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_sem -> 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_sem
 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_sem
 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                 if (EXT4_SB(sb)->s_journal &&
 526                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 527                         return;
 528                 panic("EXT4-fs (device %s): panic forced after error\n",
 529                         sb->s_id);
 530         }
 531 }
 532
 533 #define ext4_error_ratelimit(sb)                                        \
 534                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 535                              "EXT4-fs error")
 536
 537 void __ext4_error(struct super_block *sb, const char *function,
 538                   unsigned int line, int error, __u64 block,
 539                   const char *fmt, ...)
 540 {
 541         struct va_format vaf;
 542         va_list args;
 543
 544         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 545                 return;
 546
 547         trace_ext4_error(sb, function, line);
 548         if (ext4_error_ratelimit(sb)) {
 549                 va_start(args, fmt);
 550                 vaf.fmt = fmt;
 551                 vaf.va = &args;
 552                 printk(KERN_CRIT
 553                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 554                        sb->s_id, function, line, current->comm, &vaf);
 555                 va_end(args);
 556         }
 557         save_error_info(sb, error, 0, block, function, line);
 558         ext4_handle_error(sb);
 559 }
 560
 561 void __ext4_error_inode(struct inode *inode, const char *function,
 562                         unsigned int line, ext4_fsblk_t block, int error,
 563                         const char *fmt, ...)
 564 {
 565         va_list args;
 566         struct va_format vaf;
 567
 568         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 569                 return;
 570
 571         trace_ext4_error(inode->i_sb, function, line);
 572         if (ext4_error_ratelimit(inode->i_sb)) {
 573                 va_start(args, fmt);
 574                 vaf.fmt = fmt;
 575                 vaf.va = &args;
 576                 if (block)
 577                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 578                                "inode #%lu: block %llu: comm %s: %pV\n",
 579                                inode->i_sb->s_id, function, line, inode->i_ino,
 580                                block, current->comm, &vaf);
 581                 else
 582                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 583                                "inode #%lu: comm %s: %pV\n",
 584                                inode->i_sb->s_id, function, line, inode->i_ino,
 585                                current->comm, &vaf);
 586                 va_end(args);
 587         }
 588         save_error_info(inode->i_sb, error, inode->i_ino, block,
 589                         function, line);
 590         ext4_handle_error(inode->i_sb);
 591 }
 592
 593 void __ext4_error_file(struct file *file, const char *function,
 594                        unsigned int line, ext4_fsblk_t block,
 595                        const char *fmt, ...)
 596 {
 597         va_list args;
 598         struct va_format vaf;
 599         struct inode *inode = file_inode(file);
 600         char pathname[80], *path;
 601
 602         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 603                 return;
 604
 605         trace_ext4_error(inode->i_sb, function, line);
 606         if (ext4_error_ratelimit(inode->i_sb)) {
 607                 path = file_path(file, pathname, sizeof(pathname));
 608                 if (IS_ERR(path))
 609                         path = "(unknown)";
 610                 va_start(args, fmt);
 611                 vaf.fmt = fmt;
 612                 vaf.va = &args;
 613                 if (block)
 614                         printk(KERN_CRIT
 615                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 616                                "block %llu: comm %s: path %s: %pV\n",
 617                                inode->i_sb->s_id, function, line, inode->i_ino,
 618                                block, current->comm, path, &vaf);
 619                 else
 620                         printk(KERN_CRIT
 621                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 622                                "comm %s: path %s: %pV\n",
 623                                inode->i_sb->s_id, function, line, inode->i_ino,
 624                                current->comm, path, &vaf);
 625                 va_end(args);
 626         }
 627         save_error_info(inode->i_sb, EFSCORRUPTED, inode->i_ino, block,
 628                         function, line);
 629         ext4_handle_error(inode->i_sb);
 630 }
 631
 632 const char *ext4_decode_error(struct super_block *sb, int errno,
 633                               char nbuf[16])
 634 {
 635         char *errstr = NULL;
 636
 637         switch (errno) {
 638         case -EFSCORRUPTED:
 639                 errstr = "Corrupt filesystem";
 640                 break;
 641         case -EFSBADCRC:
 642                 errstr = "Filesystem failed CRC";
 643                 break;
 644         case -EIO:
 645                 errstr = "IO failure";
 646                 break;
 647         case -ENOMEM:
 648                 errstr = "Out of memory";
 649                 break;
 650         case -EROFS:
 651                 if (!sb || (EXT4_SB(sb)->s_journal &&
 652                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 653                         errstr = "Journal has aborted";
 654                 else
 655                         errstr = "Readonly filesystem";
 656                 break;
 657         default:
 658                 /* If the caller passed in an extra buffer for unknown
 659                  * errors, textualise them now.  Else we just return
 660                  * NULL. */
 661                 if (nbuf) {
 662                         /* Check for truncated error codes... */
 663                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 664                                 errstr = nbuf;
 665                 }
 666                 break;
 667         }
 668
 669         return errstr;
 670 }
 671
 672 /* __ext4_std_error decodes expected errors from journaling functions
 673  * automatically and invokes the appropriate error response.  */
 674
 675 void __ext4_std_error(struct super_block *sb, const char *function,
 676                       unsigned int line, int errno)
 677 {
 678         char nbuf[16];
 679         const char *errstr;
 680
 681         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 682                 return;
 683
 684         /* Special case: if the error is EROFS, and we're not already
 685          * inside a transaction, then there's really no point in logging
 686          * an error. */
 687         if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb))
 688                 return;
 689
 690         if (ext4_error_ratelimit(sb)) {
 691                 errstr = ext4_decode_error(sb, errno, nbuf);
 692                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 693                        sb->s_id, function, line, errstr);
 694         }
 695
 696         save_error_info(sb, -errno, 0, 0, function, line);
 697         ext4_handle_error(sb);
 698 }
 699
 700 /*
 701  * ext4_abort is a much stronger failure handler than ext4_error.  The
 702  * abort function may be used to deal with unrecoverable failures such
 703  * as journal IO errors or ENOMEM at a critical moment in log management.
 704  *
 705  * We unconditionally force the filesystem into an ABORT|READONLY state,
 706  * unless the error response on the fs has been set to panic in which
 707  * case we take the easy way out and panic immediately.
 708  */
 709
 710 void __ext4_abort(struct super_block *sb, const char *function,
 711                   unsigned int line, int error, const char *fmt, ...)
 712 {
 713         struct va_format vaf;
 714         va_list args;
 715
 716         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 717                 return;
 718
 719         save_error_info(sb, error, 0, 0, function, line);
 720         va_start(args, fmt);
 721         vaf.fmt = fmt;
 722         vaf.va = &args;
 723         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: %pV\n",
 724                sb->s_id, function, line, &vaf);
 725         va_end(args);
 726
 727         if (sb_rdonly(sb) == 0) {
 728                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 729                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 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                 if (EXT4_SB(sb)->s_journal)
 737                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 738         }
 739         if (test_opt(sb, ERRORS_PANIC) && !system_going_down()) {
 740                 if (EXT4_SB(sb)->s_journal &&
 741                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 742                         return;
 743                 panic("EXT4-fs panic from previous error\n");
 744         }
 745 }
 746
 747 void __ext4_msg(struct super_block *sb,
 748                 const char *prefix, const char *fmt, ...)
 749 {
 750         struct va_format vaf;
 751         va_list args;
 752
 753         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 754                 return;
 755
 756         va_start(args, fmt);
 757         vaf.fmt = fmt;
 758         vaf.va = &args;
 759         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 760         va_end(args);
 761 }
 762
 763 #define ext4_warning_ratelimit(sb)                                      \
 764                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
 765                              "EXT4-fs warning")
 766
 767 void __ext4_warning(struct super_block *sb, const char *function,
 768                     unsigned int line, const char *fmt, ...)
 769 {
 770         struct va_format vaf;
 771         va_list args;
 772
 773         if (!ext4_warning_ratelimit(sb))
 774                 return;
 775
 776         va_start(args, fmt);
 777         vaf.fmt = fmt;
 778         vaf.va = &args;
 779         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 780                sb->s_id, function, line, &vaf);
 781         va_end(args);
 782 }
 783
 784 void __ext4_warning_inode(const struct inode *inode, const char *function,
 785                           unsigned int line, const char *fmt, ...)
 786 {
 787         struct va_format vaf;
 788         va_list args;
 789
 790         if (!ext4_warning_ratelimit(inode->i_sb))
 791                 return;
 792
 793         va_start(args, fmt);
 794         vaf.fmt = fmt;
 795         vaf.va = &args;
 796         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 797                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 798                function, line, inode->i_ino, current->comm, &vaf);
 799         va_end(args);
 800 }
 801
 802 void __ext4_grp_locked_error(const char *function, unsigned int line,
 803                              struct super_block *sb, ext4_group_t grp,
 804                              unsigned long ino, ext4_fsblk_t block,
 805                              const char *fmt, ...)
 806 __releases(bitlock)
 807 __acquires(bitlock)
 808 {
 809         struct va_format vaf;
 810         va_list args;
 811
 812         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 813                 return;
 814
 815         trace_ext4_error(sb, function, line);
 816         __save_error_info(sb, EFSCORRUPTED, ino, block, function, line);
 817
 818         if (ext4_error_ratelimit(sb)) {
 819                 va_start(args, fmt);
 820                 vaf.fmt = fmt;
 821                 vaf.va = &args;
 822                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 823                        sb->s_id, function, line, grp);
 824                 if (ino)
 825                         printk(KERN_CONT "inode %lu: ", ino);
 826                 if (block)
 827                         printk(KERN_CONT "block %llu:",
 828                                (unsigned long long) block);
 829                 printk(KERN_CONT "%pV\n", &vaf);
 830                 va_end(args);
 831         }
 832
 833         if (test_opt(sb, WARN_ON_ERROR))
 834                 WARN_ON_ONCE(1);
 835
 836         if (test_opt(sb, ERRORS_CONT)) {
 837                 ext4_commit_super(sb, 0);
 838                 return;
 839         }
 840
 841         ext4_unlock_group(sb, grp);
 842         ext4_commit_super(sb, 1);
 843         ext4_handle_error(sb);
 844         /*
 845          * We only get here in the ERRORS_RO case; relocking the group
 846          * may be dangerous, but nothing bad will happen since the
 847          * filesystem will have already been marked read/only and the
 848          * journal has been aborted.  We return 1 as a hint to callers
 849          * who might what to use the return value from
 850          * ext4_grp_locked_error() to distinguish between the
 851          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 852          * aggressively from the ext4 function in question, with a
 853          * more appropriate error code.
 854          */
 855         ext4_lock_group(sb, grp);
 856         return;
 857 }
 858
 859 void ext4_mark_group_bitmap_corrupted(struct super_block *sb,
 860                                      ext4_group_t group,
 861                                      unsigned int flags)
 862 {
 863         struct ext4_sb_info *sbi = EXT4_SB(sb);
 864         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
 865         struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
 866         int ret;
 867
 868         if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) {
 869                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT,
 870                                             &grp->bb_state);
 871                 if (!ret)
 872                         percpu_counter_sub(&sbi->s_freeclusters_counter,
 873                                            grp->bb_free);
 874         }
 875
 876         if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) {
 877                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT,
 878                                             &grp->bb_state);
 879                 if (!ret && gdp) {
 880                         int count;
 881
 882                         count = ext4_free_inodes_count(sb, gdp);
 883                         percpu_counter_sub(&sbi->s_freeinodes_counter,
 884                                            count);
 885                 }
 886         }
 887 }
 888
 889 void ext4_update_dynamic_rev(struct super_block *sb)
 890 {
 891         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 892
 893         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 894                 return;
 895
 896         ext4_warning(sb,
 897                      "updating to rev %d because of new feature flag, "
 898                      "running e2fsck is recommended",
 899                      EXT4_DYNAMIC_REV);
 900
 901         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 902         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 903         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 904         /* leave es->s_feature_*compat flags alone */
 905         /* es->s_uuid will be set by e2fsck if empty */
 906
 907         /*
 908          * The rest of the superblock fields should be zero, and if not it
 909          * means they are likely already in use, so leave them alone.  We
 910          * can leave it up to e2fsck to clean up any inconsistencies there.
 911          */
 912 }
 913
 914 /*
 915  * Open the external journal device
 916  */
 917 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 918 {
 919         struct block_device *bdev;
 920
 921         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 922         if (IS_ERR(bdev))
 923                 goto fail;
 924         return bdev;
 925
 926 fail:
 927         ext4_msg(sb, KERN_ERR,
 928                  "failed to open journal device unknown-block(%u,%u) %ld",
 929                  MAJOR(dev), MINOR(dev), PTR_ERR(bdev));
 930         return NULL;
 931 }
 932
 933 /*
 934  * Release the journal device
 935  */
 936 static void ext4_blkdev_put(struct block_device *bdev)
 937 {
 938         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 939 }
 940
 941 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 942 {
 943         struct block_device *bdev;
 944         bdev = sbi->journal_bdev;
 945         if (bdev) {
 946                 ext4_blkdev_put(bdev);
 947                 sbi->journal_bdev = NULL;
 948         }
 949 }
 950
 951 static inline struct inode *orphan_list_entry(struct list_head *l)
 952 {
 953         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 954 }
 955
 956 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 957 {
 958         struct list_head *l;
 959
 960         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 961                  le32_to_cpu(sbi->s_es->s_last_orphan));
 962
 963         printk(KERN_ERR "sb_info orphan list:\n");
 964         list_for_each(l, &sbi->s_orphan) {
 965                 struct inode *inode = orphan_list_entry(l);
 966                 printk(KERN_ERR "  "
 967                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 968                        inode->i_sb->s_id, inode->i_ino, inode,
 969                        inode->i_mode, inode->i_nlink,
 970                        NEXT_ORPHAN(inode));
 971         }
 972 }
 973
 974 #ifdef CONFIG_QUOTA
 975 static int ext4_quota_off(struct super_block *sb, int type);
 976
 977 static inline void ext4_quota_off_umount(struct super_block *sb)
 978 {
 979         int type;
 980
 981         /* Use our quota_off function to clear inode flags etc. */
 982         for (type = 0; type < EXT4_MAXQUOTAS; type++)
 983                 ext4_quota_off(sb, type);
 984 }
 985
 986 /*
 987  * This is a helper function which is used in the mount/remount
 988  * codepaths (which holds s_umount) to fetch the quota file name.
 989  */
 990 static inline char *get_qf_name(struct super_block *sb,
 991                                 struct ext4_sb_info *sbi,
 992                                 int type)
 993 {
 994         return rcu_dereference_protected(sbi->s_qf_names[type],
 995                                          lockdep_is_held(&sb->s_umount));
 996 }
 997 #else
 998 static inline void ext4_quota_off_umount(struct super_block *sb)
 999 {
1000 }
1001 #endif
1002
1003 static void ext4_put_super(struct super_block *sb)
1004 {
1005         struct ext4_sb_info *sbi = EXT4_SB(sb);
1006         struct ext4_super_block *es = sbi->s_es;
1007         struct buffer_head **group_desc;
1008         struct flex_groups **flex_groups;
1009         int aborted = 0;
1010         int i, err;
1011
1012         ext4_unregister_li_request(sb);
1013         ext4_quota_off_umount(sb);
1014
1015         destroy_workqueue(sbi->rsv_conversion_wq);
1016
1017         /*
1018          * Unregister sysfs before destroying jbd2 journal.
1019          * Since we could still access attr_journal_task attribute via sysfs
1020          * path which could have sbi->s_journal->j_task as NULL
1021          */
1022         ext4_unregister_sysfs(sb);
1023
1024         if (sbi->s_journal) {
1025                 aborted = is_journal_aborted(sbi->s_journal);
1026                 err = jbd2_journal_destroy(sbi->s_journal);
1027                 sbi->s_journal = NULL;
1028                 if ((err < 0) && !aborted) {
1029                         ext4_abort(sb, -err, "Couldn't clean up the journal");
1030                 }
1031         }
1032
1033         ext4_es_unregister_shrinker(sbi);
1034         del_timer_sync(&sbi->s_err_report);
1035         ext4_release_system_zone(sb);
1036         ext4_mb_release(sb);
1037         ext4_ext_release(sb);
1038
1039         if (!sb_rdonly(sb) && !aborted) {
1040                 ext4_clear_feature_journal_needs_recovery(sb);
1041                 es->s_state = cpu_to_le16(sbi->s_mount_state);
1042         }
1043         if (!sb_rdonly(sb))
1044                 ext4_commit_super(sb, 1);
1045
1046         rcu_read_lock();
1047         group_desc = rcu_dereference(sbi->s_group_desc);
1048         for (i = 0; i < sbi->s_gdb_count; i++)
1049                 brelse(group_desc[i]);
1050         kvfree(group_desc);
1051         flex_groups = rcu_dereference(sbi->s_flex_groups);
1052         if (flex_groups) {
1053                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
1054                         kvfree(flex_groups[i]);
1055                 kvfree(flex_groups);
1056         }
1057         rcu_read_unlock();
1058         percpu_counter_destroy(&sbi->s_freeclusters_counter);
1059         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1060         percpu_counter_destroy(&sbi->s_dirs_counter);
1061         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
1062         percpu_free_rwsem(&sbi->s_writepages_rwsem);
1063 #ifdef CONFIG_QUOTA
1064         for (i = 0; i < EXT4_MAXQUOTAS; i++)
1065                 kfree(get_qf_name(sb, sbi, i));
1066 #endif
1067
1068         /* Debugging code just in case the in-memory inode orphan list
1069          * isn't empty.  The on-disk one can be non-empty if we've
1070          * detected an error and taken the fs readonly, but the
1071          * in-memory list had better be clean by this point. */
1072         if (!list_empty(&sbi->s_orphan))
1073                 dump_orphan_list(sb, sbi);
1074         J_ASSERT(list_empty(&sbi->s_orphan));
1075
1076         sync_blockdev(sb->s_bdev);
1077         invalidate_bdev(sb->s_bdev);
1078         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
1079                 /*
1080                  * Invalidate the journal device's buffers.  We don't want them
1081                  * floating about in memory - the physical journal device may
1082                  * hotswapped, and it breaks the `ro-after' testing code.
1083                  */
1084                 sync_blockdev(sbi->journal_bdev);
1085                 invalidate_bdev(sbi->journal_bdev);
1086                 ext4_blkdev_remove(sbi);
1087         }
1088
1089         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
1090         sbi->s_ea_inode_cache = NULL;
1091
1092         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
1093         sbi->s_ea_block_cache = NULL;
1094
1095         if (sbi->s_mmp_tsk)
1096                 kthread_stop(sbi->s_mmp_tsk);
1097         brelse(sbi->s_sbh);
1098         sb->s_fs_info = NULL;
1099         /*
1100          * Now that we are completely done shutting down the
1101          * superblock, we need to actually destroy the kobject.
1102          */
1103         kobject_put(&sbi->s_kobj);
1104         wait_for_completion(&sbi->s_kobj_unregister);
1105         if (sbi->s_chksum_driver)
1106                 crypto_free_shash(sbi->s_chksum_driver);
1107         kfree(sbi->s_blockgroup_lock);
1108         fs_put_dax(sbi->s_daxdev);
1109         fscrypt_free_dummy_context(&sbi->s_dummy_enc_ctx);
1110 #ifdef CONFIG_UNICODE
1111         utf8_unload(sbi->s_encoding);
1112 #endif
1113         kfree(sbi);
1114 }
1115
1116 static struct kmem_cache *ext4_inode_cachep;
1117
1118 /*
1119  * Called inside transaction, so use GFP_NOFS
1120  */
1121 static struct inode *ext4_alloc_inode(struct super_block *sb)
1122 {
1123         struct ext4_inode_info *ei;
1124
1125         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
1126         if (!ei)
1127                 return NULL;
1128
1129         inode_set_iversion(&ei->vfs_inode, 1);
1130         spin_lock_init(&ei->i_raw_lock);
1131         INIT_LIST_HEAD(&ei->i_prealloc_list);
1132         spin_lock_init(&ei->i_prealloc_lock);
1133         ext4_es_init_tree(&ei->i_es_tree);
1134         rwlock_init(&ei->i_es_lock);
1135         INIT_LIST_HEAD(&ei->i_es_list);
1136         ei->i_es_all_nr = 0;
1137         ei->i_es_shk_nr = 0;
1138         ei->i_es_shrink_lblk = 0;
1139         ei->i_reserved_data_blocks = 0;
1140         spin_lock_init(&(ei->i_block_reservation_lock));
1141         ext4_init_pending_tree(&ei->i_pending_tree);
1142 #ifdef CONFIG_QUOTA
1143         ei->i_reserved_quota = 0;
1144         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
1145 #endif
1146         ei->jinode = NULL;
1147         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
1148         spin_lock_init(&ei->i_completed_io_lock);
1149         ei->i_sync_tid = 0;
1150         ei->i_datasync_tid = 0;
1151         atomic_set(&ei->i_unwritten, 0);
1152         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
1153         return &ei->vfs_inode;
1154 }
1155
1156 static int ext4_drop_inode(struct inode *inode)
1157 {
1158         int drop = generic_drop_inode(inode);
1159
1160         if (!drop)
1161                 drop = fscrypt_drop_inode(inode);
1162
1163         trace_ext4_drop_inode(inode, drop);
1164         return drop;
1165 }
1166
1167 static void ext4_free_in_core_inode(struct inode *inode)
1168 {
1169         fscrypt_free_inode(inode);
1170         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1171 }
1172
1173 static void ext4_destroy_inode(struct inode *inode)
1174 {
1175         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1176                 ext4_msg(inode->i_sb, KERN_ERR,
1177                          "Inode %lu (%p): orphan list check failed!",
1178                          inode->i_ino, EXT4_I(inode));
1179                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1180                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1181                                 true);
1182                 dump_stack();
1183         }
1184 }
1185
1186 static void init_once(void *foo)
1187 {
1188         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1189
1190         INIT_LIST_HEAD(&ei->i_orphan);
1191         init_rwsem(&ei->xattr_sem);
1192         init_rwsem(&ei->i_data_sem);
1193         init_rwsem(&ei->i_mmap_sem);
1194         inode_init_once(&ei->vfs_inode);
1195 }
1196
1197 static int __init init_inodecache(void)
1198 {
1199         ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache",
1200                                 sizeof(struct ext4_inode_info), 0,
1201                                 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
1202                                         SLAB_ACCOUNT),
1203                                 offsetof(struct ext4_inode_info, i_data),
1204                                 sizeof_field(struct ext4_inode_info, i_data),
1205                                 init_once);
1206         if (ext4_inode_cachep == NULL)
1207                 return -ENOMEM;
1208         return 0;
1209 }
1210
1211 static void destroy_inodecache(void)
1212 {
1213         /*
1214          * Make sure all delayed rcu free inodes are flushed before we
1215          * destroy cache.
1216          */
1217         rcu_barrier();
1218         kmem_cache_destroy(ext4_inode_cachep);
1219 }
1220
1221 void ext4_clear_inode(struct inode *inode)
1222 {
1223         invalidate_inode_buffers(inode);
1224         clear_inode(inode);
1225         ext4_discard_preallocations(inode);
1226         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1227         dquot_drop(inode);
1228         if (EXT4_I(inode)->jinode) {
1229                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1230                                                EXT4_I(inode)->jinode);
1231                 jbd2_free_inode(EXT4_I(inode)->jinode);
1232                 EXT4_I(inode)->jinode = NULL;
1233         }
1234         fscrypt_put_encryption_info(inode);
1235         fsverity_cleanup_inode(inode);
1236 }
1237
1238 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1239                                         u64 ino, u32 generation)
1240 {
1241         struct inode *inode;
1242
1243         /*
1244          * Currently we don't know the generation for parent directory, so
1245          * a generation of 0 means "accept any"
1246          */
1247         inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE);
1248         if (IS_ERR(inode))
1249                 return ERR_CAST(inode);
1250         if (generation && inode->i_generation != generation) {
1251                 iput(inode);
1252                 return ERR_PTR(-ESTALE);
1253         }
1254
1255         return inode;
1256 }
1257
1258 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1259                                         int fh_len, int fh_type)
1260 {
1261         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1262                                     ext4_nfs_get_inode);
1263 }
1264
1265 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1266                                         int fh_len, int fh_type)
1267 {
1268         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1269                                     ext4_nfs_get_inode);
1270 }
1271
1272 static int ext4_nfs_commit_metadata(struct inode *inode)
1273 {
1274         struct writeback_control wbc = {
1275                 .sync_mode = WB_SYNC_ALL
1276         };
1277
1278         trace_ext4_nfs_commit_metadata(inode);
1279         return ext4_write_inode(inode, &wbc);
1280 }
1281
1282 /*
1283  * Try to release metadata pages (indirect blocks, directories) which are
1284  * mapped via the block device.  Since these pages could have journal heads
1285  * which would prevent try_to_free_buffers() from freeing them, we must use
1286  * jbd2 layer's try_to_free_buffers() function to release them.
