ext4_get_journal_inode(struct super_block *sb,
4934                                              unsigned int journal_inum)
4935 {
4936         struct inode *journal_inode;
4937
4938         /*
4939          * Test for the existence of a valid inode on disk.  Bad things
4940          * happen if we iget() an unused inode, as the subsequent iput()
4941          * will try to delete it.
4942          */
4943         journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL);
4944         if (IS_ERR(journal_inode)) {
4945                 ext4_msg(sb, KERN_ERR, "no journal found");
4946                 return NULL;
4947         }
4948         if (!journal_inode->i_nlink) {
4949                 make_bad_inode(journal_inode);
4950                 iput(journal_inode);
4951                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4952                 return NULL;
4953         }
4954
4955         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4956                   journal_inode, journal_inode->i_size);
4957         if (!S_ISREG(journal_inode->i_mode)) {
4958                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4959                 iput(journal_inode);
4960                 return NULL;
4961         }
4962         return journal_inode;
4963 }
4964
4965 static journal_t *ext4_get_journal(struct super_block *sb,
4966                                    unsigned int journal_inum)
4967 {
4968         struct inode *journal_inode;
4969         journal_t *journal;
4970
4971         BUG_ON(!ext4_has_feature_journal(sb));
4972
4973         journal_inode = ext4_get_journal_inode(sb, journal_inum);
4974         if (!journal_inode)
4975                 return NULL;
4976
4977         journal = jbd2_journal_init_inode(journal_inode);
4978         if (!journal) {
4979                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4980                 iput(journal_inode);
4981                 return NULL;
4982         }
4983         journal->j_private = sb;
4984         ext4_init_journal_params(sb, journal);
4985         return journal;
4986 }
4987
4988 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4989                                        dev_t j_dev)
4990 {
4991         struct buffer_head *bh;
4992         journal_t *journal;
4993         ext4_fsblk_t start;
4994         ext4_fsblk_t len;
4995         int hblock, blocksize;
4996         ext4_fsblk_t sb_block;
4997         unsigned long offset;
4998         struct ext4_super_block *es;
4999         struct block_device *bdev;
5000
5001         BUG_ON(!ext4_has_feature_journal(sb));
5002
5003         bdev = ext4_blkdev_get(j_dev, sb);
5004         if (bdev == NULL)
5005                 return NULL;
5006
5007         blocksize = sb->s_blocksize;
5008         hblock = bdev_logical_block_size(bdev);
5009         if (blocksize < hblock) {
5010                 ext4_msg(sb, KERN_ERR,
5011                         "blocksize too small for journal device");
5012                 goto out_bdev;
5013         }
5014
5015         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
5016         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
5017         set_blocksize(bdev, blocksize);
5018         if (!(bh = __bread(bdev, sb_block, blocksize))) {
5019                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
5020                        "external journal");
5021                 goto out_bdev;
5022         }
5023
5024         es = (struct ext4_super_block *) (bh->b_data + offset);
5025         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
5026             !(le32_to_cpu(es->s_feature_incompat) &
5027               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
5028                 ext4_msg(sb, KERN_ERR, "external journal has "
5029                                         "bad superblock");
5030                 brelse(bh);
5031                 goto out_bdev;
5032         }
5033
5034         if ((le32_to_cpu(es->s_feature_ro_compat) &
5035              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
5036             es->s_checksum != ext4_superblock_csum(sb, es)) {
5037                 ext4_msg(sb, KERN_ERR, "external journal has "
5038                                        "corrupt superblock");
5039                 brelse(bh);
5040                 goto out_bdev;
5041         }
5042
5043         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
5044                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
5045                 brelse(bh);
5046                 goto out_bdev;
5047         }
5048
5049         len = ext4_blocks_count(es);
5050         start = sb_block + 1;
5051         brelse(bh);     /* we're done with the superblock */
5052
5053         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
5054                                         start, len, blocksize);
5055         if (!journal) {
5056                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
5057                 goto out_bdev;
5058         }
5059         journal->j_private = sb;
5060         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
5061         wait_on_buffer(journal->j_sb_buffer);
5062         if (!buffer_uptodate(journal->j_sb_buffer)) {
5063                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
5064                 goto out_journal;
5065         }
5066         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
5067                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
5068                                         "user (unsupported) - %d",
5069                         be32_to_cpu(journal->j_superblock->s_nr_users));
5070                 goto out_journal;
5071         }
5072         EXT4_SB(sb)->journal_bdev = bdev;
5073         ext4_init_journal_params(sb, journal);
5074         return journal;
5075
5076 out_journal:
5077         jbd2_journal_destroy(journal);
5078 out_bdev:
5079         ext4_blkdev_put(bdev);
5080         return NULL;
5081 }
5082
5083 static int ext4_load_journal(struct super_block *sb,
5084                              struct ext4_super_block *es,
5085                              unsigned long journal_devnum)
5086 {
5087         journal_t *journal;
5088         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
5089         dev_t journal_dev;
5090         int err = 0;
5091         int really_read_only;
5092
5093         BUG_ON(!ext4_has_feature_journal(sb));
5094
5095         if (journal_devnum &&
5096             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5097                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
5098                         "numbers have changed");
5099                 journal_dev = new_decode_dev(journal_devnum);
5100         } else
5101                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
5102
5103         really_read_only = bdev_read_only(sb->s_bdev);
5104
5105         /*
5106          * Are we loading a blank journal or performing recovery after a
5107          * crash?  For recovery, we need to check in advance whether we
5108          * can get read-write access to the device.
