fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/slab.h>
  25 #include <linux/init.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/ctype.h>
  38 #include <linux/log2.h>
  39 #include <linux/crc16.h>
  40 #include <linux/dax.h>
  41 #include <linux/cleancache.h>
  42 #include <linux/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"       /* Needed for trace points definition */
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53 #include "fsmap.h"
  54
  55 #define CREATE_TRACE_POINTS
  56 #include <trace/events/ext4.h>
  57
  58 static struct ext4_lazy_init *ext4_li_info;
  59 static struct mutex ext4_li_mtx;
  60 static struct ratelimit_state ext4_mount_msg_ratelimit;
  61
  62 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  63                              unsigned long journal_devnum);
  64 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  65 static int ext4_commit_super(struct super_block *sb, int sync);
  66 static int ext4_mark_recovery_complete(struct super_block *sb,
  67                                         struct ext4_super_block *es);
  68 static int ext4_clear_journal_err(struct super_block *sb,
  69                                   struct ext4_super_block *es);
  70 static int ext4_sync_fs(struct super_block *sb, int wait);
  71 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  72 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  73 static int ext4_unfreeze(struct super_block *sb);
  74 static int ext4_freeze(struct super_block *sb);
  75 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  76                        const char *dev_name, void *data);
  77 static inline int ext2_feature_set_ok(struct super_block *sb);
  78 static inline int ext3_feature_set_ok(struct super_block *sb);
  79 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  80 static void ext4_destroy_lazyinit_thread(void);
  81 static void ext4_unregister_li_request(struct super_block *sb);
  82 static void ext4_clear_request_list(void);
  83 static struct inode *ext4_get_journal_inode(struct super_block *sb,
  84                                             unsigned int journal_inum);
  85
  86 /*
  87  * Lock ordering
  88  *
  89  * Note the difference between i_mmap_sem (EXT4_I(inode)->i_mmap_sem) and
  90  * i_mmap_rwsem (inode->i_mmap_rwsem)!
  91  *
  92  * page fault path:
  93  * mmap_sem -> sb_start_pagefault -> i_mmap_sem (r) -> transaction start ->
  94  *   page lock -> i_data_sem (rw)
  95  *
  96  * buffered write path:
  97  * sb_start_write -> i_mutex -> mmap_sem
  98  * sb_start_write -> i_mutex -> transaction start -> page lock ->
  99  *   i_data_sem (rw)
 100  *
 101  * truncate:
 102  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (w) -> i_mmap_sem (w) ->
 103  *   i_mmap_rwsem (w) -> page lock
 104  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (w) -> i_mmap_sem (w) ->
 105  *   transaction start -> i_data_sem (rw)
 106  *
 107  * direct IO:
 108  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (r) -> mmap_sem
 109  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (r) ->
 110  *   transaction start -> i_data_sem (rw)
 111  *
 112  * writepages:
 113  * transaction start -> page lock(s) -> i_data_sem (rw)
 114  */
 115
 116 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
 117 static struct file_system_type ext2_fs_type = {
 118         .owner          = THIS_MODULE,
 119         .name           = "ext2",
 120         .mount          = ext4_mount,
 121         .kill_sb        = kill_block_super,
 122         .fs_flags       = FS_REQUIRES_DEV,
 123 };
 124 MODULE_ALIAS_FS("ext2");
 125 MODULE_ALIAS("ext2");
 126 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
 127 #else
 128 #define IS_EXT2_SB(sb) (0)
 129 #endif
 130
 131
 132 static struct file_system_type ext3_fs_type = {
 133         .owner          = THIS_MODULE,
 134         .name           = "ext3",
 135         .mount          = ext4_mount,
 136         .kill_sb        = kill_block_super,
 137         .fs_flags       = FS_REQUIRES_DEV,
 138 };
 139 MODULE_ALIAS_FS("ext3");
 140 MODULE_ALIAS("ext3");
 141 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 142
 143 /*
 144  * This works like sb_bread() except it uses ERR_PTR for error
 145  * returns.  Currently with sb_bread it's impossible to distinguish
 146  * between ENOMEM and EIO situations (since both result in a NULL
 147  * return.
 148  */
 149 struct buffer_head *
 150 ext4_sb_bread(struct super_block *sb, sector_t block, int op_flags)
 151 {
 152         struct buffer_head *bh = sb_getblk(sb, block);
 153
 154         if (bh == NULL)
 155                 return ERR_PTR(-ENOMEM);
 156         if (buffer_uptodate(bh))
 157                 return bh;
 158         ll_rw_block(REQ_OP_READ, REQ_META | op_flags, 1, &bh);
 159         wait_on_buffer(bh);
 160         if (buffer_uptodate(bh))
 161                 return bh;
 162         put_bh(bh);
 163         return ERR_PTR(-EIO);
 164 }
 165
 166 static int ext4_verify_csum_type(struct super_block *sb,
 167                                  struct ext4_super_block *es)
 168 {
 169         if (!ext4_has_feature_metadata_csum(sb))
 170                 return 1;
 171
 172         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 173 }
 174
 175 static __le32 ext4_superblock_csum(struct super_block *sb,
 176                                    struct ext4_super_block *es)
 177 {
 178         struct ext4_sb_info *sbi = EXT4_SB(sb);
 179         int offset = offsetof(struct ext4_super_block, s_checksum);
 180         __u32 csum;
 181
 182         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 183
 184         return cpu_to_le32(csum);
 185 }
 186
 187 static int ext4_superblock_csum_verify(struct super_block *sb,
 188                                        struct ext4_super_block *es)
 189 {
 190         if (!ext4_has_metadata_csum(sb))
 191                 return 1;
 192
 193         return es->s_checksum == ext4_superblock_csum(sb, es);
 194 }
 195
 196 void ext4_superblock_csum_set(struct super_block *sb)
 197 {
 198         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 199
 200         if (!ext4_has_metadata_csum(sb))
 201                 return;
 202
 203         es->s_checksum = ext4_superblock_csum(sb, es);
 204 }
 205
 206 void *ext4_kvmalloc(size_t size, gfp_t flags)
 207 {
 208         void *ret;
 209
 210         ret = kmalloc(size, flags | __GFP_NOWARN);
 211         if (!ret)
 212                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 213         return ret;
 214 }
 215
 216 void *ext4_kvzalloc(size_t size, gfp_t flags)
 217 {
 218         void *ret;
 219
 220         ret = kzalloc(size, flags | __GFP_NOWARN);
 221         if (!ret)
 222                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 223         return ret;
 224 }
 225
 226 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 227                                struct ext4_group_desc *bg)
 228 {
 229         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 230                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 231                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 232 }
 233
 234 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 235                                struct ext4_group_desc *bg)
 236 {
 237         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 238                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 239                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 240 }
 241
 242 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 243                               struct ext4_group_desc *bg)
 244 {
 245         return le32_to_cpu(bg->bg_inode_table_lo) |
 246                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 247                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 248 }
 249
 250 __u32 ext4_free_group_clusters(struct super_block *sb,
 251                                struct ext4_group_desc *bg)
 252 {
 253         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 254                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 255                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 256 }
 257
 258 __u32 ext4_free_inodes_count(struct super_block *sb,
 259                               struct ext4_group_desc *bg)
 260 {
 261         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 262                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 263                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 264 }
 265
 266 __u32 ext4_used_dirs_count(struct super_block *sb,
 267                               struct ext4_group_desc *bg)
 268 {
 269         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 270                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 271                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 272 }
 273
 274 __u32 ext4_itable_unused_count(struct super_block *sb,
 275                               struct ext4_group_desc *bg)
 276 {
 277         return le16_to_cpu(bg->bg_itable_unused_lo) |
 278                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 279                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 280 }
 281
 282 void ext4_block_bitmap_set(struct super_block *sb,
 283                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 284 {
 285         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 286         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 287                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 288 }
 289
 290 void ext4_inode_bitmap_set(struct super_block *sb,
 291                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 292 {
 293         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 294         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 295                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 296 }
 297
 298 void ext4_inode_table_set(struct super_block *sb,
 299                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 300 {
 301         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 302         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 303                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 304 }
 305
 306 void ext4_free_group_clusters_set(struct super_block *sb,
 307                                   struct ext4_group_desc *bg, __u32 count)
 308 {
 309         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 310         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 311                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 312 }
 313
 314 void ext4_free_inodes_set(struct super_block *sb,
 315                           struct ext4_group_desc *bg, __u32 count)
 316 {
 317         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 318         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 319                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 320 }
 321
 322 void ext4_used_dirs_set(struct super_block *sb,
 323                           struct ext4_group_desc *bg, __u32 count)
 324 {
 325         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 326         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 327                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 328 }
 329
 330 void ext4_itable_unused_set(struct super_block *sb,
 331                           struct ext4_group_desc *bg, __u32 count)
 332 {
 333         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 334         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 335                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 336 }
 337
 338
 339 static void __save_error_info(struct super_block *sb, const char *func,
 340                             unsigned int line)
 341 {
 342         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 343
 344         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 345         if (bdev_read_only(sb->s_bdev))
 346                 return;
 347         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 348         es->s_last_error_time = cpu_to_le32(get_seconds());
 349         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 350         es->s_last_error_line = cpu_to_le32(line);
 351         if (!es->s_first_error_time) {
 352                 es->s_first_error_time = es->s_last_error_time;
 353                 strncpy(es->s_first_error_func, func,
 354                         sizeof(es->s_first_error_func));
 355                 es->s_first_error_line = cpu_to_le32(line);
 356                 es->s_first_error_ino = es->s_last_error_ino;
 357                 es->s_first_error_block = es->s_last_error_block;
 358         }
 359         /*
 360          * Start the daily error reporting function if it hasn't been
 361          * started already
 362          */
 363         if (!es->s_error_count)
 364                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 365         le32_add_cpu(&es->s_error_count, 1);
 366 }
 367
 368 static void save_error_info(struct super_block *sb, const char *func,
 369                             unsigned int line)
 370 {
 371         __save_error_info(sb, func, line);
 372         if (!bdev_read_only(sb->s_bdev))
 373                 ext4_commit_super(sb, 1);
 374 }
 375
 376 /*
 377  * The del_gendisk() function uninitializes the disk-specific data
 378  * structures, including the bdi structure, without telling anyone
 379  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 380  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 381  * This is a kludge to prevent these oops until we can put in a proper
 382  * hook in del_gendisk() to inform the VFS and file system layers.