1287  */
1288 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1289                                  gfp_t wait)
1290 {
1291         journal_t *journal = EXT4_SB(sb)->s_journal;
1292
1293         WARN_ON(PageChecked(page));
1294         if (!page_has_buffers(page))
1295                 return 0;
1296         if (journal)
1297                 return jbd2_journal_try_to_free_buffers(journal, page,
1298                                                 wait & ~__GFP_DIRECT_RECLAIM);
1299         return try_to_free_buffers(page);
1300 }
1301
1302 #ifdef CONFIG_FS_ENCRYPTION
1303 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1304 {
1305         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1306                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1307 }
1308
1309 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1310                                                         void *fs_data)
1311 {
1312         handle_t *handle = fs_data;
1313         int res, res2, credits, retries = 0;
1314
1315         /*
1316          * Encrypting the root directory is not allowed because e2fsck expects
1317          * lost+found to exist and be unencrypted, and encrypting the root
1318          * directory would imply encrypting the lost+found directory as well as
1319          * the filename "lost+found" itself.
1320          */
1321         if (inode->i_ino == EXT4_ROOT_INO)
1322                 return -EPERM;
1323
1324         if (WARN_ON_ONCE(IS_DAX(inode) && i_size_read(inode)))
1325                 return -EINVAL;
1326
1327         res = ext4_convert_inline_data(inode);
1328         if (res)
1329                 return res;
1330
1331         /*
1332          * If a journal handle was specified, then the encryption context is
1333          * being set on a new inode via inheritance and is part of a larger
1334          * transaction to create the inode.  Otherwise the encryption context is
1335          * being set on an existing inode in its own transaction.  Only in the
1336          * latter case should the "retry on ENOSPC" logic be used.
1337          */
1338
1339         if (handle) {
1340                 res = ext4_xattr_set_handle(handle, inode,
1341                                             EXT4_XATTR_INDEX_ENCRYPTION,
1342                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1343                                             ctx, len, 0);
1344                 if (!res) {
1345                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1346                         ext4_clear_inode_state(inode,
1347                                         EXT4_STATE_MAY_INLINE_DATA);
1348                         /*
1349                          * Update inode->i_flags - S_ENCRYPTED will be enabled,
1350                          * S_DAX may be disabled
1351                          */
1352                         ext4_set_inode_flags(inode);
1353                 }
1354                 return res;
1355         }
1356
1357         res = dquot_initialize(inode);
1358         if (res)
1359                 return res;
1360 retry:
1361         res = ext4_xattr_set_credits(inode, len, false /* is_create */,
1362                                      &credits);
1363         if (res)
1364                 return res;
1365
1366         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
1367         if (IS_ERR(handle))
1368                 return PTR_ERR(handle);
1369
1370         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1371                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1372                                     ctx, len, 0);
1373         if (!res) {
1374                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1375                 /*
1376                  * Update inode->i_flags - S_ENCRYPTED will be enabled,
1377                  * S_DAX may be disabled
1378                  */
1379                 ext4_set_inode_flags(inode);
1380                 res = ext4_mark_inode_dirty(handle, inode);
1381                 if (res)
1382                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1383         }
1384         res2 = ext4_journal_stop(handle);
1385
1386         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1387                 goto retry;
1388         if (!res)
1389                 res = res2;
1390         return res;
1391 }
1392
1393 static const union fscrypt_context *
1394 ext4_get_dummy_context(struct super_block *sb)
1395 {
1396         return EXT4_SB(sb)->s_dummy_enc_ctx.ctx;
1397 }
1398
1399 static bool ext4_has_stable_inodes(struct super_block *sb)
1400 {
1401         return ext4_has_feature_stable_inodes(sb);
1402 }
1403
1404 static void ext4_get_ino_and_lblk_bits(struct super_block *sb,
1405                                        int *ino_bits_ret, int *lblk_bits_ret)
1406 {
1407         *ino_bits_ret = 8 * sizeof(EXT4_SB(sb)->s_es->s_inodes_count);
1408         *lblk_bits_ret = 8 * sizeof(ext4_lblk_t);
1409 }
1410
1411 static const struct fscrypt_operations ext4_cryptops = {
1412         .key_prefix             = "ext4:",
1413         .get_context            = ext4_get_context,
1414         .set_context            = ext4_set_context,
1415         .get_dummy_context      = ext4_get_dummy_context,
1416         .empty_dir              = ext4_empty_dir,
1417         .max_namelen            = EXT4_NAME_LEN,
1418         .has_stable_inodes      = ext4_has_stable_inodes,
1419         .get_ino_and_lblk_bits  = ext4_get_ino_and_lblk_bits,
1420 };
1421 #endif
1422
1423 #ifdef CONFIG_QUOTA
1424 static const char * const quotatypes[] = INITQFNAMES;
1425 #define QTYPE2NAME(t) (quotatypes[t])
1426
1427 static int ext4_write_dquot(struct dquot *dquot);
1428 static int ext4_acquire_dquot(struct dquot *dquot);
1429 static int ext4_release_dquot(struct dquot *dquot);
1430 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1431 static int ext4_write_info(struct super_block *sb, int type);
1432 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1433                          const struct path *path);
1434 static int ext4_quota_on_mount(struct super_block *sb, int type);
1435 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1436                                size_t len, loff_t off);
1437 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1438                                 const char *data, size_t len, loff_t off);
1439 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1440                              unsigned int flags);
1441 static int ext4_enable_quotas(struct super_block *sb);
1442
1443 static struct dquot **ext4_get_dquots(struct inode *inode)
1444 {
1445         return EXT4_I(inode)->i_dquot;
1446 }
1447
1448 static const struct dquot_operations ext4_quota_operations = {
1449         .get_reserved_space     = ext4_get_reserved_space,
1450         .write_dquot            = ext4_write_dquot,
1451         .acquire_dquot          = ext4_acquire_dquot,
1452         .release_dquot          = ext4_release_dquot,
1453         .mark_dirty             = ext4_mark_dquot_dirty,
1454         .write_info             = ext4_write_info,
1455         .alloc_dquot            = dquot_alloc,
1456         .destroy_dquot          = dquot_destroy,
1457         .get_projid             = ext4_get_projid,
1458         .get_inode_usage        = ext4_get_inode_usage,
1459         .get_next_id            = dquot_get_next_id,
1460 };
1461
1462 static const struct quotactl_ops ext4_qctl_operations = {
1463         .quota_on       = ext4_quota_on,
1464         .quota_off      = ext4_quota_off,
1465         .quota_sync     = dquot_quota_sync,
1466         .get_state      = dquot_get_state,
1467         .set_info       = dquot_set_dqinfo,
1468         .get_dqblk      = dquot_get_dqblk,
1469         .set_dqblk      = dquot_set_dqblk,
1470         .get_nextdqblk  = dquot_get_next_dqblk,
1471 };
1472 #endif
1473
1474 static const struct super_operations ext4_sops = {
1475         .alloc_inode    = ext4_alloc_inode,
1476         .free_inode     = ext4_free_in_core_inode,
1477         .destroy_inode  = ext4_destroy_inode,
1478         .write_inode    = ext4_write_inode,
1479         .dirty_inode    = ext4_dirty_inode,
1480         .drop_inode     = ext4_drop_inode,
1481         .evict_inode    = ext4_evict_inode,
1482         .put_super      = ext4_put_super,
1483         .sync_fs        = ext4_sync_fs,
1484         .freeze_fs      = ext4_freeze,
1485         .unfreeze_fs    = ext4_unfreeze,
1486         .statfs         = ext4_statfs,
1487         .remount_fs     = ext4_remount,
1488         .show_options   = ext4_show_options,
1489 #ifdef CONFIG_QUOTA
1490         .quota_read     = ext4_quota_read,
1491         .quota_write    = ext4_quota_write,
1492         .get_dquots     = ext4_get_dquots,
1493 #endif
1494         .bdev_try_to_free_page = bdev_try_to_free_page,
1495 };
1496
1497 static const struct export_operations ext4_export_ops = {
1498         .fh_to_dentry = ext4_fh_to_dentry,
1499         .fh_to_parent = ext4_fh_to_parent,
1500         .get_parent = ext4_get_parent,
1501         .commit_metadata = ext4_nfs_commit_metadata,
1502 };
1503
1504 enum {
1505         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1506         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1507         Opt_nouid32, Opt_debug, Opt_removed,
1508         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1509         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1510         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1511         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1512         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1513         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1514         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1515         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1516         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1517         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, Opt_dax,
1518         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
1519         Opt_nowarn_on_error, Opt_mblk_io_submit,
1520         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1521         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1522         Opt_inode_readahead_blks, Opt_journal_ioprio,
1523         Opt_dioread_nolock, Opt_dioread_lock,
1524         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1525         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1526 };
1527
1528 static const match_table_t tokens = {
1529         {Opt_bsd_df, "bsddf"},
1530         {Opt_minix_df, "minixdf"},
1531         {Opt_grpid, "grpid"},
1532         {Opt_grpid, "bsdgroups"},
1533         {Opt_nogrpid, "nogrpid"},
1534         {Opt_nogrpid, "sysvgroups"},
1535         {Opt_resgid, "resgid=%u"},
1536         {Opt_resuid, "resuid=%u"},
1537         {Opt_sb, "sb=%u"},
1538         {Opt_err_cont, "errors=continue"},
1539         {Opt_err_panic, "errors=panic"},
1540         {Opt_err_ro, "errors=remount-ro"},
1541         {Opt_nouid32, "nouid32"},
1542         {Opt_debug, "debug"},
1543         {Opt_removed, "oldalloc"},
1544         {Opt_removed, "orlov"},
1545         {Opt_user_xattr, "user_xattr"},
1546         {Opt_nouser_xattr, "nouser_xattr"},
1547         {Opt_acl, "acl"},
1548         {Opt_noacl, "noacl"},
1549         {Opt_noload, "norecovery"},
1550         {Opt_noload, "noload"},
1551         {Opt_removed, "nobh"},
1552         {Opt_removed, "bh"},
1553         {Opt_commit, "commit=%u"},
1554         {Opt_min_batch_time, "min_batch_time=%u"},
1555         {Opt_max_batch_time, "max_batch_time=%u"},
1556         {Opt_journal_dev, "journal_dev=%u"},
1557         {Opt_journal_path, "journal_path=%s"},
1558         {Opt_journal_checksum, "journal_checksum"},
1559         {Opt_nojournal_checksum, "nojournal_checksum"},
1560         {Opt_journal_async_commit, "journal_async_commit"},
1561         {Opt_abort, "abort"},
1562         {Opt_data_journal, "data=journal"},
1563         {Opt_data_ordered, "data=ordered"},
1564         {Opt_data_writeback, "data=writeback"},
1565         {Opt_data_err_abort, "data_err=abort"},
1566         {Opt_data_err_ignore, "data_err=ignore"},
1567         {Opt_offusrjquota, "usrjquota="},
1568         {Opt_usrjquota, "usrjquota=%s"},
1569         {Opt_offgrpjquota, "grpjquota="},
1570         {Opt_grpjquota, "grpjquota=%s"},
1571         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1572         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1573         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1574         {Opt_grpquota, "grpquota"},
1575         {Opt_noquota, "noquota"},
1576         {Opt_quota, "quota"},
1577         {Opt_usrquota, "usrquota"},
1578         {Opt_prjquota, "prjquota"},
1579         {Opt_barrier, "barrier=%u"},
1580         {Opt_barrier, "barrier"},
1581         {Opt_nobarrier, "nobarrier"},
1582         {Opt_i_version, "i_version"},
1583         {Opt_dax, "dax"},
1584         {Opt_stripe, "stripe=%u"},
1585         {Opt_delalloc, "delalloc"},
1586         {Opt_warn_on_error, "warn_on_error"},
1587         {Opt_nowarn_on_error, "nowarn_on_error"},
1588         {Opt_lazytime, "lazytime"},
1589         {Opt_nolazytime, "nolazytime"},
1590         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1591         {Opt_nodelalloc, "nodelalloc"},
1592         {Opt_removed, "mblk_io_submit"},
1593         {Opt_removed, "nomblk_io_submit"},
1594         {Opt_block_validity, "block_validity"},
1595         {Opt_noblock_validity, "noblock_validity"},
1596         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1597         {Opt_journal_ioprio, "journal_ioprio=%u"},
1598         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1599         {Opt_auto_da_alloc, "auto_da_alloc"},
1600         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1601         {Opt_dioread_nolock, "dioread_nolock"},
1602         {Opt_dioread_lock, "nodioread_nolock"},
1603         {Opt_dioread_lock, "dioread_lock"},
1604         {Opt_discard, "discard"},
1605         {Opt_nodiscard, "nodiscard"},
1606         {Opt_init_itable, "init_itable=%u"},
1607         {Opt_init_itable, "init_itable"},
1608         {Opt_noinit_itable, "noinit_itable"},
1609         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1610         {Opt_test_dummy_encryption, "test_dummy_encryption=%s"},
1611         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1612         {Opt_nombcache, "nombcache"},
1613         {Opt_nombcache, "no_mbcache"},  /* for backward compatibility */
1614         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1615         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1616         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1617         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1618         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1619         {Opt_err, NULL},
1620 };
1621
1622 static ext4_fsblk_t get_sb_block(void **data)
1623 {
1624         ext4_fsblk_t    sb_block;
1625         char            *options = (char *) *data;
1626
1627         if (!options || strncmp(options, "sb=", 3) != 0)
1628                 return 1;       /* Default location */
1629
1630         options += 3;
1631         /* TODO: use simple_strtoll with >32bit ext4 */
1632         sb_block = simple_strtoul(options, &options, 0);
1633         if (*options && *options != ',') {
1634                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1635                        (char *) *data);
1636                 return 1;
1637         }
1638         if (*options == ',')
1639                 options++;
1640         *data = (void *) options;
1641
1642         return sb_block;
1643 }
1644
1645 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1646 static const char deprecated_msg[] =
1647         "Mount option \"%s\" will be removed by %s\n"
1648         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1649
1650 #ifdef CONFIG_QUOTA
1651 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1652 {
1653         struct ext4_sb_info *sbi = EXT4_SB(sb);
1654         char *qname, *old_qname = get_qf_name(sb, sbi, qtype);
1655         int ret = -1;
1656
1657         if (sb_any_quota_loaded(sb) && !old_qname) {
1658                 ext4_msg(sb, KERN_ERR,
1659                         "Cannot change journaled "
1660                         "quota options when quota turned on");
1661                 return -1;
1662         }
1663         if (ext4_has_feature_quota(sb)) {
1664                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1665                          "ignored when QUOTA feature is enabled");
1666                 return 1;
1667         }
1668         qname = match_strdup(args);
1669         if (!qname) {
1670                 ext4_msg(sb, KERN_ERR,
1671                         "Not enough memory for storing quotafile name");
1672                 return -1;
1673         }
1674         if (old_qname) {
1675                 if (strcmp(old_qname, qname) == 0)
1676                         ret = 1;
1677                 else
1678                         ext4_msg(sb, KERN_ERR,
1679                                  "%s quota file already specified",
1680                                  QTYPE2NAME(qtype));
1681                 goto errout;
1682         }
1683         if (strchr(qname, '/')) {
1684                 ext4_msg(sb, KERN_ERR,
1685                         "quotafile must be on filesystem root");
1686                 goto errout;
1687         }
1688         rcu_assign_pointer(sbi->s_qf_names[qtype], qname);
1689         set_opt(sb, QUOTA);
1690         return 1;
1691 errout:
1692         kfree(qname);
1693         return ret;
1694 }
1695
1696 static int clear_qf_name(struct super_block *sb, int qtype)
1697 {
1698
1699         struct ext4_sb_info *sbi = EXT4_SB(sb);
1700         char *old_qname = get_qf_name(sb, sbi, qtype);
1701
1702         if (sb_any_quota_loaded(sb) && old_qname) {
1703                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1704                         " when quota turned on");
1705                 return -1;
1706         }
1707         rcu_assign_pointer(sbi->s_qf_names[qtype], NULL);
1708         synchronize_rcu();
1709         kfree(old_qname);
1710         return 1;
1711 }
1712 #endif
1713
1714 #define MOPT_SET        0x0001
1715 #define MOPT_CLEAR      0x0002
1716 #define MOPT_NOSUPPORT  0x0004
1717 #define MOPT_EXPLICIT   0x0008
1718 #define MOPT_CLEAR_ERR  0x0010
1719 #define MOPT_GTE0       0x0020
1720 #ifdef CONFIG_QUOTA
1721 #define MOPT_Q          0
1722 #define MOPT_QFMT       0x0040
1723 #else
1724 #define MOPT_Q          MOPT_NOSUPPORT
1725 #define MOPT_QFMT       MOPT_NOSUPPORT
1726 #endif
1727 #define MOPT_DATAJ      0x0080
1728 #define MOPT_NO_EXT2    0x0100
1729 #define MOPT_NO_EXT3    0x0200
1730 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1731 #define MOPT_STRING     0x0400
1732
1733 static const struct mount_opts {
1734         int     token;
1735         int     mount_opt;
1736         int     flags;
1737 } ext4_mount_opts[] = {
1738         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1739         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1740         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1741         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1742         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1743         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1744         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1745          MOPT_EXT4_ONLY | MOPT_SET},
1746         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1747          MOPT_EXT4_ONLY | MOPT_CLEAR},
1748         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1749         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1750         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1751          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1752         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1753          MOPT_EXT4_ONLY | MOPT_CLEAR},
1754         {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET},
1755         {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR},
1756         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1757          MOPT_EXT4_ONLY | MOPT_CLEAR},
1758         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1759          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1760         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1761                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1762          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1763         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1764         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1765         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1766         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1767         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1768          MOPT_NO_EXT2},
1769         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1770          MOPT_NO_EXT2},
1771         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1772         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1773         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1774         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1775         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1776         {Opt_commit, 0, MOPT_GTE0},
1777         {Opt_max_batch_time, 0, MOPT_GTE0},
1778         {Opt_min_batch_time, 0, MOPT_GTE0},
1779         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1780         {Opt_init_itable, 0, MOPT_GTE0},
1781         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1782         {Opt_stripe, 0, MOPT_GTE0},
1783         {Opt_resuid, 0, MOPT_GTE0},
1784         {Opt_resgid, 0, MOPT_GTE0},
1785         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1786         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1787         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1788         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1789         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1790         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1791          MOPT_NO_EXT2 | MOPT_DATAJ},
1792         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1793         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1794 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1795         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1796         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1797 #else
1798         {Opt_acl, 0, MOPT_NOSUPPORT},
1799         {Opt_noacl, 0, MOPT_NOSUPPORT},
1800 #endif
1801         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1802         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1803         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1804         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1805         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1806                                                         MOPT_SET | MOPT_Q},
1807         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1808                                                         MOPT_SET | MOPT_Q},
1809         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1810                                                         MOPT_SET | MOPT_Q},
1811         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1812                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1813                                                         MOPT_CLEAR | MOPT_Q},
1814         {Opt_usrjquota, 0, MOPT_Q},
1815         {Opt_grpjquota, 0, MOPT_Q},
1816         {Opt_offusrjquota, 0, MOPT_Q},
1817         {Opt_offgrpjquota, 0, MOPT_Q},
1818         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1819         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1820         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1821         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1822         {Opt_test_dummy_encryption, 0, MOPT_STRING},
1823         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1824         {Opt_err, 0, 0}
1825 };
1826
1827 #ifdef CONFIG_UNICODE
1828 static const struct ext4_sb_encodings {
1829         __u16 magic;
1830         char *name;
1831         char *version;
1832 } ext4_sb_encoding_map[] = {
1833         {EXT4_ENC_UTF8_12_1, "utf8", "12.1.0"},
1834 };
1835
1836 static int ext4_sb_read_encoding(const struct ext4_super_block *es,
1837                                  const struct ext4_sb_encodings **encoding,
1838                                  __u16 *flags)
1839 {
1840         __u16 magic = le16_to_cpu(es->s_encoding);
1841         int i;
1842
1843         for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++)
1844                 if (magic == ext4_sb_encoding_map[i].magic)
1845                         break;
1846
1847         if (i >= ARRAY_SIZE(ext4_sb_encoding_map))
1848                 return -EINVAL;
1849
1850         *encoding = &ext4_sb_encoding_map[i];
1851         *flags = le16_to_cpu(es->s_encoding_flags);
1852
1853         return 0;
1854 }
1855 #endif
1856
1857 static int ext4_set_test_dummy_encryption(struct super_block *sb,
1858                                           const char *opt,
1859                                           const substring_t *arg,
1860                                           bool is_remount)
1861 {
1862 #ifdef CONFIG_FS_ENCRYPTION
1863         struct ext4_sb_info *sbi = EXT4_SB(sb);
1864         int err;
1865
1866         /*
1867          * This mount option is just for testing, and it's not worthwhile to
1868          * implement the extra complexity (e.g. RCU protection) that would be
1869          * needed to allow it to be set or changed during remount.  We do allow
1870          * it to be specified during remount, but only if there is no change.