5109          */
5110         if (ext4_has_feature_journal_needs_recovery(sb)) {
5111                 if (sb_rdonly(sb)) {
5112                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
5113                                         "required on readonly filesystem");
5114                         if (really_read_only) {
5115                                 ext4_msg(sb, KERN_ERR, "write access "
5116                                         "unavailable, cannot proceed "
5117                                         "(try mounting with noload)");
5118                                 return -EROFS;
5119                         }
5120                         ext4_msg(sb, KERN_INFO, "write access will "
5121                                "be enabled during recovery");
5122                 }
5123         }
5124
5125         if (journal_inum && journal_dev) {
5126                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
5127                        "and inode journals!");
5128                 return -EINVAL;
5129         }
5130
5131         if (journal_inum) {
5132                 if (!(journal = ext4_get_journal(sb, journal_inum)))
5133                         return -EINVAL;
5134         } else {
5135                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
5136                         return -EINVAL;
5137         }
5138
5139         if (!(journal->j_flags & JBD2_BARRIER))
5140                 ext4_msg(sb, KERN_INFO, "barriers disabled");
5141
5142         if (!ext4_has_feature_journal_needs_recovery(sb))
5143                 err = jbd2_journal_wipe(journal, !really_read_only);
5144         if (!err) {
5145                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
5146                 if (save)
5147                         memcpy(save, ((char *) es) +
5148                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
5149                 err = jbd2_journal_load(journal);
5150                 if (save)
5151                         memcpy(((char *) es) + EXT4_S_ERR_START,
5152                                save, EXT4_S_ERR_LEN);
5153                 kfree(save);
5154         }
5155
5156         if (err) {
5157                 ext4_msg(sb, KERN_ERR, "error loading journal");
5158                 jbd2_journal_destroy(journal);
5159                 return err;
5160         }
5161
5162         EXT4_SB(sb)->s_journal = journal;
5163         ext4_clear_journal_err(sb, es);
5164
5165         if (!really_read_only && journal_devnum &&
5166             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5167                 es->s_journal_dev = cpu_to_le32(journal_devnum);
5168
5169                 /* Make sure we flush the recovery flag to disk. */
5170                 ext4_commit_super(sb, 1);
5171         }
5172
5173         return 0;
5174 }
5175
5176 static int ext4_commit_super(struct super_block *sb, int sync)
5177 {
5178         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
5179         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
5180         int error = 0;
5181
5182         if (!sbh || block_device_ejected(sb))
5183                 return error;
5184
5185         /*
5186          * The superblock bh should be mapped, but it might not be if the
5187          * device was hot-removed. Not much we can do but fail the I/O.
5188          */
5189         if (!buffer_mapped(sbh))
5190                 return error;
5191
5192         /*
5193          * If the file system is mounted read-only, don't update the
5194          * superblock write time.  This avoids updating the superblock
5195          * write time when we are mounting the root file system
5196          * read/only but we need to replay the journal; at that point,
5197          * for people who are east of GMT and who make their clock
5198          * tick in localtime for Windows bug-for-bug compatibility,
5199          * the clock is set in the future, and this will cause e2fsck
5200          * to complain and force a full file system check.
5201          */
5202         if (!(sb->s_flags & SB_RDONLY))
5203                 ext4_update_tstamp(es, s_wtime);
5204         if (sb->s_bdev->bd_part)
5205                 es->s_kbytes_written =
5206                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
5207                             ((part_stat_read(sb->s_bdev->bd_part,
5208                                              sectors[STAT_WRITE]) -
5209                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
5210         else
5211                 es->s_kbytes_written =
5212                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
5213         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
5214                 ext4_free_blocks_count_set(es,
5215                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
5216                                 &EXT4_SB(sb)->s_freeclusters_counter)));
5217         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
5218                 es->s_free_inodes_count =
5219                         cpu_to_le32(percpu_counter_sum_positive(
5220                                 &EXT4_SB(sb)->s_freeinodes_counter));
5221         BUFFER_TRACE(sbh, "marking dirty");
5222         ext4_superblock_csum_set(sb);
5223         if (sync)
5224                 lock_buffer(sbh);
5225         if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
5226                 /*
5227                  * Oh, dear.  A previous attempt to write the
5228                  * superblock failed.  This could happen because the
5229                  * USB device was yanked out.  Or it could happen to
5230                  * be a transient write error and maybe the block will
5231                  * be remapped.  Nothing we can do but to retry the
5232                  * write and hope for the best.
5233                  */
5234                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
5235                        "superblock detected");
5236                 clear_buffer_write_io_error(sbh);
5237                 set_buffer_uptodate(sbh);
5238         }
5239         mark_buffer_dirty(sbh);
5240         if (sync) {
5241                 unlock_buffer(sbh);
5242                 error = __sync_dirty_buffer(sbh,
5243                         REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
5244                 if (buffer_write_io_error(sbh)) {
5245                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
5246                                "superblock");
5247                         clear_buffer_write_io_error(sbh);
5248                         set_buffer_uptodate(sbh);
5249                 }
5250         }
5251         return error;
5252 }
5253
5254 /*
5255  * Have we just finished recovery?  If so, and if we are mounting (or
5256  * remounting) the filesystem readonly, then we will end up with a
5257  * consistent fs on disk.  Record that fact.
5258  */
5259 static void ext4_mark_recovery_complete(struct super_block *sb,
5260                                         struct ext4_super_block *es)
5261 {
5262         journal_t *journal = EXT4_SB(sb)->s_journal;
5263
5264         if (!ext4_has_feature_journal(sb)) {
5265                 BUG_ON(journal != NULL);
5266                 return;
5267         }
5268         jbd2_journal_lock_updates(journal);
5269         if (jbd2_journal_flush(journal) < 0)
5270                 goto out;
5271
5272         if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) {
5273                 ext4_clear_feature_journal_needs_recovery(sb);
5274                 ext4_commit_super(sb, 1);
5275         }
5276
5277 out:
5278         jbd2_journal_unlock_updates(journal);
5279 }
5280
5281 /*
5282  * If we are mounting (or read-write remounting) a filesystem whose journal
5283  * has recorded an error from a previous lifetime, move that error to the
5284  * main filesystem now.