 383  */
 384 static int block_device_ejected(struct super_block *sb)
 385 {
 386         struct inode *bd_inode = sb->s_bdev->bd_inode;
 387         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 388
 389         return bdi->dev == NULL;
 390 }
 391
 392 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 393 {
 394         struct super_block              *sb = journal->j_private;
 395         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 396         int                             error = is_journal_aborted(journal);
 397         struct ext4_journal_cb_entry    *jce;
 398
 399         BUG_ON(txn->t_state == T_FINISHED);
 400
 401         ext4_process_freed_data(sb, txn->t_tid);
 402
 403         spin_lock(&sbi->s_md_lock);
 404         while (!list_empty(&txn->t_private_list)) {
 405                 jce = list_entry(txn->t_private_list.next,
 406                                  struct ext4_journal_cb_entry, jce_list);
 407                 list_del_init(&jce->jce_list);
 408                 spin_unlock(&sbi->s_md_lock);
 409                 jce->jce_func(sb, jce, error);
 410                 spin_lock(&sbi->s_md_lock);
 411         }
 412         spin_unlock(&sbi->s_md_lock);
 413 }
 414
 415 /* Deal with the reporting of failure conditions on a filesystem such as
 416  * inconsistencies detected or read IO failures.
 417  *
 418  * On ext2, we can store the error state of the filesystem in the
 419  * superblock.  That is not possible on ext4, because we may have other
 420  * write ordering constraints on the superblock which prevent us from
 421  * writing it out straight away; and given that the journal is about to
 422  * be aborted, we can't rely on the current, or future, transactions to
 423  * write out the superblock safely.
 424  *
 425  * We'll just use the jbd2_journal_abort() error code to record an error in
 426  * the journal instead.  On recovery, the journal will complain about
 427  * that error until we've noted it down and cleared it.
 428  */
 429
 430 static void ext4_handle_error(struct super_block *sb)
 431 {
 432         if (sb_rdonly(sb))
 433                 return;
 434
 435         if (!test_opt(sb, ERRORS_CONT)) {
 436                 journal_t *journal = EXT4_SB(sb)->s_journal;
 437
 438                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 439                 if (journal)
 440                         jbd2_journal_abort(journal, -EIO);
 441         }
 442         if (test_opt(sb, ERRORS_RO)) {
 443                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 444                 /*
 445                  * Make sure updated value of ->s_mount_flags will be visible
 446                  * before ->s_flags update
 447                  */
 448                 smp_wmb();
 449                 sb->s_flags |= MS_RDONLY;
 450         }
 451         if (test_opt(sb, ERRORS_PANIC)) {
 452                 if (EXT4_SB(sb)->s_journal &&
 453                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 454                         return;
 455                 panic("EXT4-fs (device %s): panic forced after error\n",
 456                         sb->s_id);
 457         }
 458 }
 459
 460 #define ext4_error_ratelimit(sb)                                        \
 461                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 462                              "EXT4-fs error")
 463
 464 void __ext4_error(struct super_block *sb, const char *function,
 465                   unsigned int line, const char *fmt, ...)
 466 {
 467         struct va_format vaf;
 468         va_list args;
 469
 470         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 471                 return;
 472
 473         if (ext4_error_ratelimit(sb)) {
 474                 va_start(args, fmt);
 475                 vaf.fmt = fmt;
 476                 vaf.va = &args;
 477                 printk(KERN_CRIT
 478                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 479                        sb->s_id, function, line, current->comm, &vaf);
 480                 va_end(args);
 481         }
 482         save_error_info(sb, function, line);
 483         ext4_handle_error(sb);
 484 }
 485
 486 void __ext4_error_inode(struct inode *inode, const char *function,
 487                         unsigned int line, ext4_fsblk_t block,
 488                         const char *fmt, ...)
 489 {
 490         va_list args;
 491         struct va_format vaf;
 492         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 493
 494         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 495                 return;
 496
 497         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 498         es->s_last_error_block = cpu_to_le64(block);
 499         if (ext4_error_ratelimit(inode->i_sb)) {
 500                 va_start(args, fmt);
 501                 vaf.fmt = fmt;
 502                 vaf.va = &args;
 503                 if (block)
 504                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 505                                "inode #%lu: block %llu: comm %s: %pV\n",
 506                                inode->i_sb->s_id, function, line, inode->i_ino,
 507                                block, current->comm, &vaf);
 508                 else
 509                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 510                                "inode #%lu: comm %s: %pV\n",
 511                                inode->i_sb->s_id, function, line, inode->i_ino,
 512                                current->comm, &vaf);
 513                 va_end(args);
 514         }
 515         save_error_info(inode->i_sb, function, line);
 516         ext4_handle_error(inode->i_sb);
 517 }
 518
 519 void __ext4_error_file(struct file *file, const char *function,
 520                        unsigned int line, ext4_fsblk_t block,
 521                        const char *fmt, ...)
 522 {
 523         va_list args;
 524         struct va_format vaf;
 525         struct ext4_super_block *es;
 526         struct inode *inode = file_inode(file);
 527         char pathname[80], *path;
 528
 529         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 530                 return;
 531
 532         es = EXT4_SB(inode->i_sb)->s_es;
 533         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 534         if (ext4_error_ratelimit(inode->i_sb)) {
 535                 path = file_path(file, pathname, sizeof(pathname));
 536                 if (IS_ERR(path))
 537                         path = "(unknown)";
 538                 va_start(args, fmt);
 539                 vaf.fmt = fmt;
 540                 vaf.va = &args;
 541                 if (block)
 542                         printk(KERN_CRIT
 543                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 544                                "block %llu: comm %s: path %s: %pV\n",
 545                                inode->i_sb->s_id, function, line, inode->i_ino,
 546                                block, current->comm, path, &vaf);
 547                 else
 548                         printk(KERN_CRIT
 549                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 550                                "comm %s: path %s: %pV\n",
 551                                inode->i_sb->s_id, function, line, inode->i_ino,
 552                                current->comm, path, &vaf);
 553                 va_end(args);
 554         }
 555         save_error_info(inode->i_sb, function, line);
 556         ext4_handle_error(inode->i_sb);
 557 }
 558
 559 const char *ext4_decode_error(struct super_block *sb, int errno,
 560                               char nbuf[16])
 561 {
 562         char *errstr = NULL;
 563
 564         switch (errno) {
 565         case -EFSCORRUPTED:
 566                 errstr = "Corrupt filesystem";
 567                 break;
 568         case -EFSBADCRC:
 569                 errstr = "Filesystem failed CRC";
 570                 break;
 571         case -EIO:
 572                 errstr = "IO failure";
 573                 break;
 574         case -ENOMEM:
 575                 errstr = "Out of memory";
 576                 break;
 577         case -EROFS:
 578                 if (!sb || (EXT4_SB(sb)->s_journal &&
 579                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 580                         errstr = "Journal has aborted";
 581                 else
 582                         errstr = "Readonly filesystem";
 583                 break;
 584         default:
 585                 /* If the caller passed in an extra buffer for unknown
 586                  * errors, textualise them now.  Else we just return
 587                  * NULL. */
 588                 if (nbuf) {
 589                         /* Check for truncated error codes... */
 590                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 591                                 errstr = nbuf;
 592                 }
 593                 break;
 594         }
 595
 596         return errstr;
 597 }
 598
 599 /* __ext4_std_error decodes expected errors from journaling functions
 600  * automatically and invokes the appropriate error response.  */
 601
 602 void __ext4_std_error(struct super_block *sb, const char *function,
 603                       unsigned int line, int errno)
 604 {
 605         char nbuf[16];
 606         const char *errstr;
 607
 608         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 609                 return;
 610
 611         /* Special case: if the error is EROFS, and we're not already
 612          * inside a transaction, then there's really no point in logging
 613          * an error. */
 614         if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb))
 615                 return;
 616
 617         if (ext4_error_ratelimit(sb)) {
 618                 errstr = ext4_decode_error(sb, errno, nbuf);
 619                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 620                        sb->s_id, function, line, errstr);
 621         }
 622
 623         save_error_info(sb, function, line);
 624         ext4_handle_error(sb);
 625 }
 626
 627 /*
 628  * ext4_abort is a much stronger failure handler than ext4_error.  The
 629  * abort function may be used to deal with unrecoverable failures such
 630  * as journal IO errors or ENOMEM at a critical moment in log management.
 631  *
 632  * We unconditionally force the filesystem into an ABORT|READONLY state,
 633  * unless the error response on the fs has been set to panic in which
 634  * case we take the easy way out and panic immediately.
 635  */
 636
 637 void __ext4_abort(struct super_block *sb, const char *function,
 638                 unsigned int line, const char *fmt, ...)
 639 {
 640         struct va_format vaf;
 641         va_list args;
 642
 643         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 644                 return;
 645
 646         save_error_info(sb, function, line);
 647         va_start(args, fmt);
 648         vaf.fmt = fmt;
 649         vaf.va = &args;
 650         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: %pV\n",
 651                sb->s_id, function, line, &vaf);
 652         va_end(args);
 653
 654         if (sb_rdonly(sb) == 0) {
 655                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 656                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 657                 /*
 658                  * Make sure updated value of ->s_mount_flags will be visible
 659                  * before ->s_flags update
 660                  */
 661                 smp_wmb();
 662                 sb->s_flags |= MS_RDONLY;
 663                 if (EXT4_SB(sb)->s_journal)
 664                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 665                 save_error_info(sb, function, line);
 666         }
 667         if (test_opt(sb, ERRORS_PANIC)) {
 668                 if (EXT4_SB(sb)->s_journal &&
 669                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 670                         return;
 671                 panic("EXT4-fs panic from previous error\n");
 672         }
 673 }
 674
 675 void __ext4_msg(struct super_block *sb,
 676                 const char *prefix, const char *fmt, ...)
 677 {
 678         struct va_format vaf;
 679         va_list args;
 680
 681         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 682                 return;
 683
 684         va_start(args, fmt);
 685         vaf.fmt = fmt;
 686         vaf.va = &args;
 687         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 688         va_end(args);
 689 }
 690
 691 #define ext4_warning_ratelimit(sb)                                      \
 692                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
 693                              "EXT4-fs warning")
 694
 695 void __ext4_warning(struct super_block *sb, const char *function,
 696                     unsigned int line, const char *fmt, ...)