1871          */
1872         if (is_remount && !sbi->s_dummy_enc_ctx.ctx) {
1873                 ext4_msg(sb, KERN_WARNING,
1874                          "Can't set test_dummy_encryption on remount");
1875                 return -1;
1876         }
1877         err = fscrypt_set_test_dummy_encryption(sb, arg, &sbi->s_dummy_enc_ctx);
1878         if (err) {
1879                 if (err == -EEXIST)
1880                         ext4_msg(sb, KERN_WARNING,
1881                                  "Can't change test_dummy_encryption on remount");
1882                 else if (err == -EINVAL)
1883                         ext4_msg(sb, KERN_WARNING,
1884                                  "Value of option \"%s\" is unrecognized", opt);
1885                 else
1886                         ext4_msg(sb, KERN_WARNING,
1887                                  "Error processing option \"%s\" [%d]",
1888                                  opt, err);
1889                 return -1;
1890         }
1891         ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled");
1892 #else
1893         ext4_msg(sb, KERN_WARNING,
1894                  "Test dummy encryption mount option ignored");
1895 #endif
1896         return 1;
1897 }
1898
1899 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1900                             substring_t *args, unsigned long *journal_devnum,
1901                             unsigned int *journal_ioprio, int is_remount)
1902 {
1903         struct ext4_sb_info *sbi = EXT4_SB(sb);
1904         const struct mount_opts *m;
1905         kuid_t uid;
1906         kgid_t gid;
1907         int arg = 0;
1908
1909 #ifdef CONFIG_QUOTA
1910         if (token == Opt_usrjquota)
1911                 return set_qf_name(sb, USRQUOTA, &args[0]);
1912         else if (token == Opt_grpjquota)
1913                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1914         else if (token == Opt_offusrjquota)
1915                 return clear_qf_name(sb, USRQUOTA);
1916         else if (token == Opt_offgrpjquota)
1917                 return clear_qf_name(sb, GRPQUOTA);
1918 #endif
1919         switch (token) {
1920         case Opt_noacl:
1921         case Opt_nouser_xattr:
1922                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1923                 break;
1924         case Opt_sb:
1925                 return 1;       /* handled by get_sb_block() */
1926         case Opt_removed:
1927                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1928                 return 1;
1929         case Opt_abort:
1930                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1931                 return 1;
1932         case Opt_i_version:
1933                 sb->s_flags |= SB_I_VERSION;
1934                 return 1;
1935         case Opt_lazytime:
1936                 sb->s_flags |= SB_LAZYTIME;
1937                 return 1;
1938         case Opt_nolazytime:
1939                 sb->s_flags &= ~SB_LAZYTIME;
1940                 return 1;
1941         }
1942
1943         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1944                 if (token == m->token)
1945                         break;
1946
1947         if (m->token == Opt_err) {
1948                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1949                          "or missing value", opt);
1950                 return -1;
1951         }
1952
1953         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1954                 ext4_msg(sb, KERN_ERR,
1955                          "Mount option \"%s\" incompatible with ext2", opt);
1956                 return -1;
1957         }
1958         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1959                 ext4_msg(sb, KERN_ERR,
1960                          "Mount option \"%s\" incompatible with ext3", opt);
1961                 return -1;
1962         }
1963
1964         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1965                 return -1;
1966         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1967                 return -1;
1968         if (m->flags & MOPT_EXPLICIT) {
1969                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1970                         set_opt2(sb, EXPLICIT_DELALLOC);
1971                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1972                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1973                 } else
1974                         return -1;
1975         }
1976         if (m->flags & MOPT_CLEAR_ERR)
1977                 clear_opt(sb, ERRORS_MASK);
1978         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1979                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1980                          "options when quota turned on");
1981                 return -1;
1982         }
1983
1984         if (m->flags & MOPT_NOSUPPORT) {
1985                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1986         } else if (token == Opt_commit) {
1987                 if (arg == 0)
1988                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1989                 else if (arg > INT_MAX / HZ) {
1990                         ext4_msg(sb, KERN_ERR,
1991                                  "Invalid commit interval %d, "
1992                                  "must be smaller than %d",
1993                                  arg, INT_MAX / HZ);
1994                         return -1;
1995                 }
1996                 sbi->s_commit_interval = HZ * arg;
1997         } else if (token == Opt_debug_want_extra_isize) {
1998                 if ((arg & 1) ||
1999                     (arg < 4) ||
2000                     (arg > (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE))) {
2001                         ext4_msg(sb, KERN_ERR,
2002                                  "Invalid want_extra_isize %d", arg);
2003                         return -1;
2004                 }
2005                 sbi->s_want_extra_isize = arg;
2006         } else if (token == Opt_max_batch_time) {
2007                 sbi->s_max_batch_time = arg;
2008         } else if (token == Opt_min_batch_time) {
2009                 sbi->s_min_batch_time = arg;
2010         } else if (token == Opt_inode_readahead_blks) {
2011                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
2012                         ext4_msg(sb, KERN_ERR,
2013                                  "EXT4-fs: inode_readahead_blks must be "
2014                                  "0 or a power of 2 smaller than 2^31");
2015                         return -1;
2016                 }
2017                 sbi->s_inode_readahead_blks = arg;
2018         } else if (token == Opt_init_itable) {
2019                 set_opt(sb, INIT_INODE_TABLE);
2020                 if (!args->from)
2021                         arg = EXT4_DEF_LI_WAIT_MULT;
2022                 sbi->s_li_wait_mult = arg;
2023         } else if (token == Opt_max_dir_size_kb) {
2024                 sbi->s_max_dir_size_kb = arg;
2025         } else if (token == Opt_stripe) {
2026                 sbi->s_stripe = arg;
2027         } else if (token == Opt_resuid) {
2028                 uid = make_kuid(current_user_ns(), arg);
2029                 if (!uid_valid(uid)) {
2030                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
2031                         return -1;
2032                 }
2033                 sbi->s_resuid = uid;
2034         } else if (token == Opt_resgid) {
2035                 gid = make_kgid(current_user_ns(), arg);
2036                 if (!gid_valid(gid)) {
2037                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
2038                         return -1;
2039                 }
2040                 sbi->s_resgid = gid;
2041         } else if (token == Opt_journal_dev) {
2042                 if (is_remount) {
2043                         ext4_msg(sb, KERN_ERR,
2044                                  "Cannot specify journal on remount");
2045                         return -1;
2046                 }
2047                 *journal_devnum = arg;
2048         } else if (token == Opt_journal_path) {
2049                 char *journal_path;
2050                 struct inode *journal_inode;
2051                 struct path path;
2052                 int error;
2053
2054                 if (is_remount) {
2055                         ext4_msg(sb, KERN_ERR,
2056                                  "Cannot specify journal on remount");
2057                         return -1;
2058                 }
2059                 journal_path = match_strdup(&args[0]);
2060                 if (!journal_path) {
2061                         ext4_msg(sb, KERN_ERR, "error: could not dup "
2062                                 "journal device string");
2063                         return -1;
2064                 }
2065
2066                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
2067                 if (error) {
2068                         ext4_msg(sb, KERN_ERR, "error: could not find "
2069                                 "journal device path: error %d", error);
2070                         kfree(journal_path);
2071                         return -1;
2072                 }
2073
2074                 journal_inode = d_inode(path.dentry);
2075                 if (!S_ISBLK(journal_inode->i_mode)) {
2076                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
2077                                 "is not a block device", journal_path);
2078                         path_put(&path);
2079                         kfree(journal_path);
2080                         return -1;
2081                 }
2082
2083                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
2084                 path_put(&path);
2085                 kfree(journal_path);
2086         } else if (token == Opt_journal_ioprio) {
2087                 if (arg > 7) {
2088                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
2089                                  " (must be 0-7)");
2090                         return -1;
2091                 }
2092                 *journal_ioprio =
2093                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
2094         } else if (token == Opt_test_dummy_encryption) {
2095                 return ext4_set_test_dummy_encryption(sb, opt, &args[0],
2096                                                       is_remount);
2097         } else if (m->flags & MOPT_DATAJ) {
2098                 if (is_remount) {
2099                         if (!sbi->s_journal)
2100                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
2101                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
2102                                 ext4_msg(sb, KERN_ERR,
2103                                          "Cannot change data mode on remount");
2104                                 return -1;
2105                         }
2106                 } else {
2107                         clear_opt(sb, DATA_FLAGS);
2108                         sbi->s_mount_opt |= m->mount_opt;
2109                 }
2110 #ifdef CONFIG_QUOTA
2111         } else if (m->flags & MOPT_QFMT) {
2112                 if (sb_any_quota_loaded(sb) &&
2113                     sbi->s_jquota_fmt != m->mount_opt) {
2114                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
2115                                  "quota options when quota turned on");
2116                         return -1;
2117                 }
2118                 if (ext4_has_feature_quota(sb)) {
2119                         ext4_msg(sb, KERN_INFO,
2120                                  "Quota format mount options ignored "
2121                                  "when QUOTA feature is enabled");
2122                         return 1;
2123                 }
2124                 sbi->s_jquota_fmt = m->mount_opt;
2125 #endif
2126         } else if (token == Opt_dax) {
2127 #ifdef CONFIG_FS_DAX
2128                 ext4_msg(sb, KERN_WARNING,
2129                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
2130                 sbi->s_mount_opt |= m->mount_opt;
2131 #else
2132                 ext4_msg(sb, KERN_INFO, "dax option not supported");
2133                 return -1;
2134 #endif
2135         } else if (token == Opt_data_err_abort) {
2136                 sbi->s_mount_opt |= m->mount_opt;
2137         } else if (token == Opt_data_err_ignore) {
2138                 sbi->s_mount_opt &= ~m->mount_opt;
2139         } else {
2140                 if (!args->from)
2141                         arg = 1;
2142                 if (m->flags & MOPT_CLEAR)
2143                         arg = !arg;
2144                 else if (unlikely(!(m->flags & MOPT_SET))) {
2145                         ext4_msg(sb, KERN_WARNING,
2146                                  "buggy handling of option %s", opt);
2147                         WARN_ON(1);
2148                         return -1;
2149                 }
2150                 if (arg != 0)
2151                         sbi->s_mount_opt |= m->mount_opt;
2152                 else
2153                         sbi->s_mount_opt &= ~m->mount_opt;
2154         }
2155         return 1;
2156 }
2157
2158 static int parse_options(char *options, struct super_block *sb,
2159                          unsigned long *journal_devnum,
2160                          unsigned int *journal_ioprio,
2161                          int is_remount)
2162 {
2163         struct ext4_sb_info __maybe_unused *sbi = EXT4_SB(sb);
2164         char *p, __maybe_unused *usr_qf_name, __maybe_unused *grp_qf_name;
2165         substring_t args[MAX_OPT_ARGS];
2166         int token;
2167
2168         if (!options)
2169                 return 1;
2170
2171         while ((p = strsep(&options, ",")) != NULL) {
2172                 if (!*p)
2173                         continue;
2174                 /*
2175                  * Initialize args struct so we know whether arg was
2176                  * found; some options take optional arguments.
2177                  */
2178                 args[0].to = args[0].from = NULL;
2179                 token = match_token(p, tokens, args);
2180                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
2181                                      journal_ioprio, is_remount) < 0)
2182                         return 0;
2183         }
2184 #ifdef CONFIG_QUOTA
2185         /*
2186          * We do the test below only for project quotas. 'usrquota' and
2187          * 'grpquota' mount options are allowed even without quota feature
2188          * to support legacy quotas in quota files.
2189          */
2190         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
2191                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
2192                          "Cannot enable project quota enforcement.");
2193                 return 0;
2194         }
2195         usr_qf_name = get_qf_name(sb, sbi, USRQUOTA);
2196         grp_qf_name = get_qf_name(sb, sbi, GRPQUOTA);
2197         if (usr_qf_name || grp_qf_name) {
2198                 if (test_opt(sb, USRQUOTA) && usr_qf_name)
2199                         clear_opt(sb, USRQUOTA);
2200
2201                 if (test_opt(sb, GRPQUOTA) && grp_qf_name)
2202                         clear_opt(sb, GRPQUOTA);
2203
2204                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
2205                         ext4_msg(sb, KERN_ERR, "old and new quota "
2206                                         "format mixing");
2207                         return 0;
2208                 }
2209
2210                 if (!sbi->s_jquota_fmt) {
2211                         ext4_msg(sb, KERN_ERR, "journaled quota format "
2212                                         "not specified");
2213                         return 0;
2214                 }
2215         }
2216 #endif
2217         if (test_opt(sb, DIOREAD_NOLOCK)) {
2218                 int blocksize =
2219                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
2220                 if (blocksize < PAGE_SIZE)
2221                         ext4_msg(sb, KERN_WARNING, "Warning: mounting with an "
2222                                  "experimental mount option 'dioread_nolock' "
2223                                  "for blocksize < PAGE_SIZE");
2224         }
2225         return 1;
2226 }
2227
2228 static inline void ext4_show_quota_options(struct seq_file *seq,
2229                                            struct super_block *sb)
2230 {
2231 #if defined(CONFIG_QUOTA)
2232         struct ext4_sb_info *sbi = EXT4_SB(sb);
2233         char *usr_qf_name, *grp_qf_name;
2234
2235         if (sbi->s_jquota_fmt) {
2236                 char *fmtname = "";
2237
2238                 switch (sbi->s_jquota_fmt) {
2239                 case QFMT_VFS_OLD:
2240                         fmtname = "vfsold";
2241                         break;
2242                 case QFMT_VFS_V0:
2243                         fmtname = "vfsv0";
2244                         break;
2245                 case QFMT_VFS_V1:
2246                         fmtname = "vfsv1";
2247                         break;
2248                 }
2249                 seq_printf(seq, ",jqfmt=%s", fmtname);
2250         }
2251
2252         rcu_read_lock();
2253         usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]);
2254         grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]);
2255         if (usr_qf_name)
2256                 seq_show_option(seq, "usrjquota", usr_qf_name);
2257         if (grp_qf_name)
2258                 seq_show_option(seq, "grpjquota", grp_qf_name);
2259         rcu_read_unlock();
2260 #endif
2261 }
2262
2263 static const char *token2str(int token)
2264 {
2265         const struct match_token *t;
2266
2267         for (t = tokens; t->token != Opt_err; t++)
2268                 if (t->token == token && !strchr(t->pattern, '='))
2269                         break;
2270         return t->pattern;
2271 }
2272
2273 /*
2274  * Show an option if
2275  *  - it's set to a non-default value OR
2276  *  - if the per-sb default is different from the global default
2277  */
2278 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2279                               int nodefs)
2280 {
2281         struct ext4_sb_info *sbi = EXT4_SB(sb);
2282         struct ext4_super_block *es = sbi->s_es;
2283         int def_errors, def_mount_opt = sbi->s_def_mount_opt;
2284         const struct mount_opts *m;
2285         char sep = nodefs ? '\n' : ',';
2286
2287 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2288 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2289
2290         if (sbi->s_sb_block != 1)
2291                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2292
2293         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2294                 int want_set = m->flags & MOPT_SET;
2295                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2296                     (m->flags & MOPT_CLEAR_ERR))
2297                         continue;
2298                 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2299                         continue; /* skip if same as the default */
2300                 if ((want_set &&
2301                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2302                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2303                         continue; /* select Opt_noFoo vs Opt_Foo */
2304                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2305         }
2306
2307         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2308             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2309                 SEQ_OPTS_PRINT("resuid=%u",
2310                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2311         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2312             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2313                 SEQ_OPTS_PRINT("resgid=%u",
2314                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2315         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2316         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2317                 SEQ_OPTS_PUTS("errors=remount-ro");
2318         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2319                 SEQ_OPTS_PUTS("errors=continue");
2320         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2321                 SEQ_OPTS_PUTS("errors=panic");
2322         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2323                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2324         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2325                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2326         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2327                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2328         if (sb->s_flags & SB_I_VERSION)
2329                 SEQ_OPTS_PUTS("i_version");
2330         if (nodefs || sbi->s_stripe)
2331                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2332         if (nodefs || EXT4_MOUNT_DATA_FLAGS &
2333                         (sbi->s_mount_opt ^ def_mount_opt)) {
2334                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2335                         SEQ_OPTS_PUTS("data=journal");
2336                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2337                         SEQ_OPTS_PUTS("data=ordered");
2338                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2339                         SEQ_OPTS_PUTS("data=writeback");
2340         }
2341         if (nodefs ||
2342             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2343                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2344                                sbi->s_inode_readahead_blks);
2345
2346         if (test_opt(sb, INIT_INODE_TABLE) && (nodefs ||
2347                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2348                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2349         if (nodefs || sbi->s_max_dir_size_kb)
2350                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2351         if (test_opt(sb, DATA_ERR_ABORT))
2352                 SEQ_OPTS_PUTS("data_err=abort");
2353
2354         fscrypt_show_test_dummy_encryption(seq, sep, sb);
2355
2356         ext4_show_quota_options(seq, sb);
2357         return 0;
2358 }
2359
2360 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2361 {
2362         return _ext4_show_options(seq, root->d_sb, 0);
2363 }
2364
2365 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2366 {
2367         struct super_block *sb = seq->private;
2368         int rc;
2369
2370         seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
2371         rc = _ext4_show_options(seq, sb, 1);
2372         seq_puts(seq, "\n");
2373         return rc;
2374 }
2375
2376 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2377                             int read_only)
2378 {
2379         struct ext4_sb_info *sbi = EXT4_SB(sb);
2380         int err = 0;
2381
2382         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2383                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2384                          "forcing read-only mode");
2385                 err = -EROFS;
2386         }
2387         if (read_only)
2388                 goto done;
2389         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2390                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2391                          "running e2fsck is recommended");
2392         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2393                 ext4_msg(sb, KERN_WARNING,
2394                          "warning: mounting fs with errors, "
2395                          "running e2fsck is recommended");
2396         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2397                  le16_to_cpu(es->s_mnt_count) >=
2398                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2399                 ext4_msg(sb, KERN_WARNING,
2400                          "warning: maximal mount count reached, "
2401                          "running e2fsck is recommended");
2402         else if (le32_to_cpu(es->s_checkinterval) &&
2403                  (ext4_get_tstamp(es, s_lastcheck) +
2404                   le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds()))
2405                 ext4_msg(sb, KERN_WARNING,
2406                          "warning: checktime reached, "
2407                          "running e2fsck is recommended");
2408         if (!sbi->s_journal)
2409                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2410         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2411                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2412         le16_add_cpu(&es->s_mnt_count, 1);
2413         ext4_update_tstamp(es, s_mtime);
2414         if (sbi->s_journal)
2415                 ext4_set_feature_journal_needs_recovery(sb);
2416
2417         err = ext4_commit_super(sb, 1);
2418 done:
2419         if (test_opt(sb, DEBUG))
2420                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2421                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2422                         sb->s_blocksize,
2423                         sbi->s_groups_count,
2424                         EXT4_BLOCKS_PER_GROUP(sb),
2425                         EXT4_INODES_PER_GROUP(sb),
2426                         sbi->s_mount_opt, sbi->s_mount_opt2);
2427
2428         cleancache_init_fs(sb);
2429         return err;
2430 }
2431
2432 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2433 {
2434         struct ext4_sb_info *sbi = EXT4_SB(sb);
2435         struct flex_groups **old_groups, **new_groups;
2436         int size, i, j;
2437
2438         if (!sbi->s_log_groups_per_flex)
2439                 return 0;
2440
2441         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2442         if (size <= sbi->s_flex_groups_allocated)
2443                 return 0;
2444
2445         new_groups = kvzalloc(roundup_pow_of_two(size *
2446                               sizeof(*sbi->s_flex_groups)), GFP_KERNEL);
2447         if (!new_groups) {
2448                 ext4_msg(sb, KERN_ERR,
2449                          "not enough memory for %d flex group pointers", size);
2450                 return -ENOMEM;
2451         }
2452         for (i = sbi->s_flex_groups_allocated; i < size; i++) {
2453                 new_groups[i] = kvzalloc(roundup_pow_of_two(
2454                                          sizeof(struct flex_groups)),
2455                                          GFP_KERNEL);
2456                 if (!new_groups[i]) {
2457                         for (j = sbi->s_flex_groups_allocated; j < i; j++)
2458                                 kvfree(new_groups[j]);
2459                         kvfree(new_groups);
2460                         ext4_msg(sb, KERN_ERR,
2461                                  "not enough memory for %d flex groups", size);
2462                         return -ENOMEM;
2463                 }
2464         }
2465         rcu_read_lock();
2466         old_groups = rcu_dereference(sbi->s_flex_groups);
2467         if (old_groups)
2468                 memcpy(new_groups, old_groups,
2469                        (sbi->s_flex_groups_allocated *
2470                         sizeof(struct flex_groups *)));
2471         rcu_read_unlock();
2472         rcu_assign_pointer(sbi->s_flex_groups, new_groups);
2473         sbi->s_flex_groups_allocated = size;
2474         if (old_groups)
2475                 ext4_kvfree_array_rcu(old_groups);
2476         return 0;
2477 }
2478
2479 static int ext4_fill_flex_info(struct super_block *sb)
2480 {
2481         struct ext4_sb_info *sbi = EXT4_SB(sb);
2482         struct ext4_group_desc *gdp = NULL;
2483         struct flex_groups *fg;
2484         ext4_group_t flex_group;
2485         int i, err;
2486
2487         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2488         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2489                 sbi->s_log_groups_per_flex = 0;
2490                 return 1;
2491         }
2492
2493         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2494         if (err)
2495                 goto failed;
2496
2497         for (i = 0; i < sbi->s_groups_count; i++) {
2498                 gdp = ext4_get_group_desc(sb, i, NULL);
2499
2500                 flex_group = ext4_flex_group(sbi, i);
2501                 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
2502                 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes);
2503                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2504                              &fg->free_clusters);
2505                 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs);
2506         }
2507
2508         return 1;
2509 failed:
2510         return 0;
2511 }
2512
2513 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2514                                    struct ext4_group_desc *gdp)
2515 {
2516         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2517         __u16 crc = 0;
2518         __le32 le_group = cpu_to_le32(block_group);
2519         struct ext4_sb_info *sbi = EXT4_SB(sb);
2520
2521         if (ext4_has_metadata_csum(sbi->s_sb)) {
2522                 /* Use new metadata_csum algorithm */
2523                 __u32 csum32;
2524                 __u16 dummy_csum = 0;
2525
2526                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2527                                      sizeof(le_group));
2528                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2529                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2530                                      sizeof(dummy_csum));
2531                 offset += sizeof(dummy_csum);
2532                 if (offset < sbi->s_desc_size)
2533                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2534                                              sbi->s_desc_size - offset);
2535
2536                 crc = csum32 & 0xFFFF;
2537                 goto out;
2538         }
2539
2540         /* old crc16 code */
2541         if (!ext4_has_feature_gdt_csum(sb))
2542                 return 0;
2543
2544         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2545         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2546         crc = crc16(crc, (__u8 *)gdp, offset);
2547         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2548         /* for checksum of struct ext4_group_desc do the rest...*/
2549         if (ext4_has_feature_64bit(sb) &&
2550             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2551                 crc = crc16(crc, (__u8 *)gdp + offset,
2552                             le16_to_cpu(sbi->s_es->s_desc_size) -
2553                                 offset);
2554
2555 out:
2556         return cpu_to_le16(crc);
2557 }
2558
2559 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2560                                 struct ext4_group_desc *gdp)
2561 {
2562         if (ext4_has_group_desc_csum(sb) &&
2563             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2564                 return 0;
2565
2566         return 1;
2567 }
2568
2569 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2570                               struct ext4_group_desc *gdp)
2571 {
2572         if (!ext4_has_group_desc_csum(sb))
2573                 return;
2574         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2575 }
2576
2577 /* Called at mount-time, super-block is locked */
2578 static int ext4_check_descriptors(struct super_block *sb,
2579                                   ext4_fsblk_t sb_block,
2580                                   ext4_group_t *first_not_zeroed)
2581 {
2582         struct ext4_sb_info *sbi = EXT4_SB(sb);
2583         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2584         ext4_fsblk_t last_block;
2585         ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0);
2586         ext4_fsblk_t block_bitmap;
2587         ext4_fsblk_t inode_bitmap;
2588         ext4_fsblk_t inode_table;
2589         int flexbg_flag = 0;
2590         ext4_group_t i, grp = sbi->s_groups_count;
2591
2592         if (ext4_has_feature_flex_bg(sb))
2593                 flexbg_flag = 1;
2594
2595         ext4_debug("Checking group descriptors");
2596
2597         for (i = 0; i < sbi->s_groups_count; i++) {
2598                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2599
2600                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2601                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2602                 else
2603                         last_block = first_block +
2604                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2605
2606                 if ((grp == sbi->s_groups_count) &&
2607                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2608                         grp = i;
2609
2610                 block_bitmap = ext4_block_bitmap(sb, gdp);
2611                 if (block_bitmap == sb_block) {
2612                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2613                                  "Block bitmap for group %u overlaps "
2614                                  "superblock", i);
2615                         if (!sb_rdonly(sb))
2616                                 return 0;
2617                 }
2618                 if (block_bitmap >= sb_block + 1 &&
2619                     block_bitmap <= last_bg_block) {
2620                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2621                                  "Block bitmap for group %u overlaps "
2622                                  "block group descriptors", i);
2623                         if (!sb_rdonly(sb))
2624                                 return 0;
2625                 }
2626                 if (block_bitmap < first_block || block_bitmap > last_block) {
2627                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2628                                "Block bitmap for group %u not in group "
2629                                "(block %llu)!", i, block_bitmap);
2630                         return 0;
2631                 }
2632                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2633                 if (inode_bitmap == sb_block) {
2634                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2635                                  "Inode bitmap for group %u overlaps "
2636                                  "superblock", i);
2637                         if (!sb_rdonly(sb))
2638                                 return 0;
2639                 }
2640                 if (inode_bitmap >= sb_block + 1 &&
2641                     inode_bitmap <= last_bg_block) {
2642                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2643                                  "Inode bitmap for group %u overlaps "
2644                                  "block group descriptors", i);
2645                         if (!sb_rdonly(sb))
2646                                 return 0;
2647                 }
2648                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2649                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2650                                "Inode bitmap for group %u not in group "
2651                                "(block %llu)!", i, inode_bitmap);
2652                         return 0;
2653                 }
2654                 inode_table = ext4_inode_table(sb, gdp);
2655                 if (inode_table == sb_block) {
2656                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2657                                  "Inode table for group %u overlaps "
2658                                  "superblock", i);
2659                         if (!sb_rdonly(sb))
2660                                 return 0;
2661                 }
2662                 if (inode_table >= sb_block + 1 &&
2663                     inode_table <= last_bg_block) {
2664                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2665                                  "Inode table for group %u overlaps "
2666                                  "block group descriptors", i);
2667                         if (!sb_rdonly(sb))
2668                                 return 0;
2669                 }
2670                 if (inode_table < first_block ||
2671                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2672                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2673                                "Inode table for group %u not in group "
2674                                "(block %llu)!", i, inode_table);
2675                         return 0;
2676                 }
2677                 ext4_lock_group(sb, i);
2678                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2679                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2680                                  "Checksum for group %u failed (%u!=%u)",
2681                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2682                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2683                         if (!sb_rdonly(sb)) {
2684                                 ext4_unlock_group(sb, i);
2685                                 return 0;
2686                         }
2687                 }
2688                 ext4_unlock_group(sb, i);
2689                 if (!flexbg_flag)
2690                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2691         }
2692         if (NULL != first_not_zeroed)
2693                 *first_not_zeroed = grp;
2694         return 1;
2695 }
2696
2697 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2698  * the superblock) which were deleted from all directories, but held open by
2699  * a process at the time of a crash.  We walk the list and try to delete these
2700  * inodes at recovery time (only with a read-write filesystem).