5285  */
5286 static void ext4_clear_journal_err(struct super_block *sb,
5287                                    struct ext4_super_block *es)
5288 {
5289         journal_t *journal;
5290         int j_errno;
5291         const char *errstr;
5292
5293         BUG_ON(!ext4_has_feature_journal(sb));
5294
5295         journal = EXT4_SB(sb)->s_journal;
5296
5297         /*
5298          * Now check for any error status which may have been recorded in the
5299          * journal by a prior ext4_error() or ext4_abort()
5300          */
5301
5302         j_errno = jbd2_journal_errno(journal);
5303         if (j_errno) {
5304                 char nbuf[16];
5305
5306                 errstr = ext4_decode_error(sb, j_errno, nbuf);
5307                 ext4_warning(sb, "Filesystem error recorded "
5308                              "from previous mount: %s", errstr);
5309                 ext4_warning(sb, "Marking fs in need of filesystem check.");
5310
5311                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
5312                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
5313                 ext4_commit_super(sb, 1);
5314
5315                 jbd2_journal_clear_err(journal);
5316                 jbd2_journal_update_sb_errno(journal);
5317         }
5318 }
5319
5320 /*
5321  * Force the running and committing transactions to commit,
5322  * and wait on the commit.
5323  */
5324 int ext4_force_commit(struct super_block *sb)
5325 {
5326         journal_t *journal;
5327
5328         if (sb_rdonly(sb))
5329                 return 0;
5330
5331         journal = EXT4_SB(sb)->s_journal;
5332         return ext4_journal_force_commit(journal);
5333 }
5334
5335 static int ext4_sync_fs(struct super_block *sb, int wait)
5336 {
5337         int ret = 0;
5338         tid_t target;
5339         bool needs_barrier = false;
5340         struct ext4_sb_info *sbi = EXT4_SB(sb);
5341
5342         if (unlikely(ext4_forced_shutdown(sbi)))
5343                 return 0;
5344
5345         trace_ext4_sync_fs(sb, wait);
5346         flush_workqueue(sbi->rsv_conversion_wq);
5347         /*
5348          * Writeback quota in non-journalled quota case - journalled quota has
5349          * no dirty dquots
5350          */
5351         dquot_writeback_dquots(sb, -1);
5352         /*
5353          * Data writeback is possible w/o journal transaction, so barrier must
5354          * being sent at the end of the function. But we can skip it if
5355          * transaction_commit will do it for us.
5356          */
5357         if (sbi->s_journal) {
5358                 target = jbd2_get_latest_transaction(sbi->s_journal);
5359                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
5360                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
5361                         needs_barrier = true;
5362
5363                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
5364                         if (wait)
5365                                 ret = jbd2_log_wait_commit(sbi->s_journal,
5366                                                            target);
5367                 }
5368         } else if (wait && test_opt(sb, BARRIER))
5369                 needs_barrier = true;
5370         if (needs_barrier) {
5371                 int err;
5372                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL);
5373                 if (!ret)
5374                         ret = err;
5375         }
5376
5377         return ret;
5378 }
5379
5380 /*
5381  * LVM calls this function before a (read-only) snapshot is created.  This
5382  * gives us a chance to flush the journal completely and mark the fs clean.
5383  *
5384  * Note that only this function cannot bring a filesystem to be in a clean
5385  * state independently. It relies on upper layer to stop all data & metadata
5386  * modifications.
5387  */
5388 static int ext4_freeze(struct super_block *sb)
5389 {
5390         int error = 0;
5391         journal_t *journal;
5392
5393         if (sb_rdonly(sb))
5394                 return 0;
5395
5396         journal = EXT4_SB(sb)->s_journal;
5397
5398         if (journal) {
5399                 /* Now we set up the journal barrier. */
5400                 jbd2_journal_lock_updates(journal);
5401
5402                 /*
5403                  * Don't clear the needs_recovery flag if we failed to
5404                  * flush the journal.
5405                  */
5406                 error = jbd2_journal_flush(journal);
5407                 if (error < 0)
5408                         goto out;
5409
5410                 /* Journal blocked and flushed, clear needs_recovery flag. */
5411                 ext4_clear_feature_journal_needs_recovery(sb);
5412         }
5413
5414         error = ext4_commit_super(sb, 1);
5415 out:
5416         if (journal)
5417                 /* we rely on upper layer to stop further updates */
5418                 jbd2_journal_unlock_updates(journal);
5419         return error;
5420 }
5421
5422 /*
5423  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
5424  * flag here, even though the filesystem is not technically dirty yet.