 697 {
 698         struct va_format vaf;
 699         va_list args;
 700
 701         if (!ext4_warning_ratelimit(sb))
 702                 return;
 703
 704         va_start(args, fmt);
 705         vaf.fmt = fmt;
 706         vaf.va = &args;
 707         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 708                sb->s_id, function, line, &vaf);
 709         va_end(args);
 710 }
 711
 712 void __ext4_warning_inode(const struct inode *inode, const char *function,
 713                           unsigned int line, const char *fmt, ...)
 714 {
 715         struct va_format vaf;
 716         va_list args;
 717
 718         if (!ext4_warning_ratelimit(inode->i_sb))
 719                 return;
 720
 721         va_start(args, fmt);
 722         vaf.fmt = fmt;
 723         vaf.va = &args;
 724         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 725                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 726                function, line, inode->i_ino, current->comm, &vaf);
 727         va_end(args);
 728 }
 729
 730 void __ext4_grp_locked_error(const char *function, unsigned int line,
 731                              struct super_block *sb, ext4_group_t grp,
 732                              unsigned long ino, ext4_fsblk_t block,
 733                              const char *fmt, ...)
 734 __releases(bitlock)
 735 __acquires(bitlock)
 736 {
 737         struct va_format vaf;
 738         va_list args;
 739         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 740
 741         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 742                 return;
 743
 744         es->s_last_error_ino = cpu_to_le32(ino);
 745         es->s_last_error_block = cpu_to_le64(block);
 746         __save_error_info(sb, function, line);
 747
 748         if (ext4_error_ratelimit(sb)) {
 749                 va_start(args, fmt);
 750                 vaf.fmt = fmt;
 751                 vaf.va = &args;
 752                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 753                        sb->s_id, function, line, grp);
 754                 if (ino)
 755                         printk(KERN_CONT "inode %lu: ", ino);
 756                 if (block)
 757                         printk(KERN_CONT "block %llu:",
 758                                (unsigned long long) block);
 759                 printk(KERN_CONT "%pV\n", &vaf);
 760                 va_end(args);
 761         }
 762
 763         if (test_opt(sb, ERRORS_CONT)) {
 764                 ext4_commit_super(sb, 0);
 765                 return;
 766         }
 767
 768         ext4_unlock_group(sb, grp);
 769         ext4_commit_super(sb, 1);
 770         ext4_handle_error(sb);
 771         /*
 772          * We only get here in the ERRORS_RO case; relocking the group
 773          * may be dangerous, but nothing bad will happen since the
 774          * filesystem will have already been marked read/only and the
 775          * journal has been aborted.  We return 1 as a hint to callers
 776          * who might what to use the return value from
 777          * ext4_grp_locked_error() to distinguish between the
 778          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 779          * aggressively from the ext4 function in question, with a
 780          * more appropriate error code.
 781          */
 782         ext4_lock_group(sb, grp);
 783         return;
 784 }
 785
 786 void ext4_update_dynamic_rev(struct super_block *sb)
 787 {
 788         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 789
 790         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 791                 return;
 792
 793         ext4_warning(sb,
 794                      "updating to rev %d because of new feature flag, "
 795                      "running e2fsck is recommended",
 796                      EXT4_DYNAMIC_REV);
 797
 798         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 799         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 800         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 801         /* leave es->s_feature_*compat flags alone */
 802         /* es->s_uuid will be set by e2fsck if empty */
 803
 804         /*
 805          * The rest of the superblock fields should be zero, and if not it
 806          * means they are likely already in use, so leave them alone.  We
 807          * can leave it up to e2fsck to clean up any inconsistencies there.
 808          */
 809 }
 810
 811 /*
 812  * Open the external journal device
 813  */
 814 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 815 {
 816         struct block_device *bdev;
 817         char b[BDEVNAME_SIZE];
 818
 819         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 820         if (IS_ERR(bdev))
 821                 goto fail;
 822         return bdev;
 823
 824 fail:
 825         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 826                         __bdevname(dev, b), PTR_ERR(bdev));
 827         return NULL;
 828 }
 829
 830 /*
 831  * Release the journal device
 832  */
 833 static void ext4_blkdev_put(struct block_device *bdev)
 834 {
 835         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 836 }
 837
 838 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 839 {
 840         struct block_device *bdev;
 841         bdev = sbi->journal_bdev;
 842         if (bdev) {
 843                 ext4_blkdev_put(bdev);
 844                 sbi->journal_bdev = NULL;
 845         }
 846 }
 847
 848 static inline struct inode *orphan_list_entry(struct list_head *l)
 849 {
 850         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 851 }
 852
 853 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 854 {
 855         struct list_head *l;
 856
 857         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 858                  le32_to_cpu(sbi->s_es->s_last_orphan));
 859
 860         printk(KERN_ERR "sb_info orphan list:\n");
 861         list_for_each(l, &sbi->s_orphan) {
 862                 struct inode *inode = orphan_list_entry(l);
 863                 printk(KERN_ERR "  "
 864                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 865                        inode->i_sb->s_id, inode->i_ino, inode,
 866                        inode->i_mode, inode->i_nlink,
 867                        NEXT_ORPHAN(inode));
 868         }
 869 }
 870
 871 #ifdef CONFIG_QUOTA
 872 static int ext4_quota_off(struct super_block *sb, int type);
 873
 874 static inline void ext4_quota_off_umount(struct super_block *sb)
 875 {
 876         int type;
 877
 878         /* Use our quota_off function to clear inode flags etc. */
 879         for (type = 0; type < EXT4_MAXQUOTAS; type++)
 880                 ext4_quota_off(sb, type);
 881 }
 882
 883 /*
 884  * This is a helper function which is used in the mount/remount
 885  * codepaths (which holds s_umount) to fetch the quota file name.
 886  */
 887 static inline char *get_qf_name(struct super_block *sb,
 888                                 struct ext4_sb_info *sbi,
 889                                 int type)
 890 {
 891         return rcu_dereference_protected(sbi->s_qf_names[type],
 892                                          lockdep_is_held(&sb->s_umount));
 893 }
 894 #else
 895 static inline void ext4_quota_off_umount(struct super_block *sb)
 896 {
 897 }
 898 #endif
 899
 900 static void ext4_put_super(struct super_block *sb)
 901 {
 902         struct ext4_sb_info *sbi = EXT4_SB(sb);
 903         struct ext4_super_block *es = sbi->s_es;
 904         struct buffer_head **group_desc;
 905         struct flex_groups **flex_groups;
 906         int aborted = 0;
 907         int i, err;
 908
 909         ext4_unregister_li_request(sb);
 910         ext4_quota_off_umount(sb);
 911
 912         flush_workqueue(sbi->rsv_conversion_wq);
 913         destroy_workqueue(sbi->rsv_conversion_wq);
 914
 915         if (sbi->s_journal) {
 916                 aborted = is_journal_aborted(sbi->s_journal);
 917                 err = jbd2_journal_destroy(sbi->s_journal);
 918                 sbi->s_journal = NULL;
 919                 if ((err < 0) && !aborted)
 920                         ext4_abort(sb, "Couldn't clean up the journal");
 921         }
 922
 923         ext4_unregister_sysfs(sb);
 924         ext4_es_unregister_shrinker(sbi);
 925         del_timer_sync(&sbi->s_err_report);
 926         ext4_release_system_zone(sb);
 927         ext4_mb_release(sb);
 928         ext4_ext_release(sb);
 929
 930         if (!sb_rdonly(sb) && !aborted) {
 931                 ext4_clear_feature_journal_needs_recovery(sb);
 932                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 933         }
 934         if (!sb_rdonly(sb))
 935                 ext4_commit_super(sb, 1);
 936
 937         rcu_read_lock();
 938         group_desc = rcu_dereference(sbi->s_group_desc);
 939         for (i = 0; i < sbi->s_gdb_count; i++)
 940                 brelse(group_desc[i]);
 941         kvfree(group_desc);
 942         flex_groups = rcu_dereference(sbi->s_flex_groups);
 943         if (flex_groups) {
 944                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
 945                         kvfree(flex_groups[i]);
 946                 kvfree(flex_groups);
 947         }
 948         rcu_read_unlock();
 949         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 950         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 951         percpu_counter_destroy(&sbi->s_dirs_counter);
 952         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 953         percpu_free_rwsem(&sbi->s_writepages_rwsem);
 954 #ifdef CONFIG_QUOTA
 955         for (i = 0; i < EXT4_MAXQUOTAS; i++)
 956                 kfree(get_qf_name(sb, sbi, i));
 957 #endif
 958
 959         /* Debugging code just in case the in-memory inode orphan list
 960          * isn't empty.  The on-disk one can be non-empty if we've
 961          * detected an error and taken the fs readonly, but the
 962          * in-memory list had better be clean by this point. */
 963         if (!list_empty(&sbi->s_orphan))
 964                 dump_orphan_list(sb, sbi);
 965         J_ASSERT(list_empty(&sbi->s_orphan));
 966
 967         sync_blockdev(sb->s_bdev);
 968         invalidate_bdev(sb->s_bdev);
 969         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 970                 /*
 971                  * Invalidate the journal device's buffers.  We don't want them
 972                  * floating about in memory - the physical journal device may
 973                  * hotswapped, and it breaks the `ro-after' testing code.
 974                  */
 975                 sync_blockdev(sbi->journal_bdev);
 976                 invalidate_bdev(sbi->journal_bdev);
 977                 ext4_blkdev_remove(sbi);
 978         }
 979         if (sbi->s_ea_inode_cache) {
 980                 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
 981                 sbi->s_ea_inode_cache = NULL;
 982         }
 983         if (sbi->s_ea_block_cache) {
 984                 ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
 985                 sbi->s_ea_block_cache = NULL;
 986         }
 987         if (sbi->s_mmp_tsk)
 988                 kthread_stop(sbi->s_mmp_tsk);
 989         brelse(sbi->s_sbh);
 990         sb->s_fs_info = NULL;
 991         /*
 992          * Now that we are completely done shutting down the
 993          * superblock, we need to actually destroy the kobject.