2701  *
2702  * In order to keep the orphan inode chain consistent during traversal (in
2703  * case of crash during recovery), we link each inode into the superblock
2704  * orphan list_head and handle it the same way as an inode deletion during
2705  * normal operation (which journals the operations for us).
2706  *
2707  * We only do an iget() and an iput() on each inode, which is very safe if we
2708  * accidentally point at an in-use or already deleted inode.  The worst that
2709  * can happen in this case is that we get a "bit already cleared" message from
2710  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2711  * e2fsck was run on this filesystem, and it must have already done the orphan
2712  * inode cleanup for us, so we can safely abort without any further action.
2713  */
2714 static void ext4_orphan_cleanup(struct super_block *sb,
2715                                 struct ext4_super_block *es)
2716 {
2717         unsigned int s_flags = sb->s_flags;
2718         int ret, nr_orphans = 0, nr_truncates = 0;
2719 #ifdef CONFIG_QUOTA
2720         int quota_update = 0;
2721         int i;
2722 #endif
2723         if (!es->s_last_orphan) {
2724                 jbd_debug(4, "no orphan inodes to clean up\n");
2725                 return;
2726         }
2727
2728         if (bdev_read_only(sb->s_bdev)) {
2729                 ext4_msg(sb, KERN_ERR, "write access "
2730                         "unavailable, skipping orphan cleanup");
2731                 return;
2732         }
2733
2734         /* Check if feature set would not allow a r/w mount */
2735         if (!ext4_feature_set_ok(sb, 0)) {
2736                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2737                          "unknown ROCOMPAT features");
2738                 return;
2739         }
2740
2741         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2742                 /* don't clear list on RO mount w/ errors */
2743                 if (es->s_last_orphan && !(s_flags & SB_RDONLY)) {
2744                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2745                                   "clearing orphan list.\n");
2746                         es->s_last_orphan = 0;
2747                 }
2748                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2749                 return;
2750         }
2751
2752         if (s_flags & SB_RDONLY) {
2753                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2754                 sb->s_flags &= ~SB_RDONLY;
2755         }
2756 #ifdef CONFIG_QUOTA
2757         /* Needed for iput() to work correctly and not trash data */
2758         sb->s_flags |= SB_ACTIVE;
2759
2760         /*
2761          * Turn on quotas which were not enabled for read-only mounts if
2762          * filesystem has quota feature, so that they are updated correctly.
2763          */
2764         if (ext4_has_feature_quota(sb) && (s_flags & SB_RDONLY)) {
2765                 int ret = ext4_enable_quotas(sb);
2766
2767                 if (!ret)
2768                         quota_update = 1;
2769                 else
2770                         ext4_msg(sb, KERN_ERR,
2771                                 "Cannot turn on quotas: error %d", ret);
2772         }
2773
2774         /* Turn on journaled quotas used for old sytle */
2775         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2776                 if (EXT4_SB(sb)->s_qf_names[i]) {
2777                         int ret = ext4_quota_on_mount(sb, i);
2778
2779                         if (!ret)
2780                                 quota_update = 1;
2781                         else
2782                                 ext4_msg(sb, KERN_ERR,
2783                                         "Cannot turn on journaled "
2784                                         "quota: type %d: error %d", i, ret);
2785                 }
2786         }
2787 #endif
2788
2789         while (es->s_last_orphan) {
2790                 struct inode *inode;
2791
2792                 /*
2793                  * We may have encountered an error during cleanup; if
2794                  * so, skip the rest.
2795                  */
2796                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2797                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2798                         es->s_last_orphan = 0;
2799                         break;
2800                 }
2801
2802                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2803                 if (IS_ERR(inode)) {
2804                         es->s_last_orphan = 0;
2805                         break;
2806                 }
2807
2808                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2809                 dquot_initialize(inode);
2810                 if (inode->i_nlink) {
2811                         if (test_opt(sb, DEBUG))
2812                                 ext4_msg(sb, KERN_DEBUG,
2813                                         "%s: truncating inode %lu to %lld bytes",
2814                                         __func__, inode->i_ino, inode->i_size);
2815                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2816                                   inode->i_ino, inode->i_size);
2817                         inode_lock(inode);
2818                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2819                         ret = ext4_truncate(inode);
2820                         if (ret)
2821                                 ext4_std_error(inode->i_sb, ret);
2822                         inode_unlock(inode);
2823                         nr_truncates++;
2824                 } else {
2825                         if (test_opt(sb, DEBUG))
2826                                 ext4_msg(sb, KERN_DEBUG,
2827                                         "%s: deleting unreferenced inode %lu",
2828                                         __func__, inode->i_ino);
2829                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2830                                   inode->i_ino);
2831                         nr_orphans++;
2832                 }
2833                 iput(inode);  /* The delete magic happens here! */
2834         }
2835
2836 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2837
2838         if (nr_orphans)
2839                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2840                        PLURAL(nr_orphans));
2841         if (nr_truncates)
2842                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2843                        PLURAL(nr_truncates));
2844 #ifdef CONFIG_QUOTA
2845         /* Turn off quotas if they were enabled for orphan cleanup */
2846         if (quota_update) {
2847                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2848                         if (sb_dqopt(sb)->files[i])
2849                                 dquot_quota_off(sb, i);
2850                 }
2851         }
2852 #endif
2853         sb->s_flags = s_flags; /* Restore SB_RDONLY status */
2854 }
2855
2856 /*
2857  * Maximal extent format file size.
2858  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2859  * extent format containers, within a sector_t, and within i_blocks
2860  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2861  * so that won't be a limiting factor.
2862  *
2863  * However there is other limiting factor. We do store extents in the form
2864  * of starting block and length, hence the resulting length of the extent
2865  * covering maximum file size must fit into on-disk format containers as
2866  * well. Given that length is always by 1 unit bigger than max unit (because
2867  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2868  *
2869  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2870  */
2871 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2872 {
2873         loff_t res;
2874         loff_t upper_limit = MAX_LFS_FILESIZE;
2875
2876         BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64));
2877
2878         if (!has_huge_files) {
2879                 upper_limit = (1LL << 32) - 1;
2880
2881                 /* total blocks in file system block size */
2882                 upper_limit >>= (blkbits - 9);
2883                 upper_limit <<= blkbits;
2884         }
2885
2886         /*
2887          * 32-bit extent-start container, ee_block. We lower the maxbytes
2888          * by one fs block, so ee_len can cover the extent of maximum file
2889          * size
2890          */
2891         res = (1LL << 32) - 1;
2892         res <<= blkbits;
2893
2894         /* Sanity check against vm- & vfs- imposed limits */
2895         if (res > upper_limit)
2896                 res = upper_limit;
2897
2898         return res;
2899 }
2900
2901 /*
2902  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2903  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2904  * We need to be 1 filesystem block less than the 2^48 sector limit.
2905  */
2906 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2907 {
2908         loff_t res = EXT4_NDIR_BLOCKS;
2909         int meta_blocks;
2910         loff_t upper_limit;
2911         /* This is calculated to be the largest file size for a dense, block
2912          * mapped file such that the file's total number of 512-byte sectors,
2913          * including data and all indirect blocks, does not exceed (2^48 - 1).
2914          *
2915          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2916          * number of 512-byte sectors of the file.
2917          */
2918
2919         if (!has_huge_files) {
2920                 /*
2921                  * !has_huge_files or implies that the inode i_block field
2922                  * represents total file blocks in 2^32 512-byte sectors ==
2923                  * size of vfs inode i_blocks * 8
2924                  */
2925                 upper_limit = (1LL << 32) - 1;
2926
2927                 /* total blocks in file system block size */
2928                 upper_limit >>= (bits - 9);
2929
2930         } else {
2931                 /*
2932                  * We use 48 bit ext4_inode i_blocks
2933                  * With EXT4_HUGE_FILE_FL set the i_blocks
2934                  * represent total number of blocks in
2935                  * file system block size
2936                  */
2937                 upper_limit = (1LL << 48) - 1;
2938
2939         }
2940
2941         /* indirect blocks */
2942         meta_blocks = 1;
2943         /* double indirect blocks */
2944         meta_blocks += 1 + (1LL << (bits-2));
2945         /* tripple indirect blocks */
2946         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2947
2948         upper_limit -= meta_blocks;
2949         upper_limit <<= bits;
2950
2951         res += 1LL << (bits-2);
2952         res += 1LL << (2*(bits-2));
2953         res += 1LL << (3*(bits-2));
2954         res <<= bits;
2955         if (res > upper_limit)
2956                 res = upper_limit;
2957
2958         if (res > MAX_LFS_FILESIZE)
2959                 res = MAX_LFS_FILESIZE;
2960
2961         return res;
2962 }
2963
2964 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2965                                    ext4_fsblk_t logical_sb_block, int nr)
2966 {
2967         struct ext4_sb_info *sbi = EXT4_SB(sb);
2968         ext4_group_t bg, first_meta_bg;
2969         int has_super = 0;
2970
2971         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2972
2973         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2974                 return logical_sb_block + nr + 1;
2975         bg = sbi->s_desc_per_block * nr;
2976         if (ext4_bg_has_super(sb, bg))
2977                 has_super = 1;
2978
2979         /*
2980          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2981          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2982          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2983          * compensate.
2984          */
2985         if (sb->s_blocksize == 1024 && nr == 0 &&
2986             le32_to_cpu(sbi->s_es->s_first_data_block) == 0)
2987                 has_super++;
2988
2989         return (has_super + ext4_group_first_block_no(sb, bg));
2990 }
2991
2992 /**
2993  * ext4_get_stripe_size: Get the stripe size.
2994  * @sbi: In memory super block info
2995  *
2996  * If we have specified it via mount option, then
2997  * use the mount option value. If the value specified at mount time is
2998  * greater than the blocks per group use the super block value.
2999  * If the super block value is greater than blocks per group return 0.
3000  * Allocator needs it be less than blocks per group.
3001  *
3002  */
3003 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
3004 {
3005         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
3006         unsigned long stripe_width =
3007                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
3008         int ret;
3009
3010         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
3011                 ret = sbi->s_stripe;
3012         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
3013                 ret = stripe_width;
3014         else if (stride && stride <= sbi->s_blocks_per_group)
3015                 ret = stride;
3016         else
3017                 ret = 0;
3018
3019         /*
3020          * If the stripe width is 1, this makes no sense and
3021          * we set it to 0 to turn off stripe handling code.
3022          */
3023         if (ret <= 1)
3024                 ret = 0;
3025
3026         return ret;
3027 }
3028
3029 /*
3030  * Check whether this filesystem can be mounted based on
3031  * the features present and the RDONLY/RDWR mount requested.
3032  * Returns 1 if this filesystem can be mounted as requested,
3033  * 0 if it cannot be.
3034  */
3035 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
3036 {
3037         if (ext4_has_unknown_ext4_incompat_features(sb)) {
3038                 ext4_msg(sb, KERN_ERR,
3039                         "Couldn't mount because of "
3040                         "unsupported optional features (%x)",
3041                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
3042                         ~EXT4_FEATURE_INCOMPAT_SUPP));
3043                 return 0;
3044         }
3045
3046 #ifndef CONFIG_UNICODE
3047         if (ext4_has_feature_casefold(sb)) {
3048                 ext4_msg(sb, KERN_ERR,
3049                          "Filesystem with casefold feature cannot be "
3050                          "mounted without CONFIG_UNICODE");
3051                 return 0;
3052         }
3053 #endif
3054
3055         if (readonly)
3056                 return 1;
3057
3058         if (ext4_has_feature_readonly(sb)) {
3059                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
3060                 sb->s_flags |= SB_RDONLY;
3061                 return 1;
3062         }
3063
3064         /* Check that feature set is OK for a read-write mount */
3065         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
3066                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
3067                          "unsupported optional features (%x)",
3068                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
3069                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
3070                 return 0;
3071         }
3072         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
3073                 ext4_msg(sb, KERN_ERR,
3074                          "Can't support bigalloc feature without "
3075                          "extents feature\n");
3076                 return 0;
3077         }
3078
3079 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2)
3080         if (!readonly && (ext4_has_feature_quota(sb) ||
3081                           ext4_has_feature_project(sb))) {
3082                 ext4_msg(sb, KERN_ERR,
3083                          "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2");
3084                 return 0;
3085         }
3086 #endif  /* CONFIG_QUOTA */
3087         return 1;
3088 }
3089
3090 /*
3091  * This function is called once a day if we have errors logged
3092  * on the file system
3093  */
3094 static void print_daily_error_info(struct timer_list *t)
3095 {
3096         struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report);
3097         struct super_block *sb = sbi->s_sb;
3098         struct ext4_super_block *es = sbi->s_es;
3099
3100         if (es->s_error_count)
3101                 /* fsck newer than v1.41.13 is needed to clean this condition. */
3102                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
3103                          le32_to_cpu(es->s_error_count));
3104         if (es->s_first_error_time) {
3105                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d",
3106                        sb->s_id,
3107                        ext4_get_tstamp(es, s_first_error_time),
3108                        (int) sizeof(es->s_first_error_func),
3109                        es->s_first_error_func,
3110                        le32_to_cpu(es->s_first_error_line));
3111                 if (es->s_first_error_ino)
3112                         printk(KERN_CONT ": inode %u",
3113                                le32_to_cpu(es->s_first_error_ino));
3114                 if (es->s_first_error_block)
3115                         printk(KERN_CONT ": block %llu", (unsigned long long)
3116                                le64_to_cpu(es->s_first_error_block));
3117                 printk(KERN_CONT "\n");
3118         }
3119         if (es->s_last_error_time) {
3120                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d",
3121                        sb->s_id,
3122                        ext4_get_tstamp(es, s_last_error_time),
3123                        (int) sizeof(es->s_last_error_func),
3124                        es->s_last_error_func,
3125                        le32_to_cpu(es->s_last_error_line));
3126                 if (es->s_last_error_ino)
3127                         printk(KERN_CONT ": inode %u",
3128                                le32_to_cpu(es->s_last_error_ino));
3129                 if (es->s_last_error_block)
3130                         printk(KERN_CONT ": block %llu", (unsigned long long)
3131                                le64_to_cpu(es->s_last_error_block));
3132                 printk(KERN_CONT "\n");
3133         }
3134         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
3135 }
3136
3137 /* Find next suitable group and run ext4_init_inode_table */
3138 static int ext4_run_li_request(struct ext4_li_request *elr)
3139 {
3140         struct ext4_group_desc *gdp = NULL;
3141         ext4_group_t group, ngroups;
3142         struct super_block *sb;
3143         unsigned long timeout = 0;
3144         int ret = 0;
3145
3146         sb = elr->lr_super;
3147         ngroups = EXT4_SB(sb)->s_groups_count;
3148
3149         for (group = elr->lr_next_group; group < ngroups; group++) {
3150                 gdp = ext4_get_group_desc(sb, group, NULL);
3151                 if (!gdp) {
3152                         ret = 1;
3153                         break;
3154                 }
3155
3156                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3157                         break;
3158         }
3159
3160         if (group >= ngroups)
3161                 ret = 1;
3162
3163         if (!ret) {
3164                 timeout = jiffies;
3165                 ret = ext4_init_inode_table(sb, group,
3166                                             elr->lr_timeout ? 0 : 1);
3167                 if (elr->lr_timeout == 0) {
3168                         timeout = (jiffies - timeout) *
3169                                   elr->lr_sbi->s_li_wait_mult;
3170                         elr->lr_timeout = timeout;
3171                 }
3172                 elr->lr_next_sched = jiffies + elr->lr_timeout;
3173                 elr->lr_next_group = group + 1;
3174         }
3175         return ret;
3176 }
3177
3178 /*
3179  * Remove lr_request from the list_request and free the
3180  * request structure. Should be called with li_list_mtx held
3181  */
3182 static void ext4_remove_li_request(struct ext4_li_request *elr)
3183 {
3184         struct ext4_sb_info *sbi;
3185
3186         if (!elr)
3187                 return;
3188
3189         sbi = elr->lr_sbi;
3190
3191         list_del(&elr->lr_request);
3192         sbi->s_li_request = NULL;
3193         kfree(elr);
3194 }
3195
3196 static void ext4_unregister_li_request(struct super_block *sb)
3197 {
3198         mutex_lock(&ext4_li_mtx);
3199         if (!ext4_li_info) {
3200                 mutex_unlock(&ext4_li_mtx);
3201                 return;
3202         }
3203
3204         mutex_lock(&ext4_li_info->li_list_mtx);
3205         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
3206         mutex_unlock(&ext4_li_info->li_list_mtx);
3207         mutex_unlock(&ext4_li_mtx);
3208 }
3209
3210 static struct task_struct *ext4_lazyinit_task;
3211
3212 /*
3213  * This is the function where ext4lazyinit thread lives. It walks
3214  * through the request list searching for next scheduled filesystem.
3215  * When such a fs is found, run the lazy initialization request
3216  * (ext4_rn_li_request) and keep track of the time spend in this
3217  * function. Based on that time we compute next schedule time of
3218  * the request. When walking through the list is complete, compute
3219  * next waking time and put itself into sleep.
3220  */
3221 static int ext4_lazyinit_thread(void *arg)
3222 {
3223         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
3224         struct list_head *pos, *n;
3225         struct ext4_li_request *elr;
3226         unsigned long next_wakeup, cur;
3227
3228         BUG_ON(NULL == eli);
3229
3230 cont_thread:
3231         while (true) {
3232                 next_wakeup = MAX_JIFFY_OFFSET;
3233
3234                 mutex_lock(&eli->li_list_mtx);
3235                 if (list_empty(&eli->li_request_list)) {
3236                         mutex_unlock(&eli->li_list_mtx);
3237                         goto exit_thread;
3238                 }
3239                 list_for_each_safe(pos, n, &eli->li_request_list) {
3240                         int err = 0;
3241                         int progress = 0;
3242                         elr = list_entry(pos, struct ext4_li_request,
3243                                          lr_request);
3244
3245                         if (time_before(jiffies, elr->lr_next_sched)) {
3246                                 if (time_before(elr->lr_next_sched, next_wakeup))
3247                                         next_wakeup = elr->lr_next_sched;
3248                                 continue;
3249                         }
3250                         if (down_read_trylock(&elr->lr_super->s_umount)) {
3251                                 if (sb_start_write_trylock(elr->lr_super)) {
3252                                         progress = 1;
3253                                         /*
3254                                          * We hold sb->s_umount, sb can not
3255                                          * be removed from the list, it is
3256                                          * now safe to drop li_list_mtx
3257                                          */
3258                                         mutex_unlock(&eli->li_list_mtx);
3259                                         err = ext4_run_li_request(elr);
3260                                         sb_end_write(elr->lr_super);
3261                                         mutex_lock(&eli->li_list_mtx);
3262                                         n = pos->next;
3263                                 }
3264                                 up_read((&elr->lr_super->s_umount));
3265                         }
3266                         /* error, remove the lazy_init job */
3267                         if (err) {
3268                                 ext4_remove_li_request(elr);
3269                                 continue;
3270                         }
3271                         if (!progress) {
3272                                 elr->lr_next_sched = jiffies +
3273                                         (prandom_u32()
3274                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3275                         }
3276                         if (time_before(elr->lr_next_sched, next_wakeup))
3277                                 next_wakeup = elr->lr_next_sched;
3278                 }
3279                 mutex_unlock(&eli->li_list_mtx);
3280
3281                 try_to_freeze();
3282
3283                 cur = jiffies;
3284                 if ((time_after_eq(cur, next_wakeup)) ||
3285                     (MAX_JIFFY_OFFSET == next_wakeup)) {
3286                         cond_resched();
3287                         continue;
3288                 }
3289
3290                 schedule_timeout_interruptible(next_wakeup - cur);
3291
3292                 if (kthread_should_stop()) {
3293                         ext4_clear_request_list();
3294                         goto exit_thread;
3295                 }
3296         }
3297
3298 exit_thread:
3299         /*
3300          * It looks like the request list is empty, but we need
3301          * to check it under the li_list_mtx lock, to prevent any
3302          * additions into it, and of course we should lock ext4_li_mtx
3303          * to atomically free the list and ext4_li_info, because at
3304          * this point another ext4 filesystem could be registering
3305          * new one.
3306          */
3307         mutex_lock(&ext4_li_mtx);
3308         mutex_lock(&eli->li_list_mtx);
3309         if (!list_empty(&eli->li_request_list)) {
3310                 mutex_unlock(&eli->li_list_mtx);
3311                 mutex_unlock(&ext4_li_mtx);
3312                 goto cont_thread;
3313         }
3314         mutex_unlock(&eli->li_list_mtx);
3315         kfree(ext4_li_info);
3316         ext4_li_info = NULL;
3317         mutex_unlock(&ext4_li_mtx);
3318
3319         return 0;
3320 }
3321
3322 static void ext4_clear_request_list(void)
3323 {
3324         struct list_head *pos, *n;
3325         struct ext4_li_request *elr;
3326
3327         mutex_lock(&ext4_li_info->li_list_mtx);
3328         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3329                 elr = list_entry(pos, struct ext4_li_request,
3330                                  lr_request);
3331                 ext4_remove_li_request(elr);
3332         }
3333         mutex_unlock(&ext4_li_info->li_list_mtx);
3334 }
3335
3336 static int ext4_run_lazyinit_thread(void)
3337 {
3338         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3339                                          ext4_li_info, "ext4lazyinit");
3340         if (IS_ERR(ext4_lazyinit_task)) {
3341                 int err = PTR_ERR(ext4_lazyinit_task);
3342                 ext4_clear_request_list();
3343                 kfree(ext4_li_info);
3344                 ext4_li_info = NULL;
3345                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3346                                  "initialization thread\n",
3347                                  err);
3348                 return err;
3349         }
3350         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3351         return 0;
3352 }
3353
3354 /*
3355  * Check whether it make sense to run itable init. thread or not.