5425  */
5426 static int ext4_unfreeze(struct super_block *sb)
5427 {
5428         if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
5429                 return 0;
5430
5431         if (EXT4_SB(sb)->s_journal) {
5432                 /* Reset the needs_recovery flag before the fs is unlocked. */
5433                 ext4_set_feature_journal_needs_recovery(sb);
5434         }
5435
5436         ext4_commit_super(sb, 1);
5437         return 0;
5438 }
5439
5440 /*
5441  * Structure to save mount options for ext4_remount's benefit
5442  */
5443 struct ext4_mount_options {
5444         unsigned long s_mount_opt;
5445         unsigned long s_mount_opt2;
5446         kuid_t s_resuid;
5447         kgid_t s_resgid;
5448         unsigned long s_commit_interval;
5449         u32 s_min_batch_time, s_max_batch_time;
5450 #ifdef CONFIG_QUOTA
5451         int s_jquota_fmt;
5452         char *s_qf_names[EXT4_MAXQUOTAS];
5453 #endif
5454 };
5455
5456 static int ext4_remount(struct super_block *sb, int *flags, char *data)
5457 {
5458         struct ext4_super_block *es;
5459         struct ext4_sb_info *sbi = EXT4_SB(sb);
5460         unsigned long old_sb_flags;
5461         struct ext4_mount_options old_opts;
5462         int enable_quota = 0;
5463         ext4_group_t g;
5464         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
5465         int err = 0;
5466 #ifdef CONFIG_QUOTA
5467         int i, j;
5468         char *to_free[EXT4_MAXQUOTAS];
5469 #endif
5470         char *orig_data = kstrdup(data, GFP_KERNEL);
5471
5472         if (data && !orig_data)
5473                 return -ENOMEM;
5474
5475         /* Store the original options */
5476         old_sb_flags = sb->s_flags;
5477         old_opts.s_mount_opt = sbi->s_mount_opt;
5478         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
5479         old_opts.s_resuid = sbi->s_resuid;
5480         old_opts.s_resgid = sbi->s_resgid;
5481         old_opts.s_commit_interval = sbi->s_commit_interval;
5482         old_opts.s_min_batch_time = sbi->s_min_batch_time;
5483         old_opts.s_max_batch_time = sbi->s_max_batch_time;
5484 #ifdef CONFIG_QUOTA
5485         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
5486         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5487                 if (sbi->s_qf_names[i]) {
5488                         char *qf_name = get_qf_name(sb, sbi, i);
5489
5490                         old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL);
5491                         if (!old_opts.s_qf_names[i]) {
5492                                 for (j = 0; j < i; j++)
5493                                         kfree(old_opts.s_qf_names[j]);
5494                                 kfree(orig_data);
5495                                 return -ENOMEM;
5496                         }
5497                 } else
5498                         old_opts.s_qf_names[i] = NULL;
5499 #endif
5500         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
5501                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
5502
5503         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
5504                 err = -EINVAL;
5505                 goto restore_opts;
5506         }
5507
5508         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
5509             test_opt(sb, JOURNAL_CHECKSUM)) {
5510                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
5511                          "during remount not supported; ignoring");
5512                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
5513         }
5514
5515         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
5516                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
5517                         ext4_msg(sb, KERN_ERR, "can't mount with "
5518                                  "both data=journal and delalloc");
5519                         err = -EINVAL;
5520                         goto restore_opts;
5521                 }
5522                 if (test_opt(sb, DIOREAD_NOLOCK)) {
5523                         ext4_msg(sb, KERN_ERR, "can't mount with "
5524                                  "both data=journal and dioread_nolock");
5525                         err = -EINVAL;
5526                         goto restore_opts;
5527                 }
5528         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
5529                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5530                         ext4_msg(sb, KERN_ERR, "can't mount with "
5531                                 "journal_async_commit in data=ordered mode");
5532                         err = -EINVAL;
5533                         goto restore_opts;
5534                 }
5535         }
5536
5537         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
5538                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
5539                 err = -EINVAL;
5540                 goto restore_opts;
5541         }
5542
5543         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
5544                 ext4_abort(sb, EXT4_ERR_ESHUTDOWN, "Abort forced by user");
5545
5546         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
5547                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
5548
5549         es = sbi->s_es;
5550
5551         if (sbi->s_journal) {
5552                 ext4_init_journal_params(sb, sbi->s_journal);
5553                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
5554         }
5555
5556         if (*flags & SB_LAZYTIME)
5557                 sb->s_flags |= SB_LAZYTIME;
5558
5559         if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) {
5560                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
5561                         err = -EROFS;
5562                         goto restore_opts;
5563                 }
5564
5565                 if (*flags & SB_RDONLY) {
5566                         err = sync_filesystem(sb);
5567                         if (err < 0)
5568                                 goto restore_opts;
5569                         err = dquot_suspend(sb, -1);
5570                         if (err < 0)
5571                                 goto restore_opts;
5572
5573                         /*
5574                          * First of all, the unconditional stuff we have to do
5575                          * to disable replay of the journal when we next remount
5576                          */
5577                         sb->s_flags |= SB_RDONLY;
5578
5579                         /*
5580                          * OK, test if we are remounting a valid rw partition
5581                          * readonly, and if so set the rdonly flag and then
5582                          * mark the partition as valid again.
5583                          */
5584                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5585                             (sbi->s_mount_state & EXT4_VALID_FS))
5586                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5587
5588                         if (sbi->s_journal)
5589                                 ext4_mark_recovery_complete(sb, es);
5590                         if (sbi->s_mmp_tsk)
5591                                 kthread_stop(sbi->s_mmp_tsk);
5592                 } else {
5593                         /* Make sure we can mount this feature set readwrite */
5594                         if (ext4_has_feature_readonly(sb) ||
5595                             !ext4_feature_set_ok(sb, 0)) {
5596                                 err = -EROFS;
5597                                 goto restore_opts;
5598                         }
5599                         /*
5600                          * Make sure the group descriptor checksums
5601                          * are sane.  If they aren't, refuse to remount r/w.
5602                          */
5603                         for (g = 0; g < sbi->s_groups_count; g++) {
5604                                 struct ext4_group_desc *gdp =
5605                                         ext4_get_group_desc(sb, g, NULL);
5606
5607                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5608                                         ext4_msg(sb, KERN_ERR,
5609                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5610                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
5611                                                le16_to_cpu(gdp->bg_checksum));
5612                                         err = -EFSBADCRC;
5613                                         goto restore_opts;
5614                                 }
5615                         }
5616
5617                         /*
5618                          * If we have an unprocessed orphan list hanging
5619                          * around from a previously readonly bdev mount,
5620                          * require a full umount/remount for now.
5621                          */
5622                         if (es->s_last_orphan) {
5623                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5624                                        "remount RDWR because of unprocessed "
5625                                        "orphan inode list.  Please "
5626                                        "umount/remount instead");
5627                                 err = -EINVAL;
5628                                 goto restore_opts;
5629                         }
5630
5631                         /*
5632                          * Mounting a RDONLY partition read-write, so reread
5633                          * and store the current valid flag.  (It may have
5634                          * been changed by e2fsck since we originally mounted
5635                          * the partition.)