 994          */
 995         kobject_put(&sbi->s_kobj);
 996         wait_for_completion(&sbi->s_kobj_unregister);
 997         if (sbi->s_chksum_driver)
 998                 crypto_free_shash(sbi->s_chksum_driver);
 999         kfree(sbi->s_blockgroup_lock);
1000         fs_put_dax(sbi->s_daxdev);
1001         kfree(sbi);
1002 }
1003
1004 static struct kmem_cache *ext4_inode_cachep;
1005
1006 /*
1007  * Called inside transaction, so use GFP_NOFS
1008  */
1009 static struct inode *ext4_alloc_inode(struct super_block *sb)
1010 {
1011         struct ext4_inode_info *ei;
1012
1013         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
1014         if (!ei)
1015                 return NULL;
1016
1017         ei->vfs_inode.i_version = 1;
1018         spin_lock_init(&ei->i_raw_lock);
1019         INIT_LIST_HEAD(&ei->i_prealloc_list);
1020         spin_lock_init(&ei->i_prealloc_lock);
1021         ext4_es_init_tree(&ei->i_es_tree);
1022         rwlock_init(&ei->i_es_lock);
1023         INIT_LIST_HEAD(&ei->i_es_list);
1024         ei->i_es_all_nr = 0;
1025         ei->i_es_shk_nr = 0;
1026         ei->i_es_shrink_lblk = 0;
1027         ei->i_reserved_data_blocks = 0;
1028         ei->i_da_metadata_calc_len = 0;
1029         ei->i_da_metadata_calc_last_lblock = 0;
1030         spin_lock_init(&(ei->i_block_reservation_lock));
1031 #ifdef CONFIG_QUOTA
1032         ei->i_reserved_quota = 0;
1033         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
1034 #endif
1035         ei->jinode = NULL;
1036         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
1037         spin_lock_init(&ei->i_completed_io_lock);
1038         ei->i_sync_tid = 0;
1039         ei->i_datasync_tid = 0;
1040         atomic_set(&ei->i_unwritten, 0);
1041         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
1042         return &ei->vfs_inode;
1043 }
1044
1045 static int ext4_drop_inode(struct inode *inode)
1046 {
1047         int drop = generic_drop_inode(inode);
1048
1049         trace_ext4_drop_inode(inode, drop);
1050         return drop;
1051 }
1052
1053 static void ext4_i_callback(struct rcu_head *head)
1054 {
1055         struct inode *inode = container_of(head, struct inode, i_rcu);
1056         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1057 }
1058
1059 static void ext4_destroy_inode(struct inode *inode)
1060 {
1061         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1062                 ext4_msg(inode->i_sb, KERN_ERR,
1063                          "Inode %lu (%p): orphan list check failed!",
1064                          inode->i_ino, EXT4_I(inode));
1065                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1066                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1067                                 true);
1068                 dump_stack();
1069         }
1070         call_rcu(&inode->i_rcu, ext4_i_callback);
1071 }
1072
1073 static void init_once(void *foo)
1074 {
1075         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1076
1077         INIT_LIST_HEAD(&ei->i_orphan);
1078         init_rwsem(&ei->xattr_sem);
1079         init_rwsem(&ei->i_data_sem);
1080         init_rwsem(&ei->i_mmap_sem);
1081         inode_init_once(&ei->vfs_inode);
1082 }
1083
1084 static int __init init_inodecache(void)
1085 {
1086         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
1087                                              sizeof(struct ext4_inode_info),
1088                                              0, (SLAB_RECLAIM_ACCOUNT|
1089                                                 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1090                                              init_once);
1091         if (ext4_inode_cachep == NULL)
1092                 return -ENOMEM;
1093         return 0;
1094 }
1095
1096 static void destroy_inodecache(void)
1097 {
1098         /*
1099          * Make sure all delayed rcu free inodes are flushed before we
1100          * destroy cache.
1101          */
1102         rcu_barrier();
1103         kmem_cache_destroy(ext4_inode_cachep);
1104 }
1105
1106 void ext4_clear_inode(struct inode *inode)
1107 {
1108         invalidate_inode_buffers(inode);
1109         clear_inode(inode);
1110         dquot_drop(inode);
1111         ext4_discard_preallocations(inode);
1112         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1113         if (EXT4_I(inode)->jinode) {
1114                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1115                                                EXT4_I(inode)->jinode);
1116                 jbd2_free_inode(EXT4_I(inode)->jinode);
1117                 EXT4_I(inode)->jinode = NULL;
1118         }
1119 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1120         fscrypt_put_encryption_info(inode, NULL);
1121 #endif
1122 }
1123
1124 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1125                                         u64 ino, u32 generation)
1126 {
1127         struct inode *inode;
1128
1129         /*
1130          * Currently we don't know the generation for parent directory, so
1131          * a generation of 0 means "accept any"
1132          */
1133         inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE);
1134         if (IS_ERR(inode))
1135                 return ERR_CAST(inode);
1136         if (generation && inode->i_generation != generation) {
1137                 iput(inode);
1138                 return ERR_PTR(-ESTALE);
1139         }
1140
1141         return inode;
1142 }
1143
1144 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1145                                         int fh_len, int fh_type)
1146 {
1147         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1148                                     ext4_nfs_get_inode);
1149 }
1150
1151 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1152                                         int fh_len, int fh_type)
1153 {
1154         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1155                                     ext4_nfs_get_inode);
1156 }
1157
1158 static int ext4_nfs_commit_metadata(struct inode *inode)
1159 {
1160         struct writeback_control wbc = {
1161                 .sync_mode = WB_SYNC_ALL
1162         };
1163
1164         trace_ext4_nfs_commit_metadata(inode);
1165         return ext4_write_inode(inode, &wbc);
1166 }
1167
1168 /*
1169  * Try to release metadata pages (indirect blocks, directories) which are
1170  * mapped via the block device.  Since these pages could have journal heads
1171  * which would prevent try_to_free_buffers() from freeing them, we must use
1172  * jbd2 layer's try_to_free_buffers() function to release them.
1173  */
1174 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1175                                  gfp_t wait)
1176 {
1177         journal_t *journal = EXT4_SB(sb)->s_journal;
1178
1179         WARN_ON(PageChecked(page));
1180         if (!page_has_buffers(page))
1181                 return 0;
1182         if (journal)
1183                 return jbd2_journal_try_to_free_buffers(journal, page,
1184                                                 wait & ~__GFP_DIRECT_RECLAIM);
1185         return try_to_free_buffers(page);
1186 }
1187
1188 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1189 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1190 {
1191         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1192                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1193 }
1194
1195 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1196                                                         void *fs_data)
1197 {
1198         handle_t *handle = fs_data;
1199         int res, res2, credits, retries = 0;
1200
1201         /*
1202          * Encrypting the root directory is not allowed because e2fsck expects
1203          * lost+found to exist and be unencrypted, and encrypting the root
1204          * directory would imply encrypting the lost+found directory as well as
1205          * the filename "lost+found" itself.
1206          */
1207         if (inode->i_ino == EXT4_ROOT_INO)
1208                 return -EPERM;
1209
1210         res = ext4_convert_inline_data(inode);
1211         if (res)
1212                 return res;
1213
1214         /*
1215          * If a journal handle was specified, then the encryption context is
1216          * being set on a new inode via inheritance and is part of a larger
1217          * transaction to create the inode.  Otherwise the encryption context is
1218          * being set on an existing inode in its own transaction.  Only in the
1219          * latter case should the "retry on ENOSPC" logic be used.
1220          */
1221
1222         if (handle) {
1223                 res = ext4_xattr_set_handle(handle, inode,
1224                                             EXT4_XATTR_INDEX_ENCRYPTION,
1225                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1226                                             ctx, len, 0);
1227                 if (!res) {
1228                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1229                         ext4_clear_inode_state(inode,
1230                                         EXT4_STATE_MAY_INLINE_DATA);
1231                         /*
1232                          * Update inode->i_flags - e.g. S_DAX may get disabled
1233                          */
1234                         ext4_set_inode_flags(inode);
1235                 }
1236                 return res;
1237         }
1238
1239         res = dquot_initialize(inode);
1240         if (res)
1241                 return res;
1242 retry:
1243         res = ext4_xattr_set_credits(inode, len, false /* is_create */,
1244                                      &credits);
1245         if (res)
1246                 return res;
1247
1248         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
1249         if (IS_ERR(handle))
1250                 return PTR_ERR(handle);
1251
1252         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1253                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1254                                     ctx, len, 0);
1255         if (!res) {
1256                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1257                 /* Update inode->i_flags - e.g. S_DAX may get disabled */
1258                 ext4_set_inode_flags(inode);
1259                 res = ext4_mark_inode_dirty(handle, inode);
1260                 if (res)
1261                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1262         }
1263         res2 = ext4_journal_stop(handle);
1264
1265         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1266                 goto retry;
1267         if (!res)
1268                 res = res2;
1269         return res;
1270 }
1271
1272 static bool ext4_dummy_context(struct inode *inode)
1273 {
1274         return DUMMY_ENCRYPTION_ENABLED(EXT4_SB(inode->i_sb));
1275 }
1276
1277 static unsigned ext4_max_namelen(struct inode *inode)
1278 {
1279         return S_ISLNK(inode->i_mode) ? inode->i_sb->s_blocksize :
1280                 EXT4_NAME_LEN;
1281 }
1282
1283 static const struct fscrypt_operations ext4_cryptops = {
1284         .key_prefix             = "ext4:",
1285         .get_context            = ext4_get_context,
1286         .set_context            = ext4_set_context,
1287         .dummy_context          = ext4_dummy_context,
1288         .is_encrypted           = ext4_encrypted_inode,
1289         .empty_dir              = ext4_empty_dir,
1290         .max_namelen            = ext4_max_namelen,
1291 };
1292 #else
1293 static const struct fscrypt_operations ext4_cryptops = {
1294         .is_encrypted           = ext4_encrypted_inode,
1295 };
1296 #endif
1297
1298 #ifdef CONFIG_QUOTA
1299 static const char * const quotatypes[] = INITQFNAMES;
1300 #define QTYPE2NAME(t) (quotatypes[t])
1301
1302 static int ext4_write_dquot(struct dquot *dquot);
1303 static int ext4_acquire_dquot(struct dquot *dquot);
1304 static int ext4_release_dquot(struct dquot *dquot);
1305 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1306 static int ext4_write_info(struct super_block *sb, int type);
1307 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1308                          const struct path *path);
1309 static int ext4_quota_on_mount(struct super_block *sb, int type);
1310 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1311                                size_t len, loff_t off);
1312 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1313                                 const char *data, size_t len, loff_t off);
1314 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1315                              unsigned int flags);
1316 static int ext4_enable_quotas(struct super_block *sb);
1317 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid);
1318
1319 static struct dquot **ext4_get_dquots(struct inode *inode)
1320 {
1321         return EXT4_I(inode)->i_dquot;
1322 }
1323
1324 static const struct dquot_operations ext4_quota_operations = {
1325         .get_reserved_space     = ext4_get_reserved_space,
1326         .write_dquot            = ext4_write_dquot,
1327         .acquire_dquot          = ext4_acquire_dquot,
1328         .release_dquot          = ext4_release_dquot,
1329         .mark_dirty             = ext4_mark_dquot_dirty,
1330         .write_info             = ext4_write_info,
1331         .alloc_dquot            = dquot_alloc,
1332         .destroy_dquot          = dquot_destroy,
1333         .get_projid             = ext4_get_projid,
1334         .get_inode_usage        = ext4_get_inode_usage,
1335         .get_next_id            = ext4_get_next_id,
1336 };
1337
1338 static const struct quotactl_ops ext4_qctl_operations = {
1339         .quota_on       = ext4_quota_on,
1340         .quota_off      = ext4_quota_off,
1341         .quota_sync     = dquot_quota_sync,
1342         .get_state      = dquot_get_state,
1343         .set_info       = dquot_set_dqinfo,
1344         .get_dqblk      = dquot_get_dqblk,
1345         .set_dqblk      = dquot_set_dqblk,
1346         .get_nextdqblk  = dquot_get_next_dqblk,
1347 };
1348 #endif
1349
1350 static const struct super_operations ext4_sops = {
1351         .alloc_inode    = ext4_alloc_inode,