3356  * If there is at least one uninitialized inode table, return
3357  * corresponding group number, else the loop goes through all
3358  * groups and return total number of groups.
3359  */
3360 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3361 {
3362         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3363         struct ext4_group_desc *gdp = NULL;
3364
3365         if (!ext4_has_group_desc_csum(sb))
3366                 return ngroups;
3367
3368         for (group = 0; group < ngroups; group++) {
3369                 gdp = ext4_get_group_desc(sb, group, NULL);
3370                 if (!gdp)
3371                         continue;
3372
3373                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3374                         break;
3375         }
3376
3377         return group;
3378 }
3379
3380 static int ext4_li_info_new(void)
3381 {
3382         struct ext4_lazy_init *eli = NULL;
3383
3384         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3385         if (!eli)
3386                 return -ENOMEM;
3387
3388         INIT_LIST_HEAD(&eli->li_request_list);
3389         mutex_init(&eli->li_list_mtx);
3390
3391         eli->li_state |= EXT4_LAZYINIT_QUIT;
3392
3393         ext4_li_info = eli;
3394
3395         return 0;
3396 }
3397
3398 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3399                                             ext4_group_t start)
3400 {
3401         struct ext4_sb_info *sbi = EXT4_SB(sb);
3402         struct ext4_li_request *elr;
3403
3404         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3405         if (!elr)
3406                 return NULL;
3407
3408         elr->lr_super = sb;
3409         elr->lr_sbi = sbi;
3410         elr->lr_next_group = start;
3411
3412         /*
3413          * Randomize first schedule time of the request to
3414          * spread the inode table initialization requests
3415          * better.
3416          */
3417         elr->lr_next_sched = jiffies + (prandom_u32() %
3418                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3419         return elr;
3420 }
3421
3422 int ext4_register_li_request(struct super_block *sb,
3423                              ext4_group_t first_not_zeroed)
3424 {
3425         struct ext4_sb_info *sbi = EXT4_SB(sb);
3426         struct ext4_li_request *elr = NULL;
3427         ext4_group_t ngroups = sbi->s_groups_count;
3428         int ret = 0;
3429
3430         mutex_lock(&ext4_li_mtx);
3431         if (sbi->s_li_request != NULL) {
3432                 /*
3433                  * Reset timeout so it can be computed again, because
3434                  * s_li_wait_mult might have changed.
3435                  */
3436                 sbi->s_li_request->lr_timeout = 0;
3437                 goto out;
3438         }
3439
3440         if (first_not_zeroed == ngroups || sb_rdonly(sb) ||
3441             !test_opt(sb, INIT_INODE_TABLE))
3442                 goto out;
3443
3444         elr = ext4_li_request_new(sb, first_not_zeroed);
3445         if (!elr) {
3446                 ret = -ENOMEM;
3447                 goto out;
3448         }
3449
3450         if (NULL == ext4_li_info) {
3451                 ret = ext4_li_info_new();
3452                 if (ret)
3453                         goto out;
3454         }
3455
3456         mutex_lock(&ext4_li_info->li_list_mtx);
3457         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3458         mutex_unlock(&ext4_li_info->li_list_mtx);
3459
3460         sbi->s_li_request = elr;
3461         /*
3462          * set elr to NULL here since it has been inserted to
3463          * the request_list and the removal and free of it is
3464          * handled by ext4_clear_request_list from now on.
3465          */
3466         elr = NULL;
3467
3468         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3469                 ret = ext4_run_lazyinit_thread();
3470                 if (ret)
3471                         goto out;
3472         }
3473 out:
3474         mutex_unlock(&ext4_li_mtx);
3475         if (ret)
3476                 kfree(elr);
3477         return ret;
3478 }
3479
3480 /*
3481  * We do not need to lock anything since this is called on
3482  * module unload.
3483  */
3484 static void ext4_destroy_lazyinit_thread(void)
3485 {
3486         /*
3487          * If thread exited earlier
3488          * there's nothing to be done.
3489          */
3490         if (!ext4_li_info || !ext4_lazyinit_task)
3491                 return;
3492
3493         kthread_stop(ext4_lazyinit_task);
3494 }
3495
3496 static int set_journal_csum_feature_set(struct super_block *sb)
3497 {
3498         int ret = 1;
3499         int compat, incompat;
3500         struct ext4_sb_info *sbi = EXT4_SB(sb);
3501
3502         if (ext4_has_metadata_csum(sb)) {
3503                 /* journal checksum v3 */
3504                 compat = 0;
3505                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3506         } else {
3507                 /* journal checksum v1 */
3508                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3509                 incompat = 0;
3510         }
3511
3512         jbd2_journal_clear_features(sbi->s_journal,
3513                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3514                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3515                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3516         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3517                 ret = jbd2_journal_set_features(sbi->s_journal,
3518                                 compat, 0,
3519                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3520                                 incompat);
3521         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3522                 ret = jbd2_journal_set_features(sbi->s_journal,
3523                                 compat, 0,
3524                                 incompat);
3525                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3526                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3527         } else {
3528                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3529                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3530         }
3531
3532         return ret;
3533 }
3534
3535 /*
3536  * Note: calculating the overhead so we can be compatible with
3537  * historical BSD practice is quite difficult in the face of
3538  * clusters/bigalloc.  This is because multiple metadata blocks from
3539  * different block group can end up in the same allocation cluster.
3540  * Calculating the exact overhead in the face of clustered allocation
3541  * requires either O(all block bitmaps) in memory or O(number of block
3542  * groups**2) in time.  We will still calculate the superblock for
3543  * older file systems --- and if we come across with a bigalloc file
3544  * system with zero in s_overhead_clusters the estimate will be close to
3545  * correct especially for very large cluster sizes --- but for newer
3546  * file systems, it's better to calculate this figure once at mkfs
3547  * time, and store it in the superblock.  If the superblock value is
3548  * present (even for non-bigalloc file systems), we will use it.
3549  */
3550 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3551                           char *buf)
3552 {
3553         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3554         struct ext4_group_desc  *gdp;
3555         ext4_fsblk_t            first_block, last_block, b;
3556         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3557         int                     s, j, count = 0;
3558
3559         if (!ext4_has_feature_bigalloc(sb))
3560                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3561                         sbi->s_itb_per_group + 2);
3562
3563         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3564                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3565         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3566         for (i = 0; i < ngroups; i++) {
3567                 gdp = ext4_get_group_desc(sb, i, NULL);
3568                 b = ext4_block_bitmap(sb, gdp);
3569                 if (b >= first_block && b <= last_block) {
3570                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3571                         count++;
3572                 }
3573                 b = ext4_inode_bitmap(sb, gdp);
3574                 if (b >= first_block && b <= last_block) {
3575                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3576                         count++;
3577                 }
3578                 b = ext4_inode_table(sb, gdp);
3579                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3580                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3581                                 int c = EXT4_B2C(sbi, b - first_block);
3582                                 ext4_set_bit(c, buf);
3583                                 count++;
3584                         }
3585                 if (i != grp)
3586                         continue;
3587                 s = 0;
3588                 if (ext4_bg_has_super(sb, grp)) {
3589                         ext4_set_bit(s++, buf);
3590                         count++;
3591                 }
3592                 j = ext4_bg_num_gdb(sb, grp);
3593                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3594                         ext4_error(sb, "Invalid number of block group "
3595                                    "descriptor blocks: %d", j);
3596                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3597                 }
3598                 count += j;
3599                 for (; j > 0; j--)
3600                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3601         }
3602         if (!count)
3603                 return 0;
3604         return EXT4_CLUSTERS_PER_GROUP(sb) -
3605                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3606 }
3607
3608 /*
3609  * Compute the overhead and stash it in sbi->s_overhead
3610  */
3611 int ext4_calculate_overhead(struct super_block *sb)
3612 {
3613         struct ext4_sb_info *sbi = EXT4_SB(sb);
3614         struct ext4_super_block *es = sbi->s_es;
3615         struct inode *j_inode;
3616         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3617         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3618         ext4_fsblk_t overhead = 0;
3619         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3620
3621         if (!buf)
3622                 return -ENOMEM;
3623
3624         /*
3625          * Compute the overhead (FS structures).  This is constant
3626          * for a given filesystem unless the number of block groups
3627          * changes so we cache the previous value until it does.
3628          */
3629
3630         /*
3631          * All of the blocks before first_data_block are overhead
3632          */
3633         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3634
3635         /*
3636          * Add the overhead found in each block group
3637          */
3638         for (i = 0; i < ngroups; i++) {
3639                 int blks;
3640
3641                 blks = count_overhead(sb, i, buf);
3642                 overhead += blks;
3643                 if (blks)
3644                         memset(buf, 0, PAGE_SIZE);
3645                 cond_resched();
3646         }
3647
3648         /*
3649          * Add the internal journal blocks whether the journal has been
3650          * loaded or not
3651          */
3652         if (sbi->s_journal && !sbi->journal_bdev)
3653                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3654         else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) {
3655                 /* j_inum for internal journal is non-zero */
3656                 j_inode = ext4_get_journal_inode(sb, j_inum);
3657                 if (j_inode) {
3658                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3659                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
3660                         iput(j_inode);
3661                 } else {
3662                         ext4_msg(sb, KERN_ERR, "can't get journal size");
3663                 }
3664         }
3665         sbi->s_overhead = overhead;
3666         smp_wmb();
3667         free_page((unsigned long) buf);
3668         return 0;
3669 }
3670
3671 static void ext4_set_resv_clusters(struct super_block *sb)
3672 {
3673         ext4_fsblk_t resv_clusters;
3674         struct ext4_sb_info *sbi = EXT4_SB(sb);
3675
3676         /*
3677          * There's no need to reserve anything when we aren't using extents.
3678          * The space estimates are exact, there are no unwritten extents,
3679          * hole punching doesn't need new metadata... This is needed especially
3680          * to keep ext2/3 backward compatibility.
3681          */
3682         if (!ext4_has_feature_extents(sb))
3683                 return;
3684         /*
3685          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3686          * This should cover the situations where we can not afford to run
3687          * out of space like for example punch hole, or converting
3688          * unwritten extents in delalloc path. In most cases such
3689          * allocation would require 1, or 2 blocks, higher numbers are
3690          * very rare.
3691          */
3692         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3693                          sbi->s_cluster_bits);
3694
3695         do_div(resv_clusters, 50);
3696         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3697
3698         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3699 }
3700
3701 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3702 {
3703         struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev);
3704         char *orig_data = kstrdup(data, GFP_KERNEL);
3705         struct buffer_head *bh, **group_desc;
3706         struct ext4_super_block *es = NULL;
3707         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3708         struct flex_groups **flex_groups;
3709         ext4_fsblk_t block;
3710         ext4_fsblk_t sb_block = get_sb_block(&data);
3711         ext4_fsblk_t logical_sb_block;
3712         unsigned long offset = 0;
3713         unsigned long journal_devnum = 0;
3714         unsigned long def_mount_opts;
3715         struct inode *root;
3716         const char *descr;
3717         int ret = -ENOMEM;
3718         int blocksize, clustersize;
3719         unsigned int db_count;
3720         unsigned int i;
3721         int needs_recovery, has_huge_files, has_bigalloc;
3722         __u64 blocks_count;
3723         int err = 0;
3724         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3725         ext4_group_t first_not_zeroed;
3726
3727         if ((data && !orig_data) || !sbi)
3728                 goto out_free_base;
3729
3730         sbi->s_daxdev = dax_dev;
3731         sbi->s_blockgroup_lock =
3732                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3733         if (!sbi->s_blockgroup_lock)
3734                 goto out_free_base;
3735
3736         sb->s_fs_info = sbi;
3737         sbi->s_sb = sb;
3738         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3739         sbi->s_sb_block = sb_block;
3740         if (sb->s_bdev->bd_part)
3741                 sbi->s_sectors_written_start =
3742                         part_stat_read(sb->s_bdev->bd_part, sectors[STAT_WRITE]);
3743
3744         /* Cleanup superblock name */
3745         strreplace(sb->s_id, '/', '!');
3746
3747         /* -EINVAL is default */
3748         ret = -EINVAL;
3749         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3750         if (!blocksize) {
3751                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3752                 goto out_fail;
3753         }
3754
3755         /*
3756          * The ext4 superblock will not be buffer aligned for other than 1kB
3757          * block sizes.  We need to calculate the offset from buffer start.
3758          */
3759         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3760                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3761                 offset = do_div(logical_sb_block, blocksize);
3762         } else {
3763                 logical_sb_block = sb_block;
3764         }
3765
3766         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3767                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3768                 goto out_fail;
3769         }
3770         /*
3771          * Note: s_es must be initialized as soon as possible because
3772          *       some ext4 macro-instructions depend on its value
3773          */
3774         es = (struct ext4_super_block *) (bh->b_data + offset);
3775         sbi->s_es = es;
3776         sb->s_magic = le16_to_cpu(es->s_magic);
3777         if (sb->s_magic != EXT4_SUPER_MAGIC)
3778                 goto cantfind_ext4;
3779         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3780
3781         /* Warn if metadata_csum and gdt_csum are both set. */
3782         if (ext4_has_feature_metadata_csum(sb) &&
3783             ext4_has_feature_gdt_csum(sb))
3784                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3785                              "redundant flags; please run fsck.");
3786
3787         /* Check for a known checksum algorithm */
3788         if (!ext4_verify_csum_type(sb, es)) {
3789                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3790                          "unknown checksum algorithm.");
3791                 silent = 1;
3792                 goto cantfind_ext4;
3793         }
3794
3795         /* Load the checksum driver */
3796         sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3797         if (IS_ERR(sbi->s_chksum_driver)) {
3798                 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3799                 ret = PTR_ERR(sbi->s_chksum_driver);
3800                 sbi->s_chksum_driver = NULL;
3801                 goto failed_mount;
3802         }
3803
3804         /* Check superblock checksum */
3805         if (!ext4_superblock_csum_verify(sb, es)) {
3806                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3807                          "invalid superblock checksum.  Run e2fsck?");
3808                 silent = 1;
3809                 ret = -EFSBADCRC;
3810                 goto cantfind_ext4;
3811         }
3812
3813         /* Precompute checksum seed for all metadata */
3814         if (ext4_has_feature_csum_seed(sb))
3815                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3816         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
3817                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3818                                                sizeof(es->s_uuid));
3819
3820         /* Set defaults before we parse the mount options */
3821         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3822         set_opt(sb, INIT_INODE_TABLE);
3823         if (def_mount_opts & EXT4_DEFM_DEBUG)
3824                 set_opt(sb, DEBUG);
3825         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3826                 set_opt(sb, GRPID);
3827         if (def_mount_opts & EXT4_DEFM_UID16)
3828                 set_opt(sb, NO_UID32);
3829         /* xattr user namespace & acls are now defaulted on */
3830         set_opt(sb, XATTR_USER);
3831 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3832         set_opt(sb, POSIX_ACL);
3833 #endif
3834         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3835         if (ext4_has_metadata_csum(sb))
3836                 set_opt(sb, JOURNAL_CHECKSUM);
3837
3838         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3839                 set_opt(sb, JOURNAL_DATA);
3840         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3841                 set_opt(sb, ORDERED_DATA);
3842         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3843                 set_opt(sb, WRITEBACK_DATA);
3844
3845         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3846                 set_opt(sb, ERRORS_PANIC);
3847         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3848                 set_opt(sb, ERRORS_CONT);
3849         else
3850                 set_opt(sb, ERRORS_RO);
3851         /* block_validity enabled by default; disable with noblock_validity */
3852         set_opt(sb, BLOCK_VALIDITY);
3853         if (def_mount_opts & EXT4_DEFM_DISCARD)
3854                 set_opt(sb, DISCARD);
3855
3856         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3857         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3858         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3859         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3860         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3861
3862         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3863                 set_opt(sb, BARRIER);
3864
3865         /*
3866          * enable delayed allocation by default
3867          * Use -o nodelalloc to turn it off
3868          */
3869         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3870             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3871                 set_opt(sb, DELALLOC);
3872
3873         /*
3874          * set default s_li_wait_mult for lazyinit, for the case there is
3875          * no mount option specified.
3876          */
3877         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3878
3879         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3880
3881         if (blocksize == PAGE_SIZE)
3882                 set_opt(sb, DIOREAD_NOLOCK);
3883
3884         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3885             blocksize > EXT4_MAX_BLOCK_SIZE) {
3886                 ext4_msg(sb, KERN_ERR,
3887                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3888                          blocksize, le32_to_cpu(es->s_log_block_size));
3889                 goto failed_mount;
3890         }
3891
3892         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3893                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3894                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3895         } else {
3896                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3897                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3898                 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) {
3899                         ext4_msg(sb, KERN_ERR, "invalid first ino: %u",
3900                                  sbi->s_first_ino);
3901                         goto failed_mount;
3902                 }
3903                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3904                     (!is_power_of_2(sbi->s_inode_size)) ||
3905                     (sbi->s_inode_size > blocksize)) {
3906                         ext4_msg(sb, KERN_ERR,
3907                                "unsupported inode size: %d",
3908                                sbi->s_inode_size);
3909                         ext4_msg(sb, KERN_ERR, "blocksize: %d", blocksize);
3910                         goto failed_mount;
3911                 }
3912                 /*
3913                  * i_atime_extra is the last extra field available for
3914                  * [acm]times in struct ext4_inode. Checking for that
3915                  * field should suffice to ensure we have extra space
3916                  * for all three.
3917                  */
3918                 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) +
3919                         sizeof(((struct ext4_inode *)0)->i_atime_extra)) {
3920                         sb->s_time_gran = 1;
3921                         sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX;
3922                 } else {
3923                         sb->s_time_gran = NSEC_PER_SEC;
3924                         sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX;
3925                 }
3926                 sb->s_time_min = EXT4_TIMESTAMP_MIN;
3927         }
3928         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3929                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3930                         EXT4_GOOD_OLD_INODE_SIZE;
3931                 if (ext4_has_feature_extra_isize(sb)) {
3932                         unsigned v, max = (sbi->s_inode_size -
3933                                            EXT4_GOOD_OLD_INODE_SIZE);
3934
3935                         v = le16_to_cpu(es->s_want_extra_isize);
3936                         if (v > max) {
3937                                 ext4_msg(sb, KERN_ERR,
3938                                          "bad s_want_extra_isize: %d", v);
3939                                 goto failed_mount;
3940                         }
3941                         if (sbi->s_want_extra_isize < v)
3942                                 sbi->s_want_extra_isize = v;
3943
3944                         v = le16_to_cpu(es->s_min_extra_isize);
3945                         if (v > max) {
3946                                 ext4_msg(sb, KERN_ERR,
3947                                          "bad s_min_extra_isize: %d", v);
3948                                 goto failed_mount;
3949                         }
3950                         if (sbi->s_want_extra_isize < v)
3951                                 sbi->s_want_extra_isize = v;
3952                 }
3953         }
3954
3955         if (sbi->s_es->s_mount_opts[0]) {
3956                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3957                                               sizeof(sbi->s_es->s_mount_opts),
3958                                               GFP_KERNEL);
3959                 if (!s_mount_opts)
3960                         goto failed_mount;
3961                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3962                                    &journal_ioprio, 0)) {
3963                         ext4_msg(sb, KERN_WARNING,
3964                                  "failed to parse options in superblock: %s",
3965                                  s_mount_opts);
3966                 }
3967                 kfree(s_mount_opts);
3968         }
3969         sbi->s_def_mount_opt = sbi->s_mount_opt;
3970         if (!parse_options((char *) data, sb, &journal_devnum,
3971                            &journal_ioprio, 0))
3972                 goto failed_mount;
3973
3974 #ifdef CONFIG_UNICODE
3975         if (ext4_has_feature_casefold(sb) && !sbi->s_encoding) {
3976                 const struct ext4_sb_encodings *encoding_info;
3977                 struct unicode_map *encoding;
3978                 __u16 encoding_flags;
3979
3980                 if (ext4_has_feature_encrypt(sb)) {
3981                         ext4_msg(sb, KERN_ERR,
3982                                  "Can't mount with encoding and encryption");
3983                         goto failed_mount;
3984                 }
3985
3986                 if (ext4_sb_read_encoding(es, &encoding_info,
3987                                           &encoding_flags)) {
3988                         ext4_msg(sb, KERN_ERR,
3989                                  "Encoding requested by superblock is unknown");
3990                         goto failed_mount;
3991                 }
3992
3993                 encoding = utf8_load(encoding_info->version);
3994                 if (IS_ERR(encoding)) {
3995                         ext4_msg(sb, KERN_ERR,
3996                                  "can't mount with superblock charset: %s-%s "
3997                                  "not supported by the kernel. flags: 0x%x.",
3998                                  encoding_info->name, encoding_info->version,
3999                                  encoding_flags);
4000                         goto failed_mount;
4001                 }
4002                 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: "
4003                          "%s-%s with flags 0x%hx", encoding_info->name,
4004                          encoding_info->version?:"\b", encoding_flags);
4005
4006                 sbi->s_encoding = encoding;
4007                 sbi->s_encoding_flags = encoding_flags;
4008         }
4009 #endif
4010
4011         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4012                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with data=journal disables delayed allocation, dioread_nolock, and O_DIRECT support!\n");
4013                 clear_opt(sb, DIOREAD_NOLOCK);
4014                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4015                         ext4_msg(sb, KERN_ERR, "can't mount with "
4016                                  "both data=journal and delalloc");
4017                         goto failed_mount;
4018                 }
4019                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4020                         ext4_msg(sb, KERN_ERR, "can't mount with "
4021                                  "both data=journal and dioread_nolock");
4022                         goto failed_mount;
4023                 }
4024                 if (test_opt(sb, DAX)) {
4025                         ext4_msg(sb, KERN_ERR, "can't mount with "
4026                                  "both data=journal and dax");
4027                         goto failed_mount;
4028                 }
4029                 if (ext4_has_feature_encrypt(sb)) {
4030                         ext4_msg(sb, KERN_WARNING,
4031                                  "encrypted files will use data=ordered "
4032                                  "instead of data journaling mode");
4033                 }
4034                 if (test_opt(sb, DELALLOC))
4035                         clear_opt(sb, DELALLOC);
4036         } else {
4037                 sb->s_iflags |= SB_I_CGROUPWB;
4038         }
4039
4040         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
4041                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
4042
4043         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
4044             (ext4_has_compat_features(sb) ||
4045              ext4_has_ro_compat_features(sb) ||
4046              ext4_has_incompat_features(sb)))
4047                 ext4_msg(sb, KERN_WARNING,
4048                        "feature flags set on rev 0 fs, "
4049                        "running e2fsck is recommended");
4050
4051         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
4052                 set_opt2(sb, HURD_COMPAT);
4053                 if (ext4_has_feature_64bit(sb)) {
4054                         ext4_msg(sb, KERN_ERR,
4055                                  "The Hurd can't support 64-bit file systems");
4056                         goto failed_mount;
4057                 }
4058
4059                 /*
4060                  * ea_inode feature uses l_i_version field which is not
4061                  * available in HURD_COMPAT mode.
4062                  */
4063                 if (ext4_has_feature_ea_inode(sb)) {
4064                         ext4_msg(sb, KERN_ERR,
4065                                  "ea_inode feature is not supported for Hurd");
4066                         goto failed_mount;
4067                 }
4068         }
4069
4070         if (IS_EXT2_SB(sb)) {
4071                 if (ext2_feature_set_ok(sb))
4072                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
4073                                  "using the ext4 subsystem");
4074                 else {
4075                         /*
4076                          * If we're probing be silent, if this looks like
4077                          * it's actually an ext[34] filesystem.
4078                          */
4079                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4080                                 goto failed_mount;
4081                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
4082                                  "to feature incompatibilities");
4083                         goto failed_mount;
4084                 }
4085         }
4086
4087         if (IS_EXT3_SB(sb)) {
4088                 if (ext3_feature_set_ok(sb))
4089                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
4090                                  "using the ext4 subsystem");
4091                 else {
4092                         /*
4093                          * If we're probing be silent, if this looks like
4094                          * it's actually an ext4 filesystem.