5636                          */
5637                         if (sbi->s_journal)
5638                                 ext4_clear_journal_err(sb, es);
5639                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5640
5641                         err = ext4_setup_super(sb, es, 0);
5642                         if (err)
5643                                 goto restore_opts;
5644
5645                         sb->s_flags &= ~SB_RDONLY;
5646                         if (ext4_has_feature_mmp(sb))
5647                                 if (ext4_multi_mount_protect(sb,
5648                                                 le64_to_cpu(es->s_mmp_block))) {
5649                                         err = -EROFS;
5650                                         goto restore_opts;
5651                                 }
5652                         enable_quota = 1;
5653                 }
5654         }
5655
5656         /*
5657          * Reinitialize lazy itable initialization thread based on
5658          * current settings
5659          */
5660         if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
5661                 ext4_unregister_li_request(sb);
5662         else {
5663                 ext4_group_t first_not_zeroed;
5664                 first_not_zeroed = ext4_has_uninit_itable(sb);
5665                 ext4_register_li_request(sb, first_not_zeroed);
5666         }
5667
5668         ext4_setup_system_zone(sb);
5669         if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) {
5670                 err = ext4_commit_super(sb, 1);
5671                 if (err)
5672                         goto restore_opts;
5673         }
5674
5675 #ifdef CONFIG_QUOTA
5676         /* Release old quota file names */
5677         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5678                 kfree(old_opts.s_qf_names[i]);
5679         if (enable_quota) {
5680                 if (sb_any_quota_suspended(sb))
5681                         dquot_resume(sb, -1);
5682                 else if (ext4_has_feature_quota(sb)) {
5683                         err = ext4_enable_quotas(sb);
5684                         if (err)
5685                                 goto restore_opts;
5686                 }
5687         }
5688 #endif
5689
5690         *flags = (*flags & ~SB_LAZYTIME) | (sb->s_flags & SB_LAZYTIME);
5691         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5692         kfree(orig_data);
5693         return 0;
5694
5695 restore_opts:
5696         sb->s_flags = old_sb_flags;
5697         sbi->s_mount_opt = old_opts.s_mount_opt;
5698         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5699         sbi->s_resuid = old_opts.s_resuid;
5700         sbi->s_resgid = old_opts.s_resgid;
5701         sbi->s_commit_interval = old_opts.s_commit_interval;
5702         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5703         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5704 #ifdef CONFIG_QUOTA
5705         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5706         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5707                 to_free[i] = get_qf_name(sb, sbi, i);
5708                 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]);
5709         }
5710         synchronize_rcu();
5711         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5712                 kfree(to_free[i]);
5713 #endif
5714         kfree(orig_data);
5715         return err;
5716 }
5717
5718 #ifdef CONFIG_QUOTA
5719 static int ext4_statfs_project(struct super_block *sb,
5720                                kprojid_t projid, struct kstatfs *buf)
5721 {
5722         struct kqid qid;
5723         struct dquot *dquot;
5724         u64 limit;
5725         u64 curblock;
5726
5727         qid = make_kqid_projid(projid);
5728         dquot = dqget(sb, qid);
5729         if (IS_ERR(dquot))
5730                 return PTR_ERR(dquot);
5731         spin_lock(&dquot->dq_dqb_lock);
5732
5733         limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit,
5734                              dquot->dq_dqb.dqb_bhardlimit);
5735         limit >>= sb->s_blocksize_bits;
5736
5737         if (limit && buf->f_blocks > limit) {
5738                 curblock = (dquot->dq_dqb.dqb_curspace +
5739                             dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits;
5740                 buf->f_blocks = limit;
5741                 buf->f_bfree = buf->f_bavail =
5742                         (buf->f_blocks > curblock) ?
5743                          (buf->f_blocks - curblock) : 0;
5744         }
5745
5746         limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit,
5747                              dquot->dq_dqb.dqb_ihardlimit);
5748         if (limit && buf->f_files > limit) {
5749                 buf->f_files = limit;
5750                 buf->f_ffree =
5751                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
5752                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
5753         }
5754
5755         spin_unlock(&dquot->dq_dqb_lock);
5756         dqput(dquot);
5757         return 0;
5758 }
5759 #endif
5760
5761 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5762 {
5763         struct super_block *sb = dentry->d_sb;
5764         struct ext4_sb_info *sbi = EXT4_SB(sb);
5765         struct ext4_super_block *es = sbi->s_es;
5766         ext4_fsblk_t overhead = 0, resv_blocks;
5767         u64 fsid;
5768         s64 bfree;
5769         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5770
5771         if (!test_opt(sb, MINIX_DF))
5772                 overhead = sbi->s_overhead;
5773
5774         buf->f_type = EXT4_SUPER_MAGIC;
5775         buf->f_bsize = sb->s_blocksize;
5776         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5777         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5778                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5779         /* prevent underflow in case that few free space is available */
5780         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5781         buf->f_bavail = buf->f_bfree -
5782                         (ext4_r_blocks_count(es) + resv_blocks);
5783         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5784                 buf->f_bavail = 0;
5785         buf->f_files = le32_to_cpu(es->s_inodes_count);
5786         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5787         buf->f_namelen = EXT4_NAME_LEN;
5788         fsid = le64_to_cpup((void *)es->s_uuid) ^
5789                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5790         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5791         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5792
5793 #ifdef CONFIG_QUOTA
5794         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
5795             sb_has_quota_limits_enabled(sb, PRJQUOTA))