1352         .destroy_inode  = ext4_destroy_inode,
1353         .write_inode    = ext4_write_inode,
1354         .dirty_inode    = ext4_dirty_inode,
1355         .drop_inode     = ext4_drop_inode,
1356         .evict_inode    = ext4_evict_inode,
1357         .put_super      = ext4_put_super,
1358         .sync_fs        = ext4_sync_fs,
1359         .freeze_fs      = ext4_freeze,
1360         .unfreeze_fs    = ext4_unfreeze,
1361         .statfs         = ext4_statfs,
1362         .remount_fs     = ext4_remount,
1363         .show_options   = ext4_show_options,
1364 #ifdef CONFIG_QUOTA
1365         .quota_read     = ext4_quota_read,
1366         .quota_write    = ext4_quota_write,
1367         .get_dquots     = ext4_get_dquots,
1368 #endif
1369         .bdev_try_to_free_page = bdev_try_to_free_page,
1370 };
1371
1372 static const struct export_operations ext4_export_ops = {
1373         .fh_to_dentry = ext4_fh_to_dentry,
1374         .fh_to_parent = ext4_fh_to_parent,
1375         .get_parent = ext4_get_parent,
1376         .commit_metadata = ext4_nfs_commit_metadata,
1377 };
1378
1379 enum {
1380         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1381         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1382         Opt_nouid32, Opt_debug, Opt_removed,
1383         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1384         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1385         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1386         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1387         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1388         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1389         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1390         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1391         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1392         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, Opt_dax,
1393         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1394         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1395         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1396         Opt_inode_readahead_blks, Opt_journal_ioprio,
1397         Opt_dioread_nolock, Opt_dioread_lock,
1398         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1399         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1400 };
1401
1402 static const match_table_t tokens = {
1403         {Opt_bsd_df, "bsddf"},
1404         {Opt_minix_df, "minixdf"},
1405         {Opt_grpid, "grpid"},
1406         {Opt_grpid, "bsdgroups"},
1407         {Opt_nogrpid, "nogrpid"},
1408         {Opt_nogrpid, "sysvgroups"},
1409         {Opt_resgid, "resgid=%u"},
1410         {Opt_resuid, "resuid=%u"},
1411         {Opt_sb, "sb=%u"},
1412         {Opt_err_cont, "errors=continue"},
1413         {Opt_err_panic, "errors=panic"},
1414         {Opt_err_ro, "errors=remount-ro"},
1415         {Opt_nouid32, "nouid32"},
1416         {Opt_debug, "debug"},
1417         {Opt_removed, "oldalloc"},
1418         {Opt_removed, "orlov"},
1419         {Opt_user_xattr, "user_xattr"},
1420         {Opt_nouser_xattr, "nouser_xattr"},
1421         {Opt_acl, "acl"},
1422         {Opt_noacl, "noacl"},
1423         {Opt_noload, "norecovery"},
1424         {Opt_noload, "noload"},
1425         {Opt_removed, "nobh"},
1426         {Opt_removed, "bh"},
1427         {Opt_commit, "commit=%u"},
1428         {Opt_min_batch_time, "min_batch_time=%u"},
1429         {Opt_max_batch_time, "max_batch_time=%u"},
1430         {Opt_journal_dev, "journal_dev=%u"},
1431         {Opt_journal_path, "journal_path=%s"},
1432         {Opt_journal_checksum, "journal_checksum"},
1433         {Opt_nojournal_checksum, "nojournal_checksum"},
1434         {Opt_journal_async_commit, "journal_async_commit"},
1435         {Opt_abort, "abort"},
1436         {Opt_data_journal, "data=journal"},
1437         {Opt_data_ordered, "data=ordered"},
1438         {Opt_data_writeback, "data=writeback"},
1439         {Opt_data_err_abort, "data_err=abort"},
1440         {Opt_data_err_ignore, "data_err=ignore"},
1441         {Opt_offusrjquota, "usrjquota="},
1442         {Opt_usrjquota, "usrjquota=%s"},
1443         {Opt_offgrpjquota, "grpjquota="},
1444         {Opt_grpjquota, "grpjquota=%s"},
1445         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1446         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1447         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1448         {Opt_grpquota, "grpquota"},
1449         {Opt_noquota, "noquota"},
1450         {Opt_quota, "quota"},
1451         {Opt_usrquota, "usrquota"},
1452         {Opt_prjquota, "prjquota"},
1453         {Opt_barrier, "barrier=%u"},
1454         {Opt_barrier, "barrier"},
1455         {Opt_nobarrier, "nobarrier"},
1456         {Opt_i_version, "i_version"},
1457         {Opt_dax, "dax"},
1458         {Opt_stripe, "stripe=%u"},
1459         {Opt_delalloc, "delalloc"},
1460         {Opt_lazytime, "lazytime"},
1461         {Opt_nolazytime, "nolazytime"},
1462         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1463         {Opt_nodelalloc, "nodelalloc"},
1464         {Opt_removed, "mblk_io_submit"},
1465         {Opt_removed, "nomblk_io_submit"},
1466         {Opt_block_validity, "block_validity"},
1467         {Opt_noblock_validity, "noblock_validity"},
1468         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1469         {Opt_journal_ioprio, "journal_ioprio=%u"},
1470         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1471         {Opt_auto_da_alloc, "auto_da_alloc"},
1472         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1473         {Opt_dioread_nolock, "dioread_nolock"},
1474         {Opt_dioread_lock, "dioread_lock"},
1475         {Opt_discard, "discard"},
1476         {Opt_nodiscard, "nodiscard"},
1477         {Opt_init_itable, "init_itable=%u"},
1478         {Opt_init_itable, "init_itable"},
1479         {Opt_noinit_itable, "noinit_itable"},
1480         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1481         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1482         {Opt_nombcache, "nombcache"},
1483         {Opt_nombcache, "no_mbcache"},  /* for backward compatibility */
1484         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1485         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1486         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1487         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1488         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1489         {Opt_err, NULL},
1490 };
1491
1492 static ext4_fsblk_t get_sb_block(void **data)
1493 {
1494         ext4_fsblk_t    sb_block;
1495         char            *options = (char *) *data;
1496
1497         if (!options || strncmp(options, "sb=", 3) != 0)
1498                 return 1;       /* Default location */
1499
1500         options += 3;
1501         /* TODO: use simple_strtoll with >32bit ext4 */
1502         sb_block = simple_strtoul(options, &options, 0);
1503         if (*options && *options != ',') {
1504                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1505                        (char *) *data);
1506                 return 1;
1507         }
1508         if (*options == ',')
1509                 options++;
1510         *data = (void *) options;
1511
1512         return sb_block;
1513 }
1514
1515 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1516 static const char deprecated_msg[] =
1517         "Mount option \"%s\" will be removed by %s\n"
1518         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1519
1520 #ifdef CONFIG_QUOTA
1521 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1522 {
1523         struct ext4_sb_info *sbi = EXT4_SB(sb);
1524         char *qname, *old_qname = get_qf_name(sb, sbi, qtype);
1525         int ret = -1;
1526
1527         if (sb_any_quota_loaded(sb) && !old_qname) {
1528                 ext4_msg(sb, KERN_ERR,
1529                         "Cannot change journaled "
1530                         "quota options when quota turned on");
1531                 return -1;
1532         }
1533         if (ext4_has_feature_quota(sb)) {
1534                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1535                          "ignored when QUOTA feature is enabled");
1536                 return 1;
1537         }
1538         qname = match_strdup(args);
1539         if (!qname) {
1540                 ext4_msg(sb, KERN_ERR,
1541                         "Not enough memory for storing quotafile name");
1542                 return -1;
1543         }
1544         if (old_qname) {
1545                 if (strcmp(old_qname, qname) == 0)
1546                         ret = 1;
1547                 else
1548                         ext4_msg(sb, KERN_ERR,
1549                                  "%s quota file already specified",
1550                                  QTYPE2NAME(qtype));
1551                 goto errout;
1552         }
1553         if (strchr(qname, '/')) {
1554                 ext4_msg(sb, KERN_ERR,
1555                         "quotafile must be on filesystem root");
1556                 goto errout;
1557         }
1558         rcu_assign_pointer(sbi->s_qf_names[qtype], qname);
1559         set_opt(sb, QUOTA);
1560         return 1;
1561 errout:
1562         kfree(qname);
1563         return ret;
1564 }
1565
1566 static int clear_qf_name(struct super_block *sb, int qtype)
1567 {
1568
1569         struct ext4_sb_info *sbi = EXT4_SB(sb);
1570         char *old_qname = get_qf_name(sb, sbi, qtype);
1571
1572         if (sb_any_quota_loaded(sb) && old_qname) {
1573                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1574                         " when quota turned on");
1575                 return -1;
1576         }
1577         rcu_assign_pointer(sbi->s_qf_names[qtype], NULL);
1578         synchronize_rcu();
1579         kfree(old_qname);
1580         return 1;
1581 }
1582 #endif
1583
1584 #define MOPT_SET        0x0001
1585 #define MOPT_CLEAR      0x0002
1586 #define MOPT_NOSUPPORT  0x0004
1587 #define MOPT_EXPLICIT   0x0008
1588 #define MOPT_CLEAR_ERR  0x0010
1589 #define MOPT_GTE0       0x0020
1590 #ifdef CONFIG_QUOTA
1591 #define MOPT_Q          0
1592 #define MOPT_QFMT       0x0040
1593 #else
1594 #define MOPT_Q          MOPT_NOSUPPORT
1595 #define MOPT_QFMT       MOPT_NOSUPPORT
1596 #endif
1597 #define MOPT_DATAJ      0x0080
1598 #define MOPT_NO_EXT2    0x0100
1599 #define MOPT_NO_EXT3    0x0200
1600 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1601 #define MOPT_STRING     0x0400
1602
1603 static const struct mount_opts {
1604         int     token;
1605         int     mount_opt;
1606         int     flags;
1607 } ext4_mount_opts[] = {
1608         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1609         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1610         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1611         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1612         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1613         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1614         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1615          MOPT_EXT4_ONLY | MOPT_SET},
1616         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1617          MOPT_EXT4_ONLY | MOPT_CLEAR},
1618         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1619         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1620         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1621          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1622         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1623          MOPT_EXT4_ONLY | MOPT_CLEAR},
1624         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1625          MOPT_EXT4_ONLY | MOPT_CLEAR},
1626         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1627          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1628         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1629                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1630          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1631         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1632         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1633         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1634         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1635         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1636          MOPT_NO_EXT2},
1637         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1638          MOPT_NO_EXT2},
1639         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1640         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1641         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1642         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1643         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1644         {Opt_commit, 0, MOPT_GTE0},
1645         {Opt_max_batch_time, 0, MOPT_GTE0},
1646         {Opt_min_batch_time, 0, MOPT_GTE0},
1647         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1648         {Opt_init_itable, 0, MOPT_GTE0},
1649         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1650         {Opt_stripe, 0, MOPT_GTE0},
1651         {Opt_resuid, 0, MOPT_GTE0},
1652         {Opt_resgid, 0, MOPT_GTE0},
1653         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1654         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1655         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1656         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1657         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1658         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1659          MOPT_NO_EXT2 | MOPT_DATAJ},
1660         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1661         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1662 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1663         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1664         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1665 #else
1666         {Opt_acl, 0, MOPT_NOSUPPORT},