4095                          */
4096                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4097                                 goto failed_mount;
4098                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
4099                                  "to feature incompatibilities");
4100                         goto failed_mount;
4101                 }
4102         }
4103
4104         /*
4105          * Check feature flags regardless of the revision level, since we
4106          * previously didn't change the revision level when setting the flags,
4107          * so there is a chance incompat flags are set on a rev 0 filesystem.
4108          */
4109         if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
4110                 goto failed_mount;
4111
4112         if (le32_to_cpu(es->s_log_block_size) >
4113             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4114                 ext4_msg(sb, KERN_ERR,
4115                          "Invalid log block size: %u",
4116                          le32_to_cpu(es->s_log_block_size));
4117                 goto failed_mount;
4118         }
4119         if (le32_to_cpu(es->s_log_cluster_size) >
4120             (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4121                 ext4_msg(sb, KERN_ERR,
4122                          "Invalid log cluster size: %u",
4123                          le32_to_cpu(es->s_log_cluster_size));
4124                 goto failed_mount;
4125         }
4126
4127         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
4128                 ext4_msg(sb, KERN_ERR,
4129                          "Number of reserved GDT blocks insanely large: %d",
4130                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
4131                 goto failed_mount;
4132         }
4133
4134         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
4135                 if (ext4_has_feature_inline_data(sb)) {
4136                         ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
4137                                         " that may contain inline data");
4138                         goto failed_mount;
4139                 }
4140                 if (!bdev_dax_supported(sb->s_bdev, blocksize)) {
4141                         ext4_msg(sb, KERN_ERR,
4142                                 "DAX unsupported by block device.");
4143                         goto failed_mount;
4144                 }
4145         }
4146
4147         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
4148                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
4149                          es->s_encryption_level);
4150                 goto failed_mount;
4151         }
4152
4153         if (sb->s_blocksize != blocksize) {
4154                 /* Validate the filesystem blocksize */
4155                 if (!sb_set_blocksize(sb, blocksize)) {
4156                         ext4_msg(sb, KERN_ERR, "bad block size %d",
4157                                         blocksize);
4158                         goto failed_mount;
4159                 }
4160
4161                 brelse(bh);
4162                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
4163                 offset = do_div(logical_sb_block, blocksize);
4164                 bh = sb_bread_unmovable(sb, logical_sb_block);
4165                 if (!bh) {
4166                         ext4_msg(sb, KERN_ERR,
4167                                "Can't read superblock on 2nd try");
4168                         goto failed_mount;
4169                 }
4170                 es = (struct ext4_super_block *)(bh->b_data + offset);
4171                 sbi->s_es = es;
4172                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
4173                         ext4_msg(sb, KERN_ERR,
4174                                "Magic mismatch, very weird!");
4175                         goto failed_mount;
4176                 }
4177         }
4178
4179         has_huge_files = ext4_has_feature_huge_file(sb);
4180         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
4181                                                       has_huge_files);
4182         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
4183
4184         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
4185         if (ext4_has_feature_64bit(sb)) {
4186                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
4187                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
4188                     !is_power_of_2(sbi->s_desc_size)) {
4189                         ext4_msg(sb, KERN_ERR,
4190                                "unsupported descriptor size %lu",
4191                                sbi->s_desc_size);
4192                         goto failed_mount;
4193                 }
4194         } else
4195                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
4196
4197         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
4198         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
4199
4200         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
4201         if (sbi->s_inodes_per_block == 0)
4202                 goto cantfind_ext4;
4203         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
4204             sbi->s_inodes_per_group > blocksize * 8) {
4205                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
4206                          sbi->s_inodes_per_group);
4207                 goto failed_mount;
4208         }
4209         sbi->s_itb_per_group = sbi->s_inodes_per_group /
4210                                         sbi->s_inodes_per_block;
4211         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
4212         sbi->s_sbh = bh;
4213         sbi->s_mount_state = le16_to_cpu(es->s_state);
4214         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
4215         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
4216
4217         for (i = 0; i < 4; i++)
4218                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
4219         sbi->s_def_hash_version = es->s_def_hash_version;
4220         if (ext4_has_feature_dir_index(sb)) {
4221                 i = le32_to_cpu(es->s_flags);
4222                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
4223                         sbi->s_hash_unsigned = 3;
4224                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
4225 #ifdef __CHAR_UNSIGNED__
4226                         if (!sb_rdonly(sb))
4227                                 es->s_flags |=
4228                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
4229                         sbi->s_hash_unsigned = 3;
4230 #else
4231                         if (!sb_rdonly(sb))
4232                                 es->s_flags |=
4233                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
4234 #endif
4235                 }
4236         }
4237
4238         /* Handle clustersize */
4239         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
4240         has_bigalloc = ext4_has_feature_bigalloc(sb);
4241         if (has_bigalloc) {
4242                 if (clustersize < blocksize) {
4243                         ext4_msg(sb, KERN_ERR,
4244                                  "cluster size (%d) smaller than "
4245                                  "block size (%d)", clustersize, blocksize);
4246                         goto failed_mount;
4247                 }
4248                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
4249                         le32_to_cpu(es->s_log_block_size);
4250                 sbi->s_clusters_per_group =
4251                         le32_to_cpu(es->s_clusters_per_group);
4252                 if (sbi->s_clusters_per_group > blocksize * 8) {
4253                         ext4_msg(sb, KERN_ERR,
4254                                  "#clusters per group too big: %lu",
4255                                  sbi->s_clusters_per_group);
4256                         goto failed_mount;
4257                 }
4258                 if (sbi->s_blocks_per_group !=
4259                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
4260                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
4261                                  "clusters per group (%lu) inconsistent",
4262                                  sbi->s_blocks_per_group,
4263                                  sbi->s_clusters_per_group);
4264                         goto failed_mount;
4265                 }
4266         } else {
4267                 if (clustersize != blocksize) {
4268                         ext4_msg(sb, KERN_ERR,
4269                                  "fragment/cluster size (%d) != "
4270                                  "block size (%d)", clustersize, blocksize);
4271                         goto failed_mount;
4272                 }
4273                 if (sbi->s_blocks_per_group > blocksize * 8) {
4274                         ext4_msg(sb, KERN_ERR,
4275                                  "#blocks per group too big: %lu",
4276                                  sbi->s_blocks_per_group);
4277                         goto failed_mount;
4278                 }
4279                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
4280                 sbi->s_cluster_bits = 0;
4281         }
4282         sbi->s_cluster_ratio = clustersize / blocksize;
4283
4284         /* Do we have standard group size of clustersize * 8 blocks ? */
4285         if (sbi->s_blocks_per_group == clustersize << 3)
4286                 set_opt2(sb, STD_GROUP_SIZE);
4287
4288         /*
4289          * Test whether we have more sectors than will fit in sector_t,
4290          * and whether the max offset is addressable by the page cache.
4291          */
4292         err = generic_check_addressable(sb->s_blocksize_bits,
4293                                         ext4_blocks_count(es));
4294         if (err) {
4295                 ext4_msg(sb, KERN_ERR, "filesystem"
4296                          " too large to mount safely on this system");
4297                 goto failed_mount;
4298         }
4299
4300         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
4301                 goto cantfind_ext4;
4302
4303         /* check blocks count against device size */
4304         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
4305         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
4306                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
4307                        "exceeds size of device (%llu blocks)",
4308                        ext4_blocks_count(es), blocks_count);
4309                 goto failed_mount;
4310         }
4311
4312         /*
4313          * It makes no sense for the first data block to be beyond the end
4314          * of the filesystem.
4315          */
4316         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
4317                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4318                          "block %u is beyond end of filesystem (%llu)",
4319                          le32_to_cpu(es->s_first_data_block),
4320                          ext4_blocks_count(es));
4321                 goto failed_mount;
4322         }
4323         if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) &&
4324             (sbi->s_cluster_ratio == 1)) {
4325                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4326                          "block is 0 with a 1k block and cluster size");
4327                 goto failed_mount;
4328         }
4329
4330         blocks_count = (ext4_blocks_count(es) -
4331                         le32_to_cpu(es->s_first_data_block) +
4332                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
4333         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
4334         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
4335                 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu "
4336                        "(block count %llu, first data block %u, "
4337                        "blocks per group %lu)", blocks_count,
4338                        ext4_blocks_count(es),
4339                        le32_to_cpu(es->s_first_data_block),
4340                        EXT4_BLOCKS_PER_GROUP(sb));
4341                 goto failed_mount;
4342         }
4343         sbi->s_groups_count = blocks_count;
4344         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
4345                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
4346         if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) !=
4347             le32_to_cpu(es->s_inodes_count)) {
4348                 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu",
4349                          le32_to_cpu(es->s_inodes_count),
4350                          ((u64)sbi->s_groups_count * sbi->s_inodes_per_group));
4351                 ret = -EINVAL;
4352                 goto failed_mount;
4353         }
4354         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
4355                    EXT4_DESC_PER_BLOCK(sb);
4356         if (ext4_has_feature_meta_bg(sb)) {
4357                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
4358                         ext4_msg(sb, KERN_WARNING,
4359                                  "first meta block group too large: %u "
4360                                  "(group descriptor block count %u)",
4361                                  le32_to_cpu(es->s_first_meta_bg), db_count);
4362                         goto failed_mount;
4363                 }
4364         }
4365         rcu_assign_pointer(sbi->s_group_desc,
4366                            kvmalloc_array(db_count,
4367                                           sizeof(struct buffer_head *),
4368                                           GFP_KERNEL));
4369         if (sbi->s_group_desc == NULL) {
4370                 ext4_msg(sb, KERN_ERR, "not enough memory");
4371                 ret = -ENOMEM;
4372                 goto failed_mount;
4373         }
4374
4375         bgl_lock_init(sbi->s_blockgroup_lock);
4376
4377         /* Pre-read the descriptors into the buffer cache */
4378         for (i = 0; i < db_count; i++) {
4379                 block = descriptor_loc(sb, logical_sb_block, i);
4380                 sb_breadahead_unmovable(sb, block);
4381         }
4382
4383         for (i = 0; i < db_count; i++) {
4384                 struct buffer_head *bh;
4385
4386                 block = descriptor_loc(sb, logical_sb_block, i);
4387                 bh = sb_bread_unmovable(sb, block);
4388                 if (!bh) {
4389                         ext4_msg(sb, KERN_ERR,
4390                                "can't read group descriptor %d", i);
4391                         db_count = i;
4392                         goto failed_mount2;
4393                 }
4394                 rcu_read_lock();
4395                 rcu_dereference(sbi->s_group_desc)[i] = bh;
4396                 rcu_read_unlock();
4397         }
4398         sbi->s_gdb_count = db_count;
4399         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
4400                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
4401                 ret = -EFSCORRUPTED;
4402                 goto failed_mount2;
4403         }
4404
4405         timer_setup(&sbi->s_err_report, print_daily_error_info, 0);
4406
4407         /* Register extent status tree shrinker */
4408         if (ext4_es_register_shrinker(sbi))
4409                 goto failed_mount3;
4410
4411         sbi->s_stripe = ext4_get_stripe_size(sbi);
4412         sbi->s_extent_max_zeroout_kb = 32;
4413
4414         /*
4415          * set up enough so that it can read an inode
4416          */
4417         sb->s_op = &ext4_sops;
4418         sb->s_export_op = &ext4_export_ops;
4419         sb->s_xattr = ext4_xattr_handlers;
4420 #ifdef CONFIG_FS_ENCRYPTION
4421         sb->s_cop = &ext4_cryptops;
4422 #endif
4423 #ifdef CONFIG_FS_VERITY
4424         sb->s_vop = &ext4_verityops;
4425 #endif
4426 #ifdef CONFIG_QUOTA
4427         sb->dq_op = &ext4_quota_operations;
4428         if (ext4_has_feature_quota(sb))
4429                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4430         else
4431                 sb->s_qcop = &ext4_qctl_operations;
4432         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
4433 #endif
4434         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
4435
4436         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
4437         mutex_init(&sbi->s_orphan_lock);
4438
4439         sb->s_root = NULL;
4440
4441         needs_recovery = (es->s_last_orphan != 0 ||
4442                           ext4_has_feature_journal_needs_recovery(sb));
4443
4444         if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb))
4445                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
4446                         goto failed_mount3a;
4447
4448         /*
4449          * The first inode we look at is the journal inode.  Don't try
4450          * root first: it may be modified in the journal!
4451          */
4452         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
4453                 err = ext4_load_journal(sb, es, journal_devnum);
4454                 if (err)
4455                         goto failed_mount3a;
4456         } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
4457                    ext4_has_feature_journal_needs_recovery(sb)) {
4458                 ext4_msg(sb, KERN_ERR, "required journal recovery "
4459                        "suppressed and not mounted read-only");
4460                 goto failed_mount_wq;
4461         } else {
4462                 /* Nojournal mode, all journal mount options are illegal */
4463                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
4464                         ext4_msg(sb, KERN_ERR, "can't mount with "
4465                                  "journal_checksum, fs mounted w/o journal");
4466                         goto failed_mount_wq;
4467                 }
4468                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4469                         ext4_msg(sb, KERN_ERR, "can't mount with "
4470                                  "journal_async_commit, fs mounted w/o journal");
4471                         goto failed_mount_wq;
4472                 }
4473                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
4474                         ext4_msg(sb, KERN_ERR, "can't mount with "
4475                                  "commit=%lu, fs mounted w/o journal",
4476                                  sbi->s_commit_interval / HZ);
4477                         goto failed_mount_wq;
4478                 }
4479                 if (EXT4_MOUNT_DATA_FLAGS &
4480                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
4481                         ext4_msg(sb, KERN_ERR, "can't mount with "
4482                                  "data=, fs mounted w/o journal");
4483                         goto failed_mount_wq;
4484                 }
4485                 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM;
4486                 clear_opt(sb, JOURNAL_CHECKSUM);
4487                 clear_opt(sb, DATA_FLAGS);
4488                 sbi->s_journal = NULL;
4489                 needs_recovery = 0;
4490                 goto no_journal;
4491         }
4492
4493         if (ext4_has_feature_64bit(sb) &&
4494             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4495                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4496                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4497                 goto failed_mount_wq;
4498         }
4499
4500         if (!set_journal_csum_feature_set(sb)) {
4501                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4502                          "feature set");
4503                 goto failed_mount_wq;
4504         }
4505
4506         /* We have now updated the journal if required, so we can
4507          * validate the data journaling mode. */
4508         switch (test_opt(sb, DATA_FLAGS)) {
4509         case 0:
4510                 /* No mode set, assume a default based on the journal
4511                  * capabilities: ORDERED_DATA if the journal can
4512                  * cope, else JOURNAL_DATA
4513                  */
4514                 if (jbd2_journal_check_available_features
4515                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4516                         set_opt(sb, ORDERED_DATA);
4517                         sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
4518                 } else {
4519                         set_opt(sb, JOURNAL_DATA);
4520                         sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
4521                 }
4522                 break;
4523
4524         case EXT4_MOUNT_ORDERED_DATA:
4525         case EXT4_MOUNT_WRITEBACK_DATA:
4526                 if (!jbd2_journal_check_available_features
4527                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4528                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4529                                "requested data journaling mode");
4530                         goto failed_mount_wq;
4531                 }
4532         default:
4533                 break;
4534         }
4535
4536         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4537             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4538                 ext4_msg(sb, KERN_ERR, "can't mount with "
4539                         "journal_async_commit in data=ordered mode");
4540                 goto failed_mount_wq;
4541         }
4542
4543         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4544
4545         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4546
4547 no_journal:
4548         if (!test_opt(sb, NO_MBCACHE)) {
4549                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
4550                 if (!sbi->s_ea_block_cache) {
4551                         ext4_msg(sb, KERN_ERR,
4552                                  "Failed to create ea_block_cache");
4553                         goto failed_mount_wq;
4554                 }
4555
4556                 if (ext4_has_feature_ea_inode(sb)) {
4557                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
4558                         if (!sbi->s_ea_inode_cache) {
4559                                 ext4_msg(sb, KERN_ERR,
4560                                          "Failed to create ea_inode_cache");
4561                                 goto failed_mount_wq;
4562                         }
4563                 }
4564         }
4565
4566         if (ext4_has_feature_verity(sb) && blocksize != PAGE_SIZE) {
4567                 ext4_msg(sb, KERN_ERR, "Unsupported blocksize for fs-verity");
4568                 goto failed_mount_wq;
4569         }
4570
4571         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) &&
4572             !ext4_has_feature_encrypt(sb)) {
4573                 ext4_set_feature_encrypt(sb);
4574                 ext4_commit_super(sb, 1);
4575         }
4576
4577         /*
4578          * Get the # of file system overhead blocks from the
4579          * superblock if present.
4580          */
4581         if (es->s_overhead_clusters)
4582                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4583         else {
4584                 err = ext4_calculate_overhead(sb);
4585                 if (err)
4586                         goto failed_mount_wq;
4587         }
4588
4589         /*
4590          * The maximum number of concurrent works can be high and
4591          * concurrency isn't really necessary.  Limit it to 1.
4592          */
4593         EXT4_SB(sb)->rsv_conversion_wq =
4594                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4595         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4596                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4597                 ret = -ENOMEM;
4598                 goto failed_mount4;
4599         }
4600
4601         /*
4602          * The jbd2_journal_load will have done any necessary log recovery,
4603          * so we can safely mount the rest of the filesystem now.
4604          */
4605
4606         root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL);
4607         if (IS_ERR(root)) {
4608                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4609                 ret = PTR_ERR(root);
4610                 root = NULL;
4611                 goto failed_mount4;
4612         }
4613         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4614                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4615                 iput(root);
4616                 goto failed_mount4;
4617         }
4618
4619 #ifdef CONFIG_UNICODE
4620         if (sbi->s_encoding)
4621                 sb->s_d_op = &ext4_dentry_ops;
4622 #endif
4623
4624         sb->s_root = d_make_root(root);
4625         if (!sb->s_root) {
4626                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4627                 ret = -ENOMEM;
4628                 goto failed_mount4;
4629         }
4630
4631         ret = ext4_setup_super(sb, es, sb_rdonly(sb));
4632         if (ret == -EROFS) {
4633                 sb->s_flags |= SB_RDONLY;
4634                 ret = 0;
4635         } else if (ret)
4636                 goto failed_mount4a;
4637
4638         ext4_set_resv_clusters(sb);
4639
4640         err = ext4_setup_system_zone(sb);
4641         if (err) {
4642                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4643                          "zone (%d)", err);
4644                 goto failed_mount4a;
4645         }
4646
4647         ext4_ext_init(sb);
4648         err = ext4_mb_init(sb);
4649         if (err) {
4650                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4651                          err);
4652                 goto failed_mount5;
4653         }
4654
4655         block = ext4_count_free_clusters(sb);
4656         ext4_free_blocks_count_set(sbi->s_es,
4657                                    EXT4_C2B(sbi, block));
4658         ext4_superblock_csum_set(sb);
4659         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4660                                   GFP_KERNEL);
4661         if (!err) {
4662                 unsigned long freei = ext4_count_free_inodes(sb);
4663                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4664                 ext4_superblock_csum_set(sb);
4665                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4666                                           GFP_KERNEL);
4667         }
4668         if (!err)
4669                 err = percpu_counter_init(&sbi->s_dirs_counter,
4670                                           ext4_count_dirs(sb), GFP_KERNEL);
4671         if (!err)
4672                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4673                                           GFP_KERNEL);
4674         if (!err)
4675                 err = percpu_init_rwsem(&sbi->s_writepages_rwsem);
4676
4677         if (err) {
4678                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4679                 goto failed_mount6;
4680         }
4681
4682         if (ext4_has_feature_flex_bg(sb))
4683                 if (!ext4_fill_flex_info(sb)) {
4684                         ext4_msg(sb, KERN_ERR,
4685                                "unable to initialize "
4686                                "flex_bg meta info!");
4687                         goto failed_mount6;
4688                 }
4689
4690         err = ext4_register_li_request(sb, first_not_zeroed);
4691         if (err)
4692                 goto failed_mount6;
4693
4694         err = ext4_register_sysfs(sb);
4695         if (err)
4696                 goto failed_mount7;
4697
4698 #ifdef CONFIG_QUOTA
4699         /* Enable quota usage during mount. */
4700         if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
4701                 err = ext4_enable_quotas(sb);
4702                 if (err)
4703                         goto failed_mount8;
4704         }
4705 #endif  /* CONFIG_QUOTA */
4706
4707         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4708         ext4_orphan_cleanup(sb, es);
4709         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4710         if (needs_recovery) {
4711                 ext4_msg(sb, KERN_INFO, "recovery complete");
4712                 ext4_mark_recovery_complete(sb, es);
4713         }
4714         if (EXT4_SB(sb)->s_journal) {
4715                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4716                         descr = " journalled data mode";
4717                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4718                         descr = " ordered data mode";
4719                 else
4720                         descr = " writeback data mode";
4721         } else
4722                 descr = "out journal";
4723
4724         if (test_opt(sb, DISCARD)) {
4725                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4726                 if (!blk_queue_discard(q))
4727                         ext4_msg(sb, KERN_WARNING,
4728                                  "mounting with \"discard\" option, but "
4729                                  "the device does not support discard");
4730         }
4731
4732         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4733                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4734                          "Opts: %.*s%s%s", descr,
4735                          (int) sizeof(sbi->s_es->s_mount_opts),
4736                          sbi->s_es->s_mount_opts,
4737                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4738
4739         if (es->s_error_count)
4740                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4741
4742         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4743         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4744         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4745         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4746
4747         kfree(orig_data);
4748         return 0;
4749
4750 cantfind_ext4:
4751         if (!silent)
4752                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4753         goto failed_mount;
4754
4755 #ifdef CONFIG_QUOTA
4756 failed_mount8:
4757         ext4_unregister_sysfs(sb);
4758 #endif
4759 failed_mount7:
4760         ext4_unregister_li_request(sb);
4761 failed_mount6:
4762         ext4_mb_release(sb);
4763         rcu_read_lock();
4764         flex_groups = rcu_dereference(sbi->s_flex_groups);
4765         if (flex_groups) {
4766                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
4767                         kvfree(flex_groups[i]);
4768                 kvfree(flex_groups);
4769         }
4770         rcu_read_unlock();
4771         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4772         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4773         percpu_counter_destroy(&sbi->s_dirs_counter);
4774         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4775         percpu_free_rwsem(&sbi->s_writepages_rwsem);
4776 failed_mount5:
4777         ext4_ext_release(sb);
4778         ext4_release_system_zone(sb);
4779 failed_mount4a:
4780         dput(sb->s_root);
4781         sb->s_root = NULL;
4782 failed_mount4:
4783         ext4_msg(sb, KERN_ERR, "mount failed");
4784         if (EXT4_SB(sb)->rsv_conversion_wq)
4785                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4786 failed_mount_wq:
4787         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
4788         sbi->s_ea_inode_cache = NULL;
4789
4790         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
4791         sbi->s_ea_block_cache = NULL;
4792
4793         if (sbi->s_journal) {
4794                 jbd2_journal_destroy(sbi->s_journal);
4795                 sbi->s_journal = NULL;
4796         }
4797 failed_mount3a:
4798         ext4_es_unregister_shrinker(sbi);
4799 failed_mount3:
4800         del_timer_sync(&sbi->s_err_report);
4801         if (sbi->s_mmp_tsk)
4802                 kthread_stop(sbi->s_mmp_tsk);
4803 failed_mount2:
4804         rcu_read_lock();
4805         group_desc = rcu_dereference(sbi->s_group_desc);
4806         for (i = 0; i < db_count; i++)
4807                 brelse(group_desc[i]);
4808         kvfree(group_desc);
4809         rcu_read_unlock();
4810 failed_mount:
4811         if (sbi->s_chksum_driver)
4812                 crypto_free_shash(sbi->s_chksum_driver);
4813
4814 #ifdef CONFIG_UNICODE
4815         utf8_unload(sbi->s_encoding);
4816 #endif
4817
4818 #ifdef CONFIG_QUOTA
4819         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4820                 kfree(get_qf_name(sb, sbi, i));
4821 #endif
4822         fscrypt_free_dummy_context(&sbi->s_dummy_enc_ctx);
4823         ext4_blkdev_remove(sbi);
4824         brelse(bh);
4825 out_fail:
4826         sb->s_fs_info = NULL;
4827         kfree(sbi->s_blockgroup_lock);
4828 out_free_base:
4829         kfree(sbi);
4830         kfree(orig_data);
4831         fs_put_dax(dax_dev);
4832         return err ? err : ret;
4833 }
4834
4835 /*
4836  * Setup any per-fs journal parameters now.  We'll do this both on
4837  * initial mount, once the journal has been initialised but before we've
4838  * done any recovery; and again on any subsequent remount.