5796                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
5797 #endif
5798         return 0;
5799 }
5800
5801
5802 #ifdef CONFIG_QUOTA
5803
5804 /*
5805  * Helper functions so that transaction is started before we acquire dqio_sem
5806  * to keep correct lock ordering of transaction > dqio_sem
5807  */
5808 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5809 {
5810         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5811 }
5812
5813 static int ext4_write_dquot(struct dquot *dquot)
5814 {
5815         int ret, err;
5816         handle_t *handle;
5817         struct inode *inode;
5818
5819         inode = dquot_to_inode(dquot);
5820         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5821                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5822         if (IS_ERR(handle))
5823                 return PTR_ERR(handle);
5824         ret = dquot_commit(dquot);
5825         err = ext4_journal_stop(handle);
5826         if (!ret)
5827                 ret = err;
5828         return ret;
5829 }
5830
5831 static int ext4_acquire_dquot(struct dquot *dquot)
5832 {
5833         int ret, err;
5834         handle_t *handle;
5835
5836         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5837                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5838         if (IS_ERR(handle))
5839                 return PTR_ERR(handle);
5840         ret = dquot_acquire(dquot);
5841         err = ext4_journal_stop(handle);
5842         if (!ret)
5843                 ret = err;
5844         return ret;
5845 }
5846
5847 static int ext4_release_dquot(struct dquot *dquot)
5848 {
5849         int ret, err;
5850         handle_t *handle;
5851
5852         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5853                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5854         if (IS_ERR(handle)) {
5855                 /* Release dquot anyway to avoid endless cycle in dqput() */
5856                 dquot_release(dquot);
5857                 return PTR_ERR(handle);
5858         }
5859         ret = dquot_release(dquot);
5860         err = ext4_journal_stop(handle);
5861         if (!ret)
5862                 ret = err;
5863         return ret;
5864 }
5865
5866 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5867 {
5868         struct super_block *sb = dquot->dq_sb;
5869         struct ext4_sb_info *sbi = EXT4_SB(sb);
5870
5871         /* Are we journaling quotas? */
5872         if (ext4_has_feature_quota(sb) ||
5873             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5874                 dquot_mark_dquot_dirty(dquot);
5875                 return ext4_write_dquot(dquot);
5876         } else {
5877                 return dquot_mark_dquot_dirty(dquot);
5878         }
5879 }
5880
5881 static int ext4_write_info(struct super_block *sb, int type)
5882 {
5883         int ret, err;
5884         handle_t *handle;
5885
5886         /* Data block + inode block */
5887         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5888         if (IS_ERR(handle))
5889                 return PTR_ERR(handle);
5890         ret = dquot_commit_info(sb, type);
5891         err = ext4_journal_stop(handle);
5892         if (!ret)
5893                 ret = err;
5894         return ret;
5895 }
5896
5897 /*
5898  * Turn on quotas during mount time - we need to find
5899  * the quota file and such...
5900  */
5901 static int ext4_quota_on_mount(struct super_block *sb, int type)
5902 {
5903         return dquot_quota_on_mount(sb, get_qf_name(sb, EXT4_SB(sb), type),
5904                                         EXT4_SB(sb)->s_jquota_fmt, type);
5905 }
5906
5907 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5908 {
5909         struct ext4_inode_info *ei = EXT4_I(inode);
5910
5911         /* The first argument of lockdep_set_subclass has to be
5912          * *exactly* the same as the argument to init_rwsem() --- in
5913          * this case, in init_once() --- or lockdep gets unhappy
5914          * because the name of the lock is set using the
5915          * stringification of the argument to init_rwsem().
5916          */
5917         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5918         lockdep_set_subclass(&ei->i_data_sem, subclass);
5919 }
5920
5921 /*
5922  * Standard function to be called on quota_on
5923  */
5924 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5925                          const struct path *path)
5926 {
5927         int err;
5928
5929         if (!test_opt(sb, QUOTA))
5930                 return -EINVAL;
5931
5932         /* Quotafile not on the same filesystem? */
5933         if (path->dentry->d_sb != sb)
5934                 return -EXDEV;
5935         /* Journaling quota? */
5936         if (EXT4_SB(sb)->s_qf_names[type]) {
5937                 /* Quotafile not in fs root? */
5938                 if (path->dentry->d_parent != sb->s_root)
5939                         ext4_msg(sb, KERN_WARNING,
5940                                 "Quota file not on filesystem root. "
5941                                 "Journaled quota will not work");
5942                 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
5943         } else {
5944                 /*
5945                  * Clear the flag just in case mount options changed since
5946                  * last time.
5947                  */
5948                 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
5949         }
5950
5951         /*
5952          * When we journal data on quota file, we have to flush journal to see
5953          * all updates to the file when we bypass pagecache...
5954          */
5955         if (EXT4_SB(sb)->s_journal &&
5956             ext4_should_journal_data(d_inode(path->dentry))) {
5957                 /*
5958                  * We don't need to lock updates but journal_flush() could
5959                  * otherwise be livelocked...
5960                  */
5961                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5962                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5963                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5964                 if (err)
5965                         return err;
5966         }
5967
5968         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5969         err = dquot_quota_on(sb, type, format_id, path);
5970         if (err) {
5971                 lockdep_set_quota_inode(path->dentry->d_inode,
5972                                              I_DATA_SEM_NORMAL);
5973         } else {
5974                 struct inode *inode = d_inode(path->dentry);
5975                 handle_t *handle;
5976
5977                 /*
5978                  * Set inode flags to prevent userspace from messing with quota
5979                  * files. If this fails, we return success anyway since quotas
5980                  * are already enabled and this is not a hard failure.