1667         {Opt_noacl, 0, MOPT_NOSUPPORT},
1668 #endif
1669         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1670         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1671         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1672         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1673         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1674                                                         MOPT_SET | MOPT_Q},
1675         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1676                                                         MOPT_SET | MOPT_Q},
1677         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1678                                                         MOPT_SET | MOPT_Q},
1679         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1680                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1681                                                         MOPT_CLEAR | MOPT_Q},
1682         {Opt_usrjquota, 0, MOPT_Q | MOPT_STRING},
1683         {Opt_grpjquota, 0, MOPT_Q | MOPT_STRING},
1684         {Opt_offusrjquota, 0, MOPT_Q},
1685         {Opt_offgrpjquota, 0, MOPT_Q},
1686         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1687         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1688         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1689         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1690         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1691         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1692         {Opt_err, 0, 0}
1693 };
1694
1695 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1696                             substring_t *args, unsigned long *journal_devnum,
1697                             unsigned int *journal_ioprio, int is_remount)
1698 {
1699         struct ext4_sb_info *sbi = EXT4_SB(sb);
1700         const struct mount_opts *m;
1701         kuid_t uid;
1702         kgid_t gid;
1703         int arg = 0;
1704
1705 #ifdef CONFIG_QUOTA
1706         if (token == Opt_usrjquota)
1707                 return set_qf_name(sb, USRQUOTA, &args[0]);
1708         else if (token == Opt_grpjquota)
1709                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1710         else if (token == Opt_offusrjquota)
1711                 return clear_qf_name(sb, USRQUOTA);
1712         else if (token == Opt_offgrpjquota)
1713                 return clear_qf_name(sb, GRPQUOTA);
1714 #endif
1715         switch (token) {
1716         case Opt_noacl:
1717         case Opt_nouser_xattr:
1718                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1719                 break;
1720         case Opt_sb:
1721                 return 1;       /* handled by get_sb_block() */
1722         case Opt_removed:
1723                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1724                 return 1;
1725         case Opt_abort:
1726                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1727                 return 1;
1728         case Opt_i_version:
1729                 sb->s_flags |= SB_I_VERSION;
1730                 return 1;
1731         case Opt_lazytime:
1732                 sb->s_flags |= MS_LAZYTIME;
1733                 return 1;
1734         case Opt_nolazytime:
1735                 sb->s_flags &= ~MS_LAZYTIME;
1736                 return 1;
1737         }
1738
1739         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1740                 if (token == m->token)
1741                         break;
1742
1743         if (m->token == Opt_err) {
1744                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1745                          "or missing value", opt);
1746                 return -1;
1747         }
1748
1749         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1750                 ext4_msg(sb, KERN_ERR,
1751                          "Mount option \"%s\" incompatible with ext2", opt);
1752                 return -1;
1753         }
1754         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1755                 ext4_msg(sb, KERN_ERR,
1756                          "Mount option \"%s\" incompatible with ext3", opt);
1757                 return -1;
1758         }
1759
1760         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1761                 return -1;
1762         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1763                 return -1;
1764         if (m->flags & MOPT_EXPLICIT) {
1765                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1766                         set_opt2(sb, EXPLICIT_DELALLOC);
1767                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1768                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1769                 } else
1770                         return -1;
1771         }
1772         if (m->flags & MOPT_CLEAR_ERR)
1773                 clear_opt(sb, ERRORS_MASK);
1774         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1775                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1776                          "options when quota turned on");
1777                 return -1;
1778         }
1779
1780         if (m->flags & MOPT_NOSUPPORT) {
1781                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1782         } else if (token == Opt_commit) {
1783                 if (arg == 0)
1784                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1785                 sbi->s_commit_interval = HZ * arg;
1786         } else if (token == Opt_debug_want_extra_isize) {
1787                 if ((arg & 1) ||
1788                     (arg < 4) ||
1789                     (arg > (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE))) {
1790                         ext4_msg(sb, KERN_ERR,
1791                                  "Invalid want_extra_isize %d", arg);
1792                         return -1;
1793                 }
1794                 sbi->s_want_extra_isize = arg;
1795         } else if (token == Opt_max_batch_time) {
1796                 sbi->s_max_batch_time = arg;
1797         } else if (token == Opt_min_batch_time) {
1798                 sbi->s_min_batch_time = arg;
1799         } else if (token == Opt_inode_readahead_blks) {
1800                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1801                         ext4_msg(sb, KERN_ERR,
1802                                  "EXT4-fs: inode_readahead_blks must be "
1803                                  "0 or a power of 2 smaller than 2^31");
1804                         return -1;
1805                 }
1806                 sbi->s_inode_readahead_blks = arg;
1807         } else if (token == Opt_init_itable) {
1808                 set_opt(sb, INIT_INODE_TABLE);
1809                 if (!args->from)
1810                         arg = EXT4_DEF_LI_WAIT_MULT;
1811                 sbi->s_li_wait_mult = arg;
1812         } else if (token == Opt_max_dir_size_kb) {
1813                 sbi->s_max_dir_size_kb = arg;
1814         } else if (token == Opt_stripe) {
1815                 sbi->s_stripe = arg;
1816         } else if (token == Opt_resuid) {
1817                 uid = make_kuid(current_user_ns(), arg);
1818                 if (!uid_valid(uid)) {
1819                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1820                         return -1;
1821                 }
1822                 sbi->s_resuid = uid;
1823         } else if (token == Opt_resgid) {
1824                 gid = make_kgid(current_user_ns(), arg);
1825                 if (!gid_valid(gid)) {
1826                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1827                         return -1;
1828                 }
1829                 sbi->s_resgid = gid;
1830         } else if (token == Opt_journal_dev) {
1831                 if (is_remount) {
1832                         ext4_msg(sb, KERN_ERR,
1833                                  "Cannot specify journal on remount");
1834                         return -1;
1835                 }
1836                 *journal_devnum = arg;
1837         } else if (token == Opt_journal_path) {
1838                 char *journal_path;
1839                 struct inode *journal_inode;
1840                 struct path path;
1841                 int error;
1842
1843                 if (is_remount) {
1844                         ext4_msg(sb, KERN_ERR,
1845                                  "Cannot specify journal on remount");
1846                         return -1;
1847                 }
1848                 journal_path = match_strdup(&args[0]);
1849                 if (!journal_path) {
1850                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1851                                 "journal device string");
1852                         return -1;
1853                 }
1854
1855                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1856                 if (error) {
1857                         ext4_msg(sb, KERN_ERR, "error: could not find "
1858                                 "journal device path: error %d", error);
1859                         kfree(journal_path);
1860                         return -1;
1861                 }
1862
1863                 journal_inode = d_inode(path.dentry);
1864                 if (!S_ISBLK(journal_inode->i_mode)) {
1865                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1866                                 "is not a block device", journal_path);
1867                         path_put(&path);
1868                         kfree(journal_path);
1869                         return -1;
1870                 }
1871
1872                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1873                 path_put(&path);
1874                 kfree(journal_path);
1875         } else if (token == Opt_journal_ioprio) {
1876                 if (arg > 7) {
1877                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1878                                  " (must be 0-7)");
1879                         return -1;
1880                 }
1881                 *journal_ioprio =
1882                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1883         } else if (token == Opt_test_dummy_encryption) {
1884 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1885                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1886                 ext4_msg(sb, KERN_WARNING,
1887                          "Test dummy encryption mode enabled");
1888 #else
1889                 ext4_msg(sb, KERN_WARNING,
1890                          "Test dummy encryption mount option ignored");
1891 #endif
1892         } else if (m->flags & MOPT_DATAJ) {
1893                 if (is_remount) {
1894                         if (!sbi->s_journal)
1895                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1896                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1897                                 ext4_msg(sb, KERN_ERR,
1898                                          "Cannot change data mode on remount");
1899                                 return -1;
1900                         }
1901                 } else {
1902                         clear_opt(sb, DATA_FLAGS);
1903                         sbi->s_mount_opt |= m->mount_opt;
1904                 }
1905 #ifdef CONFIG_QUOTA
1906         } else if (m->flags & MOPT_QFMT) {
1907                 if (sb_any_quota_loaded(sb) &&
1908                     sbi->s_jquota_fmt != m->mount_opt) {
1909                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1910                                  "quota options when quota turned on");
1911                         return -1;
1912                 }
1913                 if (ext4_has_feature_quota(sb)) {
1914                         ext4_msg(sb, KERN_INFO,
1915                                  "Quota format mount options ignored "
1916                                  "when QUOTA feature is enabled");
1917                         return 1;
1918                 }
1919                 sbi->s_jquota_fmt = m->mount_opt;
1920 #endif
1921         } else if (token == Opt_dax) {
1922 #ifdef CONFIG_FS_DAX
1923                 ext4_msg(sb, KERN_WARNING,
1924                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1925                         sbi->s_mount_opt |= m->mount_opt;
1926 #else
1927                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1928                 return -1;
1929 #endif
1930         } else if (token == Opt_data_err_abort) {
1931                 sbi->s_mount_opt |= m->mount_opt;
1932         } else if (token == Opt_data_err_ignore) {
1933                 sbi->s_mount_opt &= ~m->mount_opt;
1934         } else {
1935                 if (!args->from)
1936                         arg = 1;
1937                 if (m->flags & MOPT_CLEAR)
1938                         arg = !arg;
1939                 else if (unlikely(!(m->flags & MOPT_SET))) {
1940                         ext4_msg(sb, KERN_WARNING,
1941                                  "buggy handling of option %s", opt);
1942                         WARN_ON(1);
1943                         return -1;
1944                 }
1945                 if (arg != 0)
1946                         sbi->s_mount_opt |= m->mount_opt;
1947                 else
1948                         sbi->s_mount_opt &= ~m->mount_opt;
1949         }
1950         return 1;
1951 }
1952
1953 static int parse_options(char *options, struct super_block *sb,
1954                          unsigned long *journal_devnum,
1955                          unsigned int *journal_ioprio,
1956                          int is_remount)
1957 {
1958         struct ext4_sb_info *sbi = EXT4_SB(sb);
1959         char *p, __maybe_unused *usr_qf_name, __maybe_unused *grp_qf_name;
1960         substring_t args[MAX_OPT_ARGS];
1961         int token;
1962
1963         if (!options)
1964                 return 1;
1965
1966         while ((p = strsep(&options, ",")) != NULL) {
1967                 if (!*p)
1968                         continue;
1969                 /*
1970                  * Initialize args struct so we know whether arg was
1971                  * found; some options take optional arguments.