4839  */
4840 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4841 {
4842         struct ext4_sb_info *sbi = EXT4_SB(sb);
4843
4844         journal->j_commit_interval = sbi->s_commit_interval;
4845         journal->j_min_batch_time = sbi->s_min_batch_time;
4846         journal->j_max_batch_time = sbi->s_max_batch_time;
4847
4848         write_lock(&journal->j_state_lock);
4849         if (test_opt(sb, BARRIER))
4850                 journal->j_flags |= JBD2_BARRIER;
4851         else
4852                 journal->j_flags &= ~JBD2_BARRIER;
4853         if (test_opt(sb, DATA_ERR_ABORT))
4854                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4855         else
4856                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4857         write_unlock(&journal->j_state_lock);
4858 }
4859
4860 static struct inode *ext4_get_journal_inode(struct super_block *sb,
4861                                              unsigned int journal_inum)
4862 {
4863         struct inode *journal_inode;
4864
4865         /*
4866          * Test for the existence of a valid inode on disk.  Bad things
4867          * happen if we iget() an unused inode, as the subsequent iput()
4868          * will try to delete it.
4869          */
4870         journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL);
4871         if (IS_ERR(journal_inode)) {
4872                 ext4_msg(sb, KERN_ERR, "no journal found");
4873                 return NULL;
4874         }
4875         if (!journal_inode->i_nlink) {
4876                 make_bad_inode(journal_inode);
4877                 iput(journal_inode);
4878                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4879                 return NULL;
4880         }
4881
4882         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4883                   journal_inode, journal_inode->i_size);
4884         if (!S_ISREG(journal_inode->i_mode)) {
4885                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4886                 iput(journal_inode);
4887                 return NULL;
4888         }
4889         return journal_inode;
4890 }
4891
4892 static journal_t *ext4_get_journal(struct super_block *sb,
4893                                    unsigned int journal_inum)
4894 {
4895         struct inode *journal_inode;
4896         journal_t *journal;
4897
4898         BUG_ON(!ext4_has_feature_journal(sb));
4899
4900         journal_inode = ext4_get_journal_inode(sb, journal_inum);
4901         if (!journal_inode)
4902                 return NULL;
4903
4904         journal = jbd2_journal_init_inode(journal_inode);
4905         if (!journal) {
4906                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4907                 iput(journal_inode);
4908                 return NULL;
4909         }
4910         journal->j_private = sb;
4911         ext4_init_journal_params(sb, journal);
4912         return journal;
4913 }
4914
4915 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4916                                        dev_t j_dev)
4917 {
4918         struct buffer_head *bh;
4919         journal_t *journal;
4920         ext4_fsblk_t start;
4921         ext4_fsblk_t len;
4922         int hblock, blocksize;
4923         ext4_fsblk_t sb_block;
4924         unsigned long offset;
4925         struct ext4_super_block *es;
4926         struct block_device *bdev;
4927
4928         BUG_ON(!ext4_has_feature_journal(sb));
4929
4930         bdev = ext4_blkdev_get(j_dev, sb);
4931         if (bdev == NULL)
4932                 return NULL;
4933
4934         blocksize = sb->s_blocksize;
4935         hblock = bdev_logical_block_size(bdev);
4936         if (blocksize < hblock) {
4937                 ext4_msg(sb, KERN_ERR,
4938                         "blocksize too small for journal device");
4939                 goto out_bdev;
4940         }
4941
4942         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4943         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4944         set_blocksize(bdev, blocksize);
4945         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4946                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4947                        "external journal");
4948                 goto out_bdev;
4949         }
4950
4951         es = (struct ext4_super_block *) (bh->b_data + offset);
4952         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4953             !(le32_to_cpu(es->s_feature_incompat) &
4954               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4955                 ext4_msg(sb, KERN_ERR, "external journal has "
4956                                         "bad superblock");
4957                 brelse(bh);
4958                 goto out_bdev;
4959         }
4960
4961         if ((le32_to_cpu(es->s_feature_ro_compat) &
4962              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4963             es->s_checksum != ext4_superblock_csum(sb, es)) {
4964                 ext4_msg(sb, KERN_ERR, "external journal has "
4965                                        "corrupt superblock");
4966                 brelse(bh);
4967                 goto out_bdev;
4968         }
4969
4970         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4971                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4972                 brelse(bh);
4973                 goto out_bdev;
4974         }
4975
4976         len = ext4_blocks_count(es);
4977         start = sb_block + 1;
4978         brelse(bh);     /* we're done with the superblock */
4979
4980         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4981                                         start, len, blocksize);
4982         if (!journal) {
4983                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4984                 goto out_bdev;
4985         }
4986         journal->j_private = sb;
4987         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4988         wait_on_buffer(journal->j_sb_buffer);
4989         if (!buffer_uptodate(journal->j_sb_buffer)) {
4990                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4991                 goto out_journal;
4992         }
4993         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4994                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4995                                         "user (unsupported) - %d",
4996                         be32_to_cpu(journal->j_superblock->s_nr_users));
4997                 goto out_journal;
4998         }
4999         EXT4_SB(sb)->journal_bdev = bdev;
5000         ext4_init_journal_params(sb, journal);
5001         return journal;
5002
5003 out_journal:
5004         jbd2_journal_destroy(journal);
5005 out_bdev:
5006         ext4_blkdev_put(bdev);
5007         return NULL;
5008 }
5009
5010 static int ext4_load_journal(struct super_block *sb,
5011                              struct ext4_super_block *es,
5012                              unsigned long journal_devnum)
5013 {
5014         journal_t *journal;
5015         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
5016         dev_t journal_dev;
5017         int err = 0;
5018         int really_read_only;
5019
5020         BUG_ON(!ext4_has_feature_journal(sb));
5021
5022         if (journal_devnum &&
5023             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5024                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
5025                         "numbers have changed");
5026                 journal_dev = new_decode_dev(journal_devnum);
5027         } else
5028                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
5029
5030         really_read_only = bdev_read_only(sb->s_bdev);
5031
5032         /*
5033          * Are we loading a blank journal or performing recovery after a
5034          * crash?  For recovery, we need to check in advance whether we
5035          * can get read-write access to the device.
5036          */
5037         if (ext4_has_feature_journal_needs_recovery(sb)) {
5038                 if (sb_rdonly(sb)) {
5039                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
5040                                         "required on readonly filesystem");
5041                         if (really_read_only) {
5042                                 ext4_msg(sb, KERN_ERR, "write access "
5043                                         "unavailable, cannot proceed "
5044                                         "(try mounting with noload)");
5045                                 return -EROFS;
5046                         }
5047                         ext4_msg(sb, KERN_INFO, "write access will "
5048                                "be enabled during recovery");
5049                 }
5050         }
5051
5052         if (journal_inum && journal_dev) {
5053                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
5054                        "and inode journals!");
5055                 return -EINVAL;
5056         }
5057
5058         if (journal_inum) {
5059                 if (!(journal = ext4_get_journal(sb, journal_inum)))
5060                         return -EINVAL;
5061         } else {
5062                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
5063                         return -EINVAL;
5064         }
5065
5066         if (!(journal->j_flags & JBD2_BARRIER))
5067                 ext4_msg(sb, KERN_INFO, "barriers disabled");
5068
5069         if (!ext4_has_feature_journal_needs_recovery(sb))
5070                 err = jbd2_journal_wipe(journal, !really_read_only);
5071         if (!err) {
5072                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
5073                 if (save)
5074                         memcpy(save, ((char *) es) +
5075                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
5076                 err = jbd2_journal_load(journal);
5077                 if (save)
5078                         memcpy(((char *) es) + EXT4_S_ERR_START,
5079                                save, EXT4_S_ERR_LEN);
5080                 kfree(save);
5081         }
5082
5083         if (err) {
5084                 ext4_msg(sb, KERN_ERR, "error loading journal");
5085                 jbd2_journal_destroy(journal);
5086                 return err;
5087         }
5088
5089         EXT4_SB(sb)->s_journal = journal;
5090         ext4_clear_journal_err(sb, es);
5091
5092         if (!really_read_only && journal_devnum &&
5093             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5094                 es->s_journal_dev = cpu_to_le32(journal_devnum);
5095
5096                 /* Make sure we flush the recovery flag to disk. */
5097                 ext4_commit_super(sb, 1);
5098         }
5099
5100         return 0;
5101 }
5102
5103 static int ext4_commit_super(struct super_block *sb, int sync)
5104 {
5105         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
5106         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
5107         int error = 0;
5108
5109         if (!sbh || block_device_ejected(sb))
5110                 return error;
5111
5112         /*
5113          * The superblock bh should be mapped, but it might not be if the
5114          * device was hot-removed. Not much we can do but fail the I/O.
5115          */
5116         if (!buffer_mapped(sbh))
5117                 return error;
5118
5119         /*
5120          * If the file system is mounted read-only, don't update the
5121          * superblock write time.  This avoids updating the superblock
5122          * write time when we are mounting the root file system
5123          * read/only but we need to replay the journal; at that point,
5124          * for people who are east of GMT and who make their clock
5125          * tick in localtime for Windows bug-for-bug compatibility,
5126          * the clock is set in the future, and this will cause e2fsck
5127          * to complain and force a full file system check.
5128          */
5129         if (!(sb->s_flags & SB_RDONLY))
5130                 ext4_update_tstamp(es, s_wtime);
5131         if (sb->s_bdev->bd_part)
5132                 es->s_kbytes_written =
5133                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
5134                             ((part_stat_read(sb->s_bdev->bd_part,
5135                                              sectors[STAT_WRITE]) -
5136                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
5137         else
5138                 es->s_kbytes_written =
5139                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
5140         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
5141                 ext4_free_blocks_count_set(es,
5142                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
5143                                 &EXT4_SB(sb)->s_freeclusters_counter)));
5144         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
5145                 es->s_free_inodes_count =
5146                         cpu_to_le32(percpu_counter_sum_positive(
5147                                 &EXT4_SB(sb)->s_freeinodes_counter));
5148         BUFFER_TRACE(sbh, "marking dirty");
5149         ext4_superblock_csum_set(sb);
5150         if (sync)
5151                 lock_buffer(sbh);
5152         if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
5153                 /*
5154                  * Oh, dear.  A previous attempt to write the
5155                  * superblock failed.  This could happen because the
5156                  * USB device was yanked out.  Or it could happen to
5157                  * be a transient write error and maybe the block will
5158                  * be remapped.  Nothing we can do but to retry the
5159                  * write and hope for the best.
5160                  */
5161                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
5162                        "superblock detected");
5163                 clear_buffer_write_io_error(sbh);
5164                 set_buffer_uptodate(sbh);
5165         }
5166         mark_buffer_dirty(sbh);
5167         if (sync) {
5168                 unlock_buffer(sbh);
5169                 error = __sync_dirty_buffer(sbh,
5170                         REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
5171                 if (buffer_write_io_error(sbh)) {
5172                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
5173                                "superblock");
5174                         clear_buffer_write_io_error(sbh);
5175                         set_buffer_uptodate(sbh);
5176                 }
5177         }
5178         return error;
5179 }
5180
5181 /*
5182  * Have we just finished recovery?  If so, and if we are mounting (or
5183  * remounting) the filesystem readonly, then we will end up with a
5184  * consistent fs on disk.  Record that fact.
5185  */
5186 static void ext4_mark_recovery_complete(struct super_block *sb,
5187                                         struct ext4_super_block *es)
5188 {
5189         journal_t *journal = EXT4_SB(sb)->s_journal;
5190
5191         if (!ext4_has_feature_journal(sb)) {
5192                 BUG_ON(journal != NULL);
5193                 return;
5194         }
5195         jbd2_journal_lock_updates(journal);
5196         if (jbd2_journal_flush(journal) < 0)
5197                 goto out;
5198
5199         if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) {
5200                 ext4_clear_feature_journal_needs_recovery(sb);
5201                 ext4_commit_super(sb, 1);
5202         }
5203
5204 out:
5205         jbd2_journal_unlock_updates(journal);
5206 }
5207
5208 /*
5209  * If we are mounting (or read-write remounting) a filesystem whose journal
5210  * has recorded an error from a previous lifetime, move that error to the
5211  * main filesystem now.
5212  */
5213 static void ext4_clear_journal_err(struct super_block *sb,
5214                                    struct ext4_super_block *es)
5215 {
5216         journal_t *journal;
5217         int j_errno;
5218         const char *errstr;
5219
5220         BUG_ON(!ext4_has_feature_journal(sb));
5221
5222         journal = EXT4_SB(sb)->s_journal;
5223
5224         /*
5225          * Now check for any error status which may have been recorded in the
5226          * journal by a prior ext4_error() or ext4_abort()
5227          */
5228
5229         j_errno = jbd2_journal_errno(journal);
5230         if (j_errno) {
5231                 char nbuf[16];
5232
5233                 errstr = ext4_decode_error(sb, j_errno, nbuf);
5234                 ext4_warning(sb, "Filesystem error recorded "
5235                              "from previous mount: %s", errstr);
5236                 ext4_warning(sb, "Marking fs in need of filesystem check.");
5237
5238                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
5239                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
5240                 ext4_commit_super(sb, 1);
5241
5242                 jbd2_journal_clear_err(journal);
5243                 jbd2_journal_update_sb_errno(journal);
5244         }
5245 }
5246
5247 /*
5248  * Force the running and committing transactions to commit,
5249  * and wait on the commit.
5250  */
5251 int ext4_force_commit(struct super_block *sb)
5252 {
5253         journal_t *journal;
5254
5255         if (sb_rdonly(sb))
5256                 return 0;
5257
5258         journal = EXT4_SB(sb)->s_journal;
5259         return ext4_journal_force_commit(journal);
5260 }
5261
5262 static int ext4_sync_fs(struct super_block *sb, int wait)
5263 {
5264         int ret = 0;
5265         tid_t target;
5266         bool needs_barrier = false;
5267         struct ext4_sb_info *sbi = EXT4_SB(sb);
5268
5269         if (unlikely(ext4_forced_shutdown(sbi)))
5270                 return 0;
5271
5272         trace_ext4_sync_fs(sb, wait);
5273         flush_workqueue(sbi->rsv_conversion_wq);
5274         /*
5275          * Writeback quota in non-journalled quota case - journalled quota has
5276          * no dirty dquots
5277          */
5278         dquot_writeback_dquots(sb, -1);
5279         /*
5280          * Data writeback is possible w/o journal transaction, so barrier must
5281          * being sent at the end of the function. But we can skip it if
5282          * transaction_commit will do it for us.
5283          */
5284         if (sbi->s_journal) {
5285                 target = jbd2_get_latest_transaction(sbi->s_journal);
5286                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
5287                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
5288                         needs_barrier = true;
5289
5290                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
5291                         if (wait)
5292                                 ret = jbd2_log_wait_commit(sbi->s_journal,
5293                                                            target);
5294                 }
5295         } else if (wait && test_opt(sb, BARRIER))
5296                 needs_barrier = true;
5297         if (needs_barrier) {
5298                 int err;
5299                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
5300                 if (!ret)
5301                         ret = err;
5302         }
5303
5304         return ret;
5305 }
5306
5307 /*
5308  * LVM calls this function before a (read-only) snapshot is created.  This
5309  * gives us a chance to flush the journal completely and mark the fs clean.
5310  *
5311  * Note that only this function cannot bring a filesystem to be in a clean
5312  * state independently. It relies on upper layer to stop all data & metadata
5313  * modifications.
5314  */
5315 static int ext4_freeze(struct super_block *sb)
5316 {
5317         int error = 0;
5318         journal_t *journal;
5319
5320         if (sb_rdonly(sb))
5321                 return 0;
5322
5323         journal = EXT4_SB(sb)->s_journal;
5324
5325         if (journal) {
5326                 /* Now we set up the journal barrier. */
5327                 jbd2_journal_lock_updates(journal);
5328
5329                 /*
5330                  * Don't clear the needs_recovery flag if we failed to
5331                  * flush the journal.
5332                  */
5333                 error = jbd2_journal_flush(journal);
5334                 if (error < 0)
5335                         goto out;
5336
5337                 /* Journal blocked and flushed, clear needs_recovery flag. */
5338                 ext4_clear_feature_journal_needs_recovery(sb);
5339         }
5340
5341         error = ext4_commit_super(sb, 1);
5342 out:
5343         if (journal)
5344                 /* we rely on upper layer to stop further updates */
5345                 jbd2_journal_unlock_updates(journal);
5346         return error;
5347 }
5348
5349 /*
5350  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
5351  * flag here, even though the filesystem is not technically dirty yet.
5352  */
5353 static int ext4_unfreeze(struct super_block *sb)
5354 {
5355         if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
5356                 return 0;
5357
5358         if (EXT4_SB(sb)->s_journal) {
5359                 /* Reset the needs_recovery flag before the fs is unlocked. */
5360                 ext4_set_feature_journal_needs_recovery(sb);
5361         }
5362
5363         ext4_commit_super(sb, 1);
5364         return 0;
5365 }
5366
5367 /*
5368  * Structure to save mount options for ext4_remount's benefit
5369  */
5370 struct ext4_mount_options {
5371         unsigned long s_mount_opt;
5372         unsigned long s_mount_opt2;
5373         kuid_t s_resuid;
5374         kgid_t s_resgid;
5375         unsigned long s_commit_interval;
5376         u32 s_min_batch_time, s_max_batch_time;
5377 #ifdef CONFIG_QUOTA
5378         int s_jquota_fmt;
5379         char *s_qf_names[EXT4_MAXQUOTAS];
5380 #endif
5381 };
5382
5383 static int ext4_remount(struct super_block *sb, int *flags, char *data)
5384 {
5385         struct ext4_super_block *es;
5386         struct ext4_sb_info *sbi = EXT4_SB(sb);
5387         unsigned long old_sb_flags;
5388         struct ext4_mount_options old_opts;
5389         int enable_quota = 0;
5390         ext4_group_t g;
5391         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
5392         int err = 0;
5393 #ifdef CONFIG_QUOTA
5394         int i, j;
5395         char *to_free[EXT4_MAXQUOTAS];
5396 #endif
5397         char *orig_data = kstrdup(data, GFP_KERNEL);
5398
5399         if (data && !orig_data)
5400                 return -ENOMEM;
5401
5402         /* Store the original options */
5403         old_sb_flags = sb->s_flags;
5404         old_opts.s_mount_opt = sbi->s_mount_opt;
5405         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
5406         old_opts.s_resuid = sbi->s_resuid;
5407         old_opts.s_resgid = sbi->s_resgid;
5408         old_opts.s_commit_interval = sbi->s_commit_interval;
5409         old_opts.s_min_batch_time = sbi->s_min_batch_time;
5410         old_opts.s_max_batch_time = sbi->s_max_batch_time;
5411 #ifdef CONFIG_QUOTA
5412         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
5413         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5414                 if (sbi->s_qf_names[i]) {
5415                         char *qf_name = get_qf_name(sb, sbi, i);
5416
5417                         old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL);
5418                         if (!old_opts.s_qf_names[i]) {
5419                                 for (j = 0; j < i; j++)
5420                                         kfree(old_opts.s_qf_names[j]);
5421                                 kfree(orig_data);
5422                                 return -ENOMEM;
5423                         }
5424                 } else
5425                         old_opts.s_qf_names[i] = NULL;
5426 #endif
5427         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
5428                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
5429
5430         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
5431                 err = -EINVAL;
5432                 goto restore_opts;
5433         }
5434
5435         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
5436             test_opt(sb, JOURNAL_CHECKSUM)) {
5437                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
5438                          "during remount not supported; ignoring");
5439                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
5440         }
5441
5442         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
5443                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
5444                         ext4_msg(sb, KERN_ERR, "can't mount with "
5445                                  "both data=journal and delalloc");
5446                         err = -EINVAL;
5447                         goto restore_opts;
5448                 }
5449                 if (test_opt(sb, DIOREAD_NOLOCK)) {
5450                         ext4_msg(sb, KERN_ERR, "can't mount with "
5451                                  "both data=journal and dioread_nolock");
5452                         err = -EINVAL;
5453                         goto restore_opts;
5454                 }
5455                 if (test_opt(sb, DAX)) {
5456                         ext4_msg(sb, KERN_ERR, "can't mount with "
5457                                  "both data=journal and dax");
5458                         err = -EINVAL;
5459                         goto restore_opts;
5460                 }
5461         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
5462                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5463                         ext4_msg(sb, KERN_ERR, "can't mount with "
5464                                 "journal_async_commit in data=ordered mode");
5465                         err = -EINVAL;
5466                         goto restore_opts;
5467                 }
5468         }
5469
5470         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
5471                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
5472                 err = -EINVAL;
5473                 goto restore_opts;
5474         }
5475
5476         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
5477                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
5478                         "dax flag with busy inodes while remounting");
5479                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
5480         }
5481
5482         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
5483                 ext4_abort(sb, EXT4_ERR_ESHUTDOWN, "Abort forced by user");
5484
5485         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
5486                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
5487
5488         es = sbi->s_es;
5489
5490         if (sbi->s_journal) {
5491                 ext4_init_journal_params(sb, sbi->s_journal);
5492                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
5493         }
5494
5495         if (*flags & SB_LAZYTIME)
5496                 sb->s_flags |= SB_LAZYTIME;
5497
5498         if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) {
5499                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
5500                         err = -EROFS;
5501                         goto restore_opts;
5502                 }
5503
5504                 if (*flags & SB_RDONLY) {
5505                         err = sync_filesystem(sb);
5506                         if (err < 0)
5507                                 goto restore_opts;
5508                         err = dquot_suspend(sb, -1);
5509                         if (err < 0)
5510                                 goto restore_opts;
5511
5512                         /*
5513                          * First of all, the unconditional stuff we have to do
5514                          * to disable replay of the journal when we next remount
5515                          */
5516                         sb->s_flags |= SB_RDONLY;
5517
5518                         /*
5519                          * OK, test if we are remounting a valid rw partition
5520                          * readonly, and if so set the rdonly flag and then
5521                          * mark the partition as valid again.
5522                          */
5523                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5524                             (sbi->s_mount_state & EXT4_VALID_FS))
5525                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5526
5527                         if (sbi->s_journal)
5528                                 ext4_mark_recovery_complete(sb, es);
5529                         if (sbi->s_mmp_tsk)
5530                                 kthread_stop(sbi->s_mmp_tsk);
5531                 } else {
5532                         /* Make sure we can mount this feature set readwrite */
5533                         if (ext4_has_feature_readonly(sb) ||
5534                             !ext4_feature_set_ok(sb, 0)) {
5535                                 err = -EROFS;
5536                                 goto restore_opts;
5537                         }
5538                         /*
5539                          * Make sure the group descriptor checksums
5540                          * are sane.  If they aren't, refuse to remount r/w.