5981                  */
5982                 inode_lock(inode);
5983                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5984                 if (IS_ERR(handle))
5985                         goto unlock_inode;
5986                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
5987                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
5988                                 S_NOATIME | S_IMMUTABLE);
5989                 err = ext4_mark_inode_dirty(handle, inode);
5990                 ext4_journal_stop(handle);
5991         unlock_inode:
5992                 inode_unlock(inode);
5993         }
5994         return err;
5995 }
5996
5997 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5998                              unsigned int flags)
5999 {
6000         int err;
6001         struct inode *qf_inode;
6002         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
6003                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
6004                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
6005                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
6006         };
6007
6008         BUG_ON(!ext4_has_feature_quota(sb));
6009
6010         if (!qf_inums[type])
6011                 return -EPERM;
6012
6013         qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL);
6014         if (IS_ERR(qf_inode)) {
6015                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
6016                 return PTR_ERR(qf_inode);
6017         }
6018
6019         /* Don't account quota for quota files to avoid recursion */
6020         qf_inode->i_flags |= S_NOQUOTA;
6021         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
6022         err = dquot_load_quota_inode(qf_inode, type, format_id, flags);
6023         if (err)
6024                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
6025         iput(qf_inode);
6026
6027         return err;
6028 }
6029
6030 /* Enable usage tracking for all quota types. */
6031 static int ext4_enable_quotas(struct super_block *sb)
6032 {
6033         int type, err = 0;
6034         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
6035                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
6036                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
6037                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
6038         };
6039         bool quota_mopt[EXT4_MAXQUOTAS] = {
6040                 test_opt(sb, USRQUOTA),
6041                 test_opt(sb, GRPQUOTA),
6042                 test_opt(sb, PRJQUOTA),
6043         };
6044
6045         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
6046         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
6047                 if (qf_inums[type]) {
6048                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
6049                                 DQUOT_USAGE_ENABLED |
6050                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
6051                         if (err) {
6052                                 ext4_warning(sb,
6053                                         "Failed to enable quota tracking "
6054                                         "(type=%d, err=%d). Please run "
6055                                         "e2fsck to fix.", type, err);
6056                                 for (type--; type >= 0; type--)
6057                                         dquot_quota_off(sb, type);
6058
6059                                 return err;
6060                         }
6061                 }
6062         }
6063         return 0;
6064 }
6065
6066 static int ext4_quota_off(struct super_block *sb, int type)
6067 {
6068         struct inode *inode = sb_dqopt(sb)->files[type];
6069         handle_t *handle;
6070         int err;
6071
6072         /* Force all delayed allocation blocks to be allocated.
6073          * Caller already holds s_umount sem */
6074         if (test_opt(sb, DELALLOC))
6075                 sync_filesystem(sb);
6076
6077         if (!inode || !igrab(inode))
6078                 goto out;
6079
6080         err = dquot_quota_off(sb, type);
6081         if (err || ext4_has_feature_quota(sb))
6082                 goto out_put;
6083
6084         inode_lock(inode);
6085         /*
6086          * Update modification times of quota files when userspace can
6087          * start looking at them. If we fail, we return success anyway since
6088          * this is not a hard failure and quotas are already disabled.
6089          */
6090         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
6091         if (IS_ERR(handle)) {
6092                 err = PTR_ERR(handle);
6093                 goto out_unlock;
6094         }
6095         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
6096         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
6097         inode->i_mtime = inode->i_ctime = current_time(inode);
6098         err = ext4_mark_inode_dirty(handle, inode);
6099         ext4_journal_stop(handle);
6100 out_unlock:
6101         inode_unlock(inode);
6102 out_put:
6103         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
6104         iput(inode);
6105         return err;
6106 out:
6107         return dquot_quota_off(sb, type);
6108 }
6109
6110 /* Read data from quotafile - avoid pagecache and such because we cannot afford
6111  * acquiring the locks... As quota files are never truncated and quota code
6112  * itself serializes the operations (and no one else should touch the files)
6113  * we don't have to be afraid of races */
6114 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
6115                                size_t len, loff_t off)
6116 {
6117         struct inode *inode = sb_dqopt(sb)->files[type];
6118         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
6119         int offset = off & (sb->s_blocksize - 1);
6120         int tocopy;
6121         size_t toread;
6122         struct buffer_head *bh;
6123         loff_t i_size = i_size_read(inode);
6124
6125         if (off > i_size)
6126                 return 0;
6127         if (off+len > i_size)
6128                 len = i_size-off;
6129         toread = len;
6130         while (toread > 0) {
6131                 tocopy = sb->s_blocksize - offset < toread ?
6132                                 sb->s_blocksize - offset : toread;
6133                 bh = ext4_bread(NULL, inode, blk, 0);
6134                 if (IS_ERR(bh))
6135                         return PTR_ERR(bh);
6136                 if (!bh)        /* A hole? */
6137                         memset(data, 0, tocopy);
6138                 else
6139                         memcpy(data, bh->b_data+offset, tocopy);
6140                 brelse(bh);
6141                 offset = 0;
6142                 toread -= tocopy;
6143                 data += tocopy;
6144                 blk++;
6145         }
6146         return len;
6147 }
6148
6149 /* Write to quotafile (we know the transaction is already started and has
6150  * enough credits) */
6151 static ssize_t ext4_quota_write(struct super_block *sb, int type,
6152                                 const char *data, size_t len, loff_t off)
6153 {
6154         struct inode *inode = sb_dqopt(sb)->files[type];
6155         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
6156         int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1);
6157         int retries = 0;
6158         struct buffer_head *bh;
6159         handle_t *handle = journal_current_handle();
6160
6161         if (EXT4_SB(sb)->s_journal && !handle) {
6162                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
6163                         " cancelled because transaction is not started",
6164                         (unsigned long long)off, (unsigned long long)len);
6165                 return -EIO;
6166         }
6167         /*
6168          * Since we account only one data block in transaction credits,
6169          * then it is impossible to cross a block boundary.