1972                  */
1973                 args[0].to = args[0].from = NULL;
1974                 token = match_token(p, tokens, args);
1975                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1976                                      journal_ioprio, is_remount) < 0)
1977                         return 0;
1978         }
1979 #ifdef CONFIG_QUOTA
1980         /*
1981          * We do the test below only for project quotas. 'usrquota' and
1982          * 'grpquota' mount options are allowed even without quota feature
1983          * to support legacy quotas in quota files.
1984          */
1985         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
1986                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
1987                          "Cannot enable project quota enforcement.");
1988                 return 0;
1989         }
1990         usr_qf_name = get_qf_name(sb, sbi, USRQUOTA);
1991         grp_qf_name = get_qf_name(sb, sbi, GRPQUOTA);
1992         if (usr_qf_name || grp_qf_name) {
1993                 if (test_opt(sb, USRQUOTA) && usr_qf_name)
1994                         clear_opt(sb, USRQUOTA);
1995
1996                 if (test_opt(sb, GRPQUOTA) && grp_qf_name)
1997                         clear_opt(sb, GRPQUOTA);
1998
1999                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
2000                         ext4_msg(sb, KERN_ERR, "old and new quota "
2001                                         "format mixing");
2002                         return 0;
2003                 }
2004
2005                 if (!sbi->s_jquota_fmt) {
2006                         ext4_msg(sb, KERN_ERR, "journaled quota format "
2007                                         "not specified");
2008                         return 0;
2009                 }
2010         }
2011 #endif
2012         if (test_opt(sb, DIOREAD_NOLOCK)) {
2013                 int blocksize =
2014                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
2015
2016                 if (blocksize < PAGE_SIZE) {
2017                         ext4_msg(sb, KERN_ERR, "can't mount with "
2018                                  "dioread_nolock if block size != PAGE_SIZE");
2019                         return 0;
2020                 }
2021         }
2022         return 1;
2023 }
2024
2025 static inline void ext4_show_quota_options(struct seq_file *seq,
2026                                            struct super_block *sb)
2027 {
2028 #if defined(CONFIG_QUOTA)
2029         struct ext4_sb_info *sbi = EXT4_SB(sb);
2030         char *usr_qf_name, *grp_qf_name;
2031
2032         if (sbi->s_jquota_fmt) {
2033                 char *fmtname = "";
2034
2035                 switch (sbi->s_jquota_fmt) {
2036                 case QFMT_VFS_OLD:
2037                         fmtname = "vfsold";
2038                         break;
2039                 case QFMT_VFS_V0:
2040                         fmtname = "vfsv0";
2041                         break;
2042                 case QFMT_VFS_V1:
2043                         fmtname = "vfsv1";
2044                         break;
2045                 }
2046                 seq_printf(seq, ",jqfmt=%s", fmtname);
2047         }
2048
2049         rcu_read_lock();
2050         usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]);
2051         grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]);
2052         if (usr_qf_name)
2053                 seq_show_option(seq, "usrjquota", usr_qf_name);
2054         if (grp_qf_name)
2055                 seq_show_option(seq, "grpjquota", grp_qf_name);
2056         rcu_read_unlock();
2057 #endif
2058 }
2059
2060 static const char *token2str(int token)
2061 {
2062         const struct match_token *t;
2063
2064         for (t = tokens; t->token != Opt_err; t++)
2065                 if (t->token == token && !strchr(t->pattern, '='))
2066                         break;
2067         return t->pattern;
2068 }
2069
2070 /*
2071  * Show an option if
2072  *  - it's set to a non-default value OR
2073  *  - if the per-sb default is different from the global default
2074  */
2075 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2076                               int nodefs)
2077 {
2078         struct ext4_sb_info *sbi = EXT4_SB(sb);
2079         struct ext4_super_block *es = sbi->s_es;
2080         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
2081         const struct mount_opts *m;
2082         char sep = nodefs ? '\n' : ',';
2083
2084 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2085 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2086
2087         if (sbi->s_sb_block != 1)
2088                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2089
2090         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2091                 int want_set = m->flags & MOPT_SET;
2092                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2093                     (m->flags & MOPT_CLEAR_ERR))
2094                         continue;
2095                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2096                         continue; /* skip if same as the default */
2097                 if ((want_set &&
2098                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2099                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2100                         continue; /* select Opt_noFoo vs Opt_Foo */
2101                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2102         }
2103
2104         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2105             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2106                 SEQ_OPTS_PRINT("resuid=%u",
2107                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2108         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2109             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2110                 SEQ_OPTS_PRINT("resgid=%u",
2111                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2112         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2113         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2114                 SEQ_OPTS_PUTS("errors=remount-ro");
2115         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2116                 SEQ_OPTS_PUTS("errors=continue");
2117         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2118                 SEQ_OPTS_PUTS("errors=panic");
2119         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2120                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2121         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2122                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2123         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2124                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2125         if (sb->s_flags & SB_I_VERSION)
2126                 SEQ_OPTS_PUTS("i_version");
2127         if (nodefs || sbi->s_stripe)
2128                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2129         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
2130                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2131                         SEQ_OPTS_PUTS("data=journal");
2132                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2133                         SEQ_OPTS_PUTS("data=ordered");
2134                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2135                         SEQ_OPTS_PUTS("data=writeback");
2136         }
2137         if (nodefs ||
2138             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2139                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2140                                sbi->s_inode_readahead_blks);
2141
2142         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
2143                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2144                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2145         if (nodefs || sbi->s_max_dir_size_kb)
2146                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2147         if (test_opt(sb, DATA_ERR_ABORT))
2148                 SEQ_OPTS_PUTS("data_err=abort");
2149         if (DUMMY_ENCRYPTION_ENABLED(sbi))
2150                 SEQ_OPTS_PUTS("test_dummy_encryption");
2151
2152         ext4_show_quota_options(seq, sb);
2153         return 0;
2154 }
2155
2156 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2157 {
2158         return _ext4_show_options(seq, root->d_sb, 0);
2159 }
2160
2161 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2162 {
2163         struct super_block *sb = seq->private;
2164         int rc;
2165
2166         seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
2167         rc = _ext4_show_options(seq, sb, 1);
2168         seq_puts(seq, "\n");
2169         return rc;
2170 }
2171
2172 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2173                             int read_only)
2174 {
2175         struct ext4_sb_info *sbi = EXT4_SB(sb);
2176         int res = 0;
2177
2178         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2179                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2180                          "forcing read-only mode");
2181                 res = MS_RDONLY;
2182         }
2183         if (read_only)
2184                 goto done;
2185         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2186                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2187                          "running e2fsck is recommended");
2188         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2189                 ext4_msg(sb, KERN_WARNING,
2190                          "warning: mounting fs with errors, "
2191                          "running e2fsck is recommended");
2192         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2193                  le16_to_cpu(es->s_mnt_count) >=
2194                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2195                 ext4_msg(sb, KERN_WARNING,
2196                          "warning: maximal mount count reached, "
2197                          "running e2fsck is recommended");
2198         else if (le32_to_cpu(es->s_checkinterval) &&
2199                 (le32_to_cpu(es->s_lastcheck) +
2200                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
2201                 ext4_msg(sb, KERN_WARNING,
2202                          "warning: checktime reached, "
2203                          "running e2fsck is recommended");
2204         if (!sbi->s_journal)
2205                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2206         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2207                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2208         le16_add_cpu(&es->s_mnt_count, 1);
2209         es->s_mtime = cpu_to_le32(get_seconds());
2210         ext4_update_dynamic_rev(sb);
2211         if (sbi->s_journal)
2212                 ext4_set_feature_journal_needs_recovery(sb);
2213
2214         ext4_commit_super(sb, 1);
2215 done:
2216         if (test_opt(sb, DEBUG))
2217                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2218                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2219                         sb->s_blocksize,
2220                         sbi->s_groups_count,
2221                         EXT4_BLOCKS_PER_GROUP(sb),
2222                         EXT4_INODES_PER_GROUP(sb),
2223                         sbi->s_mount_opt, sbi->s_mount_opt2);
2224
2225         cleancache_init_fs(sb);
2226         return res;
2227 }
2228
2229 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2230 {
2231         struct ext4_sb_info *sbi = EXT4_SB(sb);
2232         struct flex_groups **old_groups, **new_groups;
2233         int size, i, j;
2234
2235         if (!sbi->s_log_groups_per_flex)
2236                 return 0;
2237
2238         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2239         if (size <= sbi->s_flex_groups_allocated)
2240                 return 0;
2241
2242         new_groups = kvzalloc(roundup_pow_of_two(size *
2243                               sizeof(*sbi->s_flex_groups)), GFP_KERNEL);
2244         if (!new_groups) {
2245                 ext4_msg(sb, KERN_ERR,
2246                          "not enough memory for %d flex group pointers", size);
2247                 return -ENOMEM;
2248         }
2249         for (i = sbi->s_flex_groups_allocated; i < size; i++) {
2250                 new_groups[i] = kvzalloc(roundup_pow_of_two(
2251                                          sizeof(struct flex_groups)),
2252                                          GFP_KERNEL);
2253                 if (!new_groups[i]) {
2254                         for (j = sbi->s_flex_groups_allocated; j < i; j++)
2255                                 kvfree(new_groups[j]);
2256                         kvfree(new_groups);
2257                         ext4_msg(sb, KERN_ERR,
2258                                  "not enough memory for %d flex groups", size);
2259                         return -ENOMEM;
2260                 }
2261         }
2262         rcu_read_lock();