5541                          */
5542                         for (g = 0; g < sbi->s_groups_count; g++) {
5543                                 struct ext4_group_desc *gdp =
5544                                         ext4_get_group_desc(sb, g, NULL);
5545
5546                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5547                                         ext4_msg(sb, KERN_ERR,
5548                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5549                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
5550                                                le16_to_cpu(gdp->bg_checksum));
5551                                         err = -EFSBADCRC;
5552                                         goto restore_opts;
5553                                 }
5554                         }
5555
5556                         /*
5557                          * If we have an unprocessed orphan list hanging
5558                          * around from a previously readonly bdev mount,
5559                          * require a full umount/remount for now.
5560                          */
5561                         if (es->s_last_orphan) {
5562                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5563                                        "remount RDWR because of unprocessed "
5564                                        "orphan inode list.  Please "
5565                                        "umount/remount instead");
5566                                 err = -EINVAL;
5567                                 goto restore_opts;
5568                         }
5569
5570                         /*
5571                          * Mounting a RDONLY partition read-write, so reread
5572                          * and store the current valid flag.  (It may have
5573                          * been changed by e2fsck since we originally mounted
5574                          * the partition.)
5575                          */
5576                         if (sbi->s_journal)
5577                                 ext4_clear_journal_err(sb, es);
5578                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5579
5580                         err = ext4_setup_super(sb, es, 0);
5581                         if (err)
5582                                 goto restore_opts;
5583
5584                         sb->s_flags &= ~SB_RDONLY;
5585                         if (ext4_has_feature_mmp(sb))
5586                                 if (ext4_multi_mount_protect(sb,
5587                                                 le64_to_cpu(es->s_mmp_block))) {
5588                                         err = -EROFS;
5589                                         goto restore_opts;
5590                                 }
5591                         enable_quota = 1;
5592                 }
5593         }
5594
5595         /*
5596          * Reinitialize lazy itable initialization thread based on
5597          * current settings
5598          */
5599         if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
5600                 ext4_unregister_li_request(sb);
5601         else {
5602                 ext4_group_t first_not_zeroed;
5603                 first_not_zeroed = ext4_has_uninit_itable(sb);
5604                 ext4_register_li_request(sb, first_not_zeroed);
5605         }
5606
5607         ext4_setup_system_zone(sb);
5608         if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) {
5609                 err = ext4_commit_super(sb, 1);
5610                 if (err)
5611                         goto restore_opts;
5612         }
5613
5614 #ifdef CONFIG_QUOTA
5615         /* Release old quota file names */
5616         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5617                 kfree(old_opts.s_qf_names[i]);
5618         if (enable_quota) {
5619                 if (sb_any_quota_suspended(sb))
5620                         dquot_resume(sb, -1);
5621                 else if (ext4_has_feature_quota(sb)) {
5622                         err = ext4_enable_quotas(sb);
5623                         if (err)
5624                                 goto restore_opts;
5625                 }
5626         }
5627 #endif
5628
5629         *flags = (*flags & ~SB_LAZYTIME) | (sb->s_flags & SB_LAZYTIME);
5630         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5631         kfree(orig_data);
5632         return 0;
5633
5634 restore_opts:
5635         sb->s_flags = old_sb_flags;
5636         sbi->s_mount_opt = old_opts.s_mount_opt;
5637         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5638         sbi->s_resuid = old_opts.s_resuid;
5639         sbi->s_resgid = old_opts.s_resgid;
5640         sbi->s_commit_interval = old_opts.s_commit_interval;
5641         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5642         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5643 #ifdef CONFIG_QUOTA
5644         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5645         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5646                 to_free[i] = get_qf_name(sb, sbi, i);
5647                 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]);
5648         }
5649         synchronize_rcu();
5650         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5651                 kfree(to_free[i]);
5652 #endif
5653         kfree(orig_data);
5654         return err;
5655 }
5656
5657 #ifdef CONFIG_QUOTA
5658 static int ext4_statfs_project(struct super_block *sb,
5659                                kprojid_t projid, struct kstatfs *buf)
5660 {
5661         struct kqid qid;
5662         struct dquot *dquot;
5663         u64 limit;
5664         u64 curblock;
5665
5666         qid = make_kqid_projid(projid);
5667         dquot = dqget(sb, qid);
5668         if (IS_ERR(dquot))
5669                 return PTR_ERR(dquot);
5670         spin_lock(&dquot->dq_dqb_lock);
5671
5672         limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit,
5673                              dquot->dq_dqb.dqb_bhardlimit);
5674         limit >>= sb->s_blocksize_bits;
5675
5676         if (limit && buf->f_blocks > limit) {
5677                 curblock = (dquot->dq_dqb.dqb_curspace +
5678                             dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits;
5679                 buf->f_blocks = limit;
5680                 buf->f_bfree = buf->f_bavail =
5681                         (buf->f_blocks > curblock) ?
5682                          (buf->f_blocks - curblock) : 0;
5683         }
5684
5685         limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit,
5686                              dquot->dq_dqb.dqb_ihardlimit);
5687         if (limit && buf->f_files > limit) {
5688                 buf->f_files = limit;
5689                 buf->f_ffree =
5690                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
5691                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
5692         }
5693
5694         spin_unlock(&dquot->dq_dqb_lock);
5695         dqput(dquot);
5696         return 0;
5697 }
5698 #endif
5699
5700 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5701 {
5702         struct super_block *sb = dentry->d_sb;
5703         struct ext4_sb_info *sbi = EXT4_SB(sb);
5704         struct ext4_super_block *es = sbi->s_es;
5705         ext4_fsblk_t overhead = 0, resv_blocks;
5706         u64 fsid;
5707         s64 bfree;
5708         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5709
5710         if (!test_opt(sb, MINIX_DF))
5711                 overhead = sbi->s_overhead;
5712
5713         buf->f_type = EXT4_SUPER_MAGIC;
5714         buf->f_bsize = sb->s_blocksize;
5715         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5716         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5717                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5718         /* prevent underflow in case that few free space is available */
5719         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5720         buf->f_bavail = buf->f_bfree -
5721                         (ext4_r_blocks_count(es) + resv_blocks);
5722         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5723                 buf->f_bavail = 0;
5724         buf->f_files = le32_to_cpu(es->s_inodes_count);
5725         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5726         buf->f_namelen = EXT4_NAME_LEN;
5727         fsid = le64_to_cpup((void *)es->s_uuid) ^
5728                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5729         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5730         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5731
5732 #ifdef CONFIG_QUOTA
5733         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
5734             sb_has_quota_limits_enabled(sb, PRJQUOTA))
5735                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
5736 #endif
5737         return 0;
5738 }
5739
5740
5741 #ifdef CONFIG_QUOTA
5742
5743 /*
5744  * Helper functions so that transaction is started before we acquire dqio_sem
5745  * to keep correct lock ordering of transaction > dqio_sem
5746  */
5747 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5748 {
5749         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5750 }
5751
5752 static int ext4_write_dquot(struct dquot *dquot)
5753 {
5754         int ret, err;
5755         handle_t *handle;
5756         struct inode *inode;
5757
5758         inode = dquot_to_inode(dquot);
5759         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5760                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5761         if (IS_ERR(handle))
5762                 return PTR_ERR(handle);
5763         ret = dquot_commit(dquot);
5764         err = ext4_journal_stop(handle);
5765         if (!ret)
5766                 ret = err;
5767         return ret;
5768 }
5769
5770 static int ext4_acquire_dquot(struct dquot *dquot)
5771 {
5772         int ret, err;
5773         handle_t *handle;
5774
5775         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5776                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5777         if (IS_ERR(handle))
5778                 return PTR_ERR(handle);
5779         ret = dquot_acquire(dquot);
5780         err = ext4_journal_stop(handle);
5781         if (!ret)
5782                 ret = err;
5783         return ret;
5784 }
5785
5786 static int ext4_release_dquot(struct dquot *dquot)
5787 {
5788         int ret, err;
5789         handle_t *handle;
5790
5791         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5792                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5793         if (IS_ERR(handle)) {
5794                 /* Release dquot anyway to avoid endless cycle in dqput() */
5795                 dquot_release(dquot);
5796                 return PTR_ERR(handle);
5797         }
5798         ret = dquot_release(dquot);
5799         err = ext4_journal_stop(handle);
5800         if (!ret)
5801                 ret = err;
5802         return ret;
5803 }
5804
5805 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5806 {
5807         struct super_block *sb = dquot->dq_sb;
5808         struct ext4_sb_info *sbi = EXT4_SB(sb);
5809
5810         /* Are we journaling quotas? */
5811         if (ext4_has_feature_quota(sb) ||
5812             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5813                 dquot_mark_dquot_dirty(dquot);
5814                 return ext4_write_dquot(dquot);
5815         } else {
5816                 return dquot_mark_dquot_dirty(dquot);
5817         }
5818 }
5819
5820 static int ext4_write_info(struct super_block *sb, int type)
5821 {
5822         int ret, err;
5823         handle_t *handle;
5824
5825         /* Data block + inode block */
5826         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5827         if (IS_ERR(handle))
5828                 return PTR_ERR(handle);
5829         ret = dquot_commit_info(sb, type);
5830         err = ext4_journal_stop(handle);
5831         if (!ret)
5832                 ret = err;
5833         return ret;
5834 }
5835
5836 /*
5837  * Turn on quotas during mount time - we need to find
5838  * the quota file and such...
5839  */
5840 static int ext4_quota_on_mount(struct super_block *sb, int type)
5841 {
5842         return dquot_quota_on_mount(sb, get_qf_name(sb, EXT4_SB(sb), type),
5843                                         EXT4_SB(sb)->s_jquota_fmt, type);
5844 }
5845
5846 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5847 {
5848         struct ext4_inode_info *ei = EXT4_I(inode);
5849
5850         /* The first argument of lockdep_set_subclass has to be
5851          * *exactly* the same as the argument to init_rwsem() --- in
5852          * this case, in init_once() --- or lockdep gets unhappy
5853          * because the name of the lock is set using the
5854          * stringification of the argument to init_rwsem().
5855          */
5856         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5857         lockdep_set_subclass(&ei->i_data_sem, subclass);
5858 }
5859
5860 /*
5861  * Standard function to be called on quota_on
5862  */
5863 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5864                          const struct path *path)
5865 {
5866         int err;
5867
5868         if (!test_opt(sb, QUOTA))
5869                 return -EINVAL;
5870
5871         /* Quotafile not on the same filesystem? */
5872         if (path->dentry->d_sb != sb)
5873                 return -EXDEV;
5874         /* Journaling quota? */
5875         if (EXT4_SB(sb)->s_qf_names[type]) {
5876                 /* Quotafile not in fs root? */
5877                 if (path->dentry->d_parent != sb->s_root)
5878                         ext4_msg(sb, KERN_WARNING,
5879                                 "Quota file not on filesystem root. "
5880                                 "Journaled quota will not work");
5881                 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
5882         } else {
5883                 /*
5884                  * Clear the flag just in case mount options changed since
5885                  * last time.
5886                  */
5887                 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
5888         }
5889
5890         /*
5891          * When we journal data on quota file, we have to flush journal to see
5892          * all updates to the file when we bypass pagecache...
5893          */
5894         if (EXT4_SB(sb)->s_journal &&
5895             ext4_should_journal_data(d_inode(path->dentry))) {
5896                 /*
5897                  * We don't need to lock updates but journal_flush() could
5898                  * otherwise be livelocked...
5899                  */
5900                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5901                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5902                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5903                 if (err)
5904                         return err;
5905         }
5906
5907         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5908         err = dquot_quota_on(sb, type, format_id, path);
5909         if (err) {
5910                 lockdep_set_quota_inode(path->dentry->d_inode,
5911                                              I_DATA_SEM_NORMAL);
5912         } else {
5913                 struct inode *inode = d_inode(path->dentry);
5914                 handle_t *handle;
5915
5916                 /*
5917                  * Set inode flags to prevent userspace from messing with quota
5918                  * files. If this fails, we return success anyway since quotas
5919                  * are already enabled and this is not a hard failure.
5920                  */
5921                 inode_lock(inode);
5922                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5923                 if (IS_ERR(handle))
5924                         goto unlock_inode;
5925                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
5926                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
5927                                 S_NOATIME | S_IMMUTABLE);
5928                 ext4_mark_inode_dirty(handle, inode);
5929                 ext4_journal_stop(handle);
5930         unlock_inode:
5931                 inode_unlock(inode);
5932         }
5933         return err;
5934 }
5935
5936 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5937                              unsigned int flags)
5938 {
5939         int err;
5940         struct inode *qf_inode;
5941         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5942                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5943                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5944                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5945         };
5946
5947         BUG_ON(!ext4_has_feature_quota(sb));
5948
5949         if (!qf_inums[type])
5950                 return -EPERM;
5951
5952         qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL);
5953         if (IS_ERR(qf_inode)) {
5954                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5955                 return PTR_ERR(qf_inode);
5956         }
5957
5958         /* Don't account quota for quota files to avoid recursion */
5959         qf_inode->i_flags |= S_NOQUOTA;
5960         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
5961         err = dquot_load_quota_inode(qf_inode, type, format_id, flags);
5962         if (err)
5963                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
5964         iput(qf_inode);
5965
5966         return err;
5967 }
5968
5969 /* Enable usage tracking for all quota types. */
5970 static int ext4_enable_quotas(struct super_block *sb)
5971 {
5972         int type, err = 0;
5973         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5974                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5975                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5976                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5977         };
5978         bool quota_mopt[EXT4_MAXQUOTAS] = {
5979                 test_opt(sb, USRQUOTA),
5980                 test_opt(sb, GRPQUOTA),
5981                 test_opt(sb, PRJQUOTA),
5982         };
5983
5984         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
5985         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5986                 if (qf_inums[type]) {
5987                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5988                                 DQUOT_USAGE_ENABLED |
5989                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
5990                         if (err) {
5991                                 ext4_warning(sb,
5992                                         "Failed to enable quota tracking "
5993                                         "(type=%d, err=%d). Please run "
5994                                         "e2fsck to fix.", type, err);
5995                                 for (type--; type >= 0; type--)
5996                                         dquot_quota_off(sb, type);
5997
5998                                 return err;
5999                         }
6000                 }
6001         }
6002         return 0;
6003 }
6004
6005 static int ext4_quota_off(struct super_block *sb, int type)
6006 {
6007         struct inode *inode = sb_dqopt(sb)->files[type];
6008         handle_t *handle;
6009         int err;
6010
6011         /* Force all delayed allocation blocks to be allocated.
6012          * Caller already holds s_umount sem */
6013         if (test_opt(sb, DELALLOC))
6014                 sync_filesystem(sb);
6015
6016         if (!inode || !igrab(inode))
6017                 goto out;
6018
6019         err = dquot_quota_off(sb, type);
6020         if (err || ext4_has_feature_quota(sb))
6021                 goto out_put;
6022
6023         inode_lock(inode);
6024         /*
6025          * Update modification times of quota files when userspace can
6026          * start looking at them. If we fail, we return success anyway since
6027          * this is not a hard failure and quotas are already disabled.
6028          */
6029         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
6030         if (IS_ERR(handle))
6031                 goto out_unlock;
6032         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
6033         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
6034         inode->i_mtime = inode->i_ctime = current_time(inode);
6035         ext4_mark_inode_dirty(handle, inode);
6036         ext4_journal_stop(handle);
6037 out_unlock:
6038         inode_unlock(inode);
6039 out_put:
6040         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
6041         iput(inode);
6042         return err;
6043 out:
6044         return dquot_quota_off(sb, type);
6045 }
6046
6047 /* Read data from quotafile - avoid pagecache and such because we cannot afford
6048  * acquiring the locks... As quota files are never truncated and quota code
6049  * itself serializes the operations (and no one else should touch the files)
6050  * we don't have to be afraid of races */
6051 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
6052                                size_t len, loff_t off)
6053 {
6054         struct inode *inode = sb_dqopt(sb)->files[type];
6055         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
6056         int offset = off & (sb->s_blocksize - 1);
6057         int tocopy;
6058         size_t toread;
6059         struct buffer_head *bh;
6060         loff_t i_size = i_size_read(inode);
6061
6062         if (off > i_size)
6063                 return 0;
6064         if (off+len > i_size)
6065                 len = i_size-off;
6066         toread = len;
6067         while (toread > 0) {
6068                 tocopy = sb->s_blocksize - offset < toread ?
6069                                 sb->s_blocksize - offset : toread;
6070                 bh = ext4_bread(NULL, inode, blk, 0);
6071                 if (IS_ERR(bh))
6072                         return PTR_ERR(bh);
6073                 if (!bh)        /* A hole? */
6074                         memset(data, 0, tocopy);
6075                 else
6076                         memcpy(data, bh->b_data+offset, tocopy);
6077                 brelse(bh);
6078                 offset = 0;
6079                 toread -= tocopy;
6080                 data += tocopy;
6081                 blk++;
6082         }
6083         return len;
6084 }
6085
6086 /* Write to quotafile (we know the transaction is already started and has
6087  * enough credits) */
6088 static ssize_t ext4_quota_write(struct super_block *sb, int type,
6089                                 const char *data, size_t len, loff_t off)
6090 {
6091         struct inode *inode = sb_dqopt(sb)->files[type];
6092         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
6093         int err, offset = off & (sb->s_blocksize - 1);
6094         int retries = 0;
6095         struct buffer_head *bh;
6096         handle_t *handle = journal_current_handle();
6097
6098         if (EXT4_SB(sb)->s_journal && !handle) {
6099                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
6100                         " cancelled because transaction is not started",
6101                         (unsigned long long)off, (unsigned long long)len);
6102                 return -EIO;
6103         }
6104         /*
6105          * Since we account only one data block in transaction credits,
6106          * then it is impossible to cross a block boundary.
6107          */
6108         if (sb->s_blocksize - offset < len) {
6109                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
6110                         " cancelled because not block aligned",
6111                         (unsigned long long)off, (unsigned long long)len);
6112                 return -EIO;
6113         }
6114
6115         do {
6116                 bh = ext4_bread(handle, inode, blk,
6117                                 EXT4_GET_BLOCKS_CREATE |
6118                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
6119         } while (PTR_ERR(bh) == -ENOSPC &&
6120                  ext4_should_retry_alloc(inode->i_sb, &retries));
6121         if (IS_ERR(bh))
6122                 return PTR_ERR(bh);
6123         if (!bh)
6124                 goto out;
6125         BUFFER_TRACE(bh, "get write access");
6126         err = ext4_journal_get_write_access(handle, bh);
6127         if (err) {
6128                 brelse(bh);
6129                 return err;
6130         }
6131         lock_buffer(bh);
6132         memcpy(bh->b_data+offset, data, len);
6133         flush_dcache_page(bh->b_page);
6134         unlock_buffer(bh);
6135         err = ext4_handle_dirty_metadata(handle, NULL, bh);
6136         brelse(bh);
6137 out:
6138         if (inode->i_size < off + len) {
6139                 i_size_write(inode, off + len);
6140                 EXT4_I(inode)->i_disksize = inode->i_size;
6141                 ext4_mark_inode_dirty(handle, inode);
6142         }
6143         return len;
6144 }
6145 #endif
6146
6147 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
6148                        const char *dev_name, void *data)
6149 {
6150         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
6151 }
6152
6153 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
6154 static inline void register_as_ext2(void)
6155 {
6156         int err = register_filesystem(&ext2_fs_type);
6157         if (err)
6158                 printk(KERN_WARNING
6159                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
6160 }
6161
6162 static inline void unregister_as_ext2(void)
6163 {
6164         unregister_filesystem(&ext2_fs_type);
6165 }
6166
6167 static inline int ext2_feature_set_ok(struct super_block *sb)
6168 {
6169         if (ext4_has_unknown_ext2_incompat_features(sb))
6170                 return 0;
6171         if (sb_rdonly(sb))
6172                 return 1;
6173         if (ext4_has_unknown_ext2_ro_compat_features(sb))
6174                 return 0;
6175         return 1;
6176 }
6177 #else
6178 static inline void register_as_ext2(void) { }
6179 static inline void unregister_as_ext2(void) { }
6180 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
6181 #endif
6182
6183 static inline void register_as_ext3(void)
6184 {
6185         int err = register_filesystem(&ext3_fs_type);
6186         if (err)
6187                 printk(KERN_WARNING
6188                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
6189 }
6190
6191 static inline void unregister_as_ext3(void)
6192 {
6193         unregister_filesystem(&ext3_fs_type);
6194 }
6195
6196 static inline int ext3_feature_set_ok(struct super_block *sb)
6197 {
6198         if (ext4_has_unknown_ext3_incompat_features(sb))
6199                 return 0;
6200         if (!ext4_has_feature_journal(sb))
6201                 return 0;
6202         if (sb_rdonly(sb))
6203                 return 1;
6204         if (ext4_has_unknown_ext3_ro_compat_features(sb))
6205                 return 0;
6206         return 1;
6207 }
6208
6209 static struct file_system_type ext4_fs_type = {
6210         .owner          = THIS_MODULE,
6211         .name           = "ext4",
6212         .mount          = ext4_mount,
6213         .kill_sb        = kill_block_super,
6214         .fs_flags       = FS_REQUIRES_DEV,
6215 };
6216 MODULE_ALIAS_FS("ext4");
6217
6218 /* Shared across all ext4 file systems */
6219 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
6220
6221 static int __init ext4_init_fs(void)
6222 {
6223         int i, err;
6224
6225         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
6226         ext4_li_info = NULL;
6227         mutex_init(&ext4_li_mtx);
6228
6229         /* Build-time check for flags consistency */
6230         ext4_check_flag_values();
6231
6232         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
6233                 init_waitqueue_head(&ext4__ioend_wq[i]);
6234
6235         err = ext4_init_es();
6236         if (err)
6237                 return err;
6238
6239         err = ext4_init_pending();
6240         if (err)
6241                 goto out7;
6242
6243         err = ext4_init_post_read_processing();
6244         if (err)
6245                 goto out6;
6246
6247         err = ext4_init_pageio();
6248         if (err)
6249                 goto out5;
6250
6251         err = ext4_init_system_zone();
6252         if (err)
6253                 goto out4;
6254
6255         err = ext4_init_sysfs();
6256         if (err)
6257                 goto out3;
6258
6259         err = ext4_init_mballoc();
6260         if (err)
6261                 goto out2;
6262         err = init_inodecache();
6263         if (err)
6264                 goto out1;
6265         register_as_ext3();
6266         register_as_ext2();
6267         err = register_filesystem(&ext4_fs_type);
6268         if (err)
6269                 goto out;
6270
6271         return 0;
6272 out:
6273         unregister_as_ext2();
6274         unregister_as_ext3();
6275         destroy_inodecache();
6276 out1:
6277         ext4_exit_mballoc();
6278 out2:
6279         ext4_exit_sysfs();
6280 out3:
6281         ext4_exit_system_zone();
6282 out4:
6283         ext4_exit_pageio();
6284 out5:
6285         ext4_exit_post_read_processing();
6286 out6:
6287         ext4_exit_pending();
6288 out7:
6289         ext4_exit_es();
6290
6291         return err;
6292 }
6293
6294 static void __exit ext4_exit_fs(void)
6295 {
6296         ext4_destroy_lazyinit_thread();
6297         unregister_as_ext2();
6298         unregister_as_ext3();
6299         unregister_filesystem(&ext4_fs_type);
6300         destroy_inodecache();
6301         ext4_exit_mballoc();
6302         ext4_exit_sysfs();
6303         ext4_exit_system_zone();
6304         ext4_exit_pageio();
6305         ext4_exit_post_read_processing();
6306         ext4_exit_es();
6307         ext4_exit_pending();
6308 }
6309
6310 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
6311 MODULE_DESCRIPTION("Fourth Extended Filesystem");
6312 MODULE_LICENSE("GPL");
6313 MODULE_SOFTDEP("pre: crc32c");
6314 module_init(ext4_init_fs)
6315 module_exit(ext4_exit_fs)