6170          */
6171         if (sb->s_blocksize - offset < len) {
6172                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
6173                         " cancelled because not block aligned",
6174                         (unsigned long long)off, (unsigned long long)len);
6175                 return -EIO;
6176         }
6177
6178         do {
6179                 bh = ext4_bread(handle, inode, blk,
6180                                 EXT4_GET_BLOCKS_CREATE |
6181                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
6182         } while (PTR_ERR(bh) == -ENOSPC &&
6183                  ext4_should_retry_alloc(inode->i_sb, &retries));
6184         if (IS_ERR(bh))
6185                 return PTR_ERR(bh);
6186         if (!bh)
6187                 goto out;
6188         BUFFER_TRACE(bh, "get write access");
6189         err = ext4_journal_get_write_access(handle, bh);
6190         if (err) {
6191                 brelse(bh);
6192                 return err;
6193         }
6194         lock_buffer(bh);
6195         memcpy(bh->b_data+offset, data, len);
6196         flush_dcache_page(bh->b_page);
6197         unlock_buffer(bh);
6198         err = ext4_handle_dirty_metadata(handle, NULL, bh);
6199         brelse(bh);
6200 out:
6201         if (inode->i_size < off + len) {
6202                 i_size_write(inode, off + len);
6203                 EXT4_I(inode)->i_disksize = inode->i_size;
6204                 err2 = ext4_mark_inode_dirty(handle, inode);
6205                 if (unlikely(err2 && !err))
6206                         err = err2;
6207         }
6208         return err ? err : len;
6209 }
6210 #endif
6211
6212 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
6213                        const char *dev_name, void *data)
6214 {
6215         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
6216 }
6217
6218 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
6219 static inline void register_as_ext2(void)
6220 {
6221         int err = register_filesystem(&ext2_fs_type);
6222         if (err)
6223                 printk(KERN_WARNING
6224                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
6225 }
6226
6227 static inline void unregister_as_ext2(void)
6228 {
6229         unregister_filesystem(&ext2_fs_type);
6230 }
6231
6232 static inline int ext2_feature_set_ok(struct super_block *sb)
6233 {
6234         if (ext4_has_unknown_ext2_incompat_features(sb))
6235                 return 0;
6236         if (sb_rdonly(sb))
6237                 return 1;
6238         if (ext4_has_unknown_ext2_ro_compat_features(sb))
6239                 return 0;
6240         return 1;
6241 }
6242 #else
6243 static inline void register_as_ext2(void) { }
6244 static inline void unregister_as_ext2(void) { }
6245 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
6246 #endif
6247
6248 static inline void register_as_ext3(void)
6249 {
6250         int err = register_filesystem(&ext3_fs_type);
6251         if (err)
6252                 printk(KERN_WARNING
6253                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
6254 }
6255
6256 static inline void unregister_as_ext3(void)
6257 {
6258         unregister_filesystem(&ext3_fs_type);
6259 }
6260
6261 static inline int ext3_feature_set_ok(struct super_block *sb)
6262 {
6263         if (ext4_has_unknown_ext3_incompat_features(sb))
6264                 return 0;
6265         if (!ext4_has_feature_journal(sb))
6266                 return 0;
6267         if (sb_rdonly(sb))
6268                 return 1;
6269         if (ext4_has_unknown_ext3_ro_compat_features(sb))
6270                 return 0;
6271         return 1;
6272 }
6273
6274 static struct file_system_type ext4_fs_type = {
6275         .owner          = THIS_MODULE,
6276         .name           = "ext4",
6277         .mount          = ext4_mount,
6278         .kill_sb        = kill_block_super,
6279         .fs_flags       = FS_REQUIRES_DEV,
6280 };
6281 MODULE_ALIAS_FS("ext4");
6282
6283 /* Shared across all ext4 file systems */
6284 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
6285
6286 static int __init ext4_init_fs(void)
6287 {
6288         int i, err;
6289
6290         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
6291         ext4_li_info = NULL;
6292         mutex_init(&ext4_li_mtx);
6293
6294         /* Build-time check for flags consistency */
6295         ext4_check_flag_values();
6296
6297         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
6298                 init_waitqueue_head(&ext4__ioend_wq[i]);
6299
6300         err = ext4_init_es();
6301         if (err)
6302                 return err;
6303
6304         err = ext4_init_pending();
6305         if (err)
6306                 goto out7;
6307
6308         err = ext4_init_post_read_processing();
6309         if (err)
6310                 goto out6;
6311
6312         err = ext4_init_pageio();
6313         if (err)
6314                 goto out5;
6315
6316         err = ext4_init_system_zone();
6317         if (err)
6318                 goto out4;
6319
6320         err = ext4_init_sysfs();
6321         if (err)
6322                 goto out3;
6323
6324         err = ext4_init_mballoc();
6325         if (err)
6326                 goto out2;
6327         err = init_inodecache();
6328         if (err)
6329                 goto out1;
6330         register_as_ext3();
6331         register_as_ext2();
6332         err = register_filesystem(&ext4_fs_type);
6333         if (err)
6334                 goto out;
6335
6336         return 0;
6337 out:
6338         unregister_as_ext2();
6339         unregister_as_ext3();
6340         destroy_inodecache();
6341 out1:
6342         ext4_exit_mballoc();
6343 out2:
6344         ext4_exit_sysfs();
6345 out3:
6346         ext4_exit_system_zone();
6347 out4:
6348         ext4_exit_pageio();
6349 out5:
6350         ext4_exit_post_read_processing();
6351 out6:
6352         ext4_exit_pending();
6353 out7:
6354         ext4_exit_es();
6355
6356         return err;
6357 }
6358
6359 static void __exit ext4_exit_fs(void)
6360 {
6361         ext4_destroy_lazyinit_thread();
6362         unregister_as_ext2();
6363         unregister_as_ext3();
6364         unregister_filesystem(&ext4_fs_type);
6365         destroy_inodecache();
6366         ext4_exit_mballoc();
6367         ext4_exit_sysfs();
6368         ext4_exit_system_zone();
6369         ext4_exit_pageio();
6370         ext4_exit_post_read_processing();
6371         ext4_exit_es();
6372         ext4_exit_pending();
6373 }
6374
6375 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
6376 MODULE_DESCRIPTION("Fourth Extended Filesystem");
6377 MODULE_LICENSE("GPL");
6378 MODULE_SOFTDEP("pre: crc32c");
6379 module_init(ext4_init_fs)
6380 module_exit(ext4_exit_fs)