2263         old_groups = rcu_dereference(sbi->s_flex_groups);
2264         if (old_groups)
2265                 memcpy(new_groups, old_groups,
2266                        (sbi->s_flex_groups_allocated *
2267                         sizeof(struct flex_groups *)));
2268         rcu_read_unlock();
2269         rcu_assign_pointer(sbi->s_flex_groups, new_groups);
2270         sbi->s_flex_groups_allocated = size;
2271         if (old_groups)
2272                 ext4_kvfree_array_rcu(old_groups);
2273         return 0;
2274 }
2275
2276 static int ext4_fill_flex_info(struct super_block *sb)
2277 {
2278         struct ext4_sb_info *sbi = EXT4_SB(sb);
2279         struct ext4_group_desc *gdp = NULL;
2280         struct flex_groups *fg;
2281         ext4_group_t flex_group;
2282         int i, err;
2283
2284         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2285         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2286                 sbi->s_log_groups_per_flex = 0;
2287                 return 1;
2288         }
2289
2290         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2291         if (err)
2292                 goto failed;
2293
2294         for (i = 0; i < sbi->s_groups_count; i++) {
2295                 gdp = ext4_get_group_desc(sb, i, NULL);
2296
2297                 flex_group = ext4_flex_group(sbi, i);
2298                 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
2299                 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes);
2300                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2301                              &fg->free_clusters);
2302                 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs);
2303         }
2304
2305         return 1;
2306 failed:
2307         return 0;
2308 }
2309
2310 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2311                                    struct ext4_group_desc *gdp)
2312 {
2313         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2314         __u16 crc = 0;
2315         __le32 le_group = cpu_to_le32(block_group);
2316         struct ext4_sb_info *sbi = EXT4_SB(sb);
2317
2318         if (ext4_has_metadata_csum(sbi->s_sb)) {
2319                 /* Use new metadata_csum algorithm */
2320                 __u32 csum32;
2321                 __u16 dummy_csum = 0;
2322
2323                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2324                                      sizeof(le_group));
2325                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2326                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2327                                      sizeof(dummy_csum));
2328                 offset += sizeof(dummy_csum);
2329                 if (offset < sbi->s_desc_size)
2330                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2331                                              sbi->s_desc_size - offset);
2332
2333                 crc = csum32 & 0xFFFF;
2334                 goto out;
2335         }
2336
2337         /* old crc16 code */
2338         if (!ext4_has_feature_gdt_csum(sb))
2339                 return 0;
2340
2341         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2342         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2343         crc = crc16(crc, (__u8 *)gdp, offset);
2344         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2345         /* for checksum of struct ext4_group_desc do the rest...*/
2346         if (ext4_has_feature_64bit(sb) &&
2347             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2348                 crc = crc16(crc, (__u8 *)gdp + offset,
2349                             le16_to_cpu(sbi->s_es->s_desc_size) -
2350                                 offset);
2351
2352 out:
2353         return cpu_to_le16(crc);
2354 }
2355
2356 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2357                                 struct ext4_group_desc *gdp)
2358 {
2359         if (ext4_has_group_desc_csum(sb) &&
2360             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2361                 return 0;
2362
2363         return 1;
2364 }
2365
2366 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2367                               struct ext4_group_desc *gdp)
2368 {
2369         if (!ext4_has_group_desc_csum(sb))
2370                 return;
2371         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2372 }
2373
2374 /* Called at mount-time, super-block is locked */
2375 static int ext4_check_descriptors(struct super_block *sb,
2376                                   ext4_fsblk_t sb_block,
2377                                   ext4_group_t *first_not_zeroed)
2378 {
2379         struct ext4_sb_info *sbi = EXT4_SB(sb);
2380         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2381         ext4_fsblk_t last_block;
2382         ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0);
2383         ext4_fsblk_t block_bitmap;
2384         ext4_fsblk_t inode_bitmap;
2385         ext4_fsblk_t inode_table;
2386         int flexbg_flag = 0;
2387         ext4_group_t i, grp = sbi->s_groups_count;
2388
2389         if (ext4_has_feature_flex_bg(sb))
2390                 flexbg_flag = 1;
2391
2392         ext4_debug("Checking group descriptors");
2393
2394         for (i = 0; i < sbi->s_groups_count; i++) {
2395                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2396
2397                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2398                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2399                 else
2400                         last_block = first_block +
2401                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2402
2403                 if ((grp == sbi->s_groups_count) &&
2404                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2405                         grp = i;
2406
2407                 block_bitmap = ext4_block_bitmap(sb, gdp);
2408                 if (block_bitmap == sb_block) {
2409                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2410                                  "Block bitmap for group %u overlaps "
2411                                  "superblock", i);
2412                         if (!sb_rdonly(sb))
2413                                 return 0;
2414                 }
2415                 if (block_bitmap >= sb_block + 1 &&
2416                     block_bitmap <= last_bg_block) {
2417                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2418                                  "Block bitmap for group %u overlaps "
2419                                  "block group descriptors", i);
2420                         if (!sb_rdonly(sb))
2421                                 return 0;
2422                 }
2423                 if (block_bitmap < first_block || block_bitmap > last_block) {
2424                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2425                                "Block bitmap for group %u not in group "
2426                                "(block %llu)!", i, block_bitmap);
2427                         return 0;
2428                 }
2429                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2430                 if (inode_bitmap == sb_block) {
2431                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2432                                  "Inode bitmap for group %u overlaps "
2433                                  "superblock", i);
2434                         if (!sb_rdonly(sb))
2435                                 return 0;
2436                 }
2437                 if (inode_bitmap >= sb_block + 1 &&
2438                     inode_bitmap <= last_bg_block) {
2439                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2440                                  "Inode bitmap for group %u overlaps "
2441                                  "block group descriptors", i);
2442                         if (!sb_rdonly(sb))
2443                                 return 0;
2444                 }
2445                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2446                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2447                                "Inode bitmap for group %u not in group "
2448                                "(block %llu)!", i, inode_bitmap);
2449                         return 0;
2450                 }
2451                 inode_table = ext4_inode_table(sb, gdp);
2452                 if (inode_table == sb_block) {
2453                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2454                                  "Inode table for group %u overlaps "
2455                                  "superblock", i);
2456                         if (!sb_rdonly(sb))
2457                                 return 0;
2458                 }
2459                 if (inode_table >= sb_block + 1 &&
2460                     inode_table <= last_bg_block) {
2461                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2462                                  "Inode table for group %u overlaps "
2463                                  "block group descriptors", i);
2464                         if (!sb_rdonly(sb))
2465                                 return 0;
2466                 }
2467                 if (inode_table < first_block ||
2468                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2469                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2470                                "Inode table for group %u not in group "
2471                                "(block %llu)!", i, inode_table);
2472                         return 0;
2473                 }
2474                 ext4_lock_group(sb, i);
2475                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2476                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2477                                  "Checksum for group %u failed (%u!=%u)",
2478                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2479                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2480                         if (!sb_rdonly(sb)) {
2481                                 ext4_unlock_group(sb, i);
2482                                 return 0;
2483                         }
2484                 }
2485                 ext4_unlock_group(sb, i);
2486                 if (!flexbg_flag)
2487                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2488         }
2489         if (NULL != first_not_zeroed)
2490                 *first_not_zeroed = grp;
2491         return 1;
2492 }
2493
2494 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2495  * the superblock) which were deleted from all directories, but held open by
2496  * a process at the time of a crash.  We walk the list and try to delete these
2497  * inodes at recovery time (only with a read-write filesystem).
2498  *
2499  * In order to keep the orphan inode chain consistent during traversal (in
2500  * case of crash during recovery), we link each inode into the superblock
2501  * orphan list_head and handle it the same way as an inode deletion during
2502  * normal operation (which journals the operations for us).
2503  *
2504  * We only do an iget() and an iput() on each inode, which is very safe if we
2505  * accidentally point at an in-use or already deleted inode.  The worst that
2506  * can happen in this case is that we get a "bit already cleared" message from
2507  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2508  * e2fsck was run on this filesystem, and it must have already done the orphan
2509  * inode cleanup for us, so we can safely abort without any further action.
2510  */
2511 static void ext4_orphan_cleanup(struct super_block *sb,
2512                                 struct ext4_super_block *es)
2513 {
2514         unsigned int s_flags = sb->s_flags;
2515         int ret, nr_orphans = 0, nr_truncates = 0;
2516 #ifdef CONFIG_QUOTA
2517         int quota_update = 0;
2518         int i;
2519 #endif
2520         if (!es->s_last_orphan) {
2521                 jbd_debug(4, "no orphan inodes to clean up\n");
2522                 return;
2523         }
2524
2525         if (bdev_read_only(sb->s_bdev)) {
2526                 ext4_msg(sb, KERN_ERR, "write access "
2527                         "unavailable, skipping orphan cleanup");
2528                 return;
2529         }
2530
2531         /* Check if feature set would not allow a r/w mount */
2532         if (!ext4_feature_set_ok(sb, 0)) {
2533                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2534                          "unknown ROCOMPAT features");
2535                 return;
2536         }
2537
2538         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2539                 /* don't clear list on RO mount w/ errors */
2540                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2541                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2542                                   "clearing orphan list.\n");
2543                         es->s_last_orphan = 0;
2544                 }
2545                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2546                 return;
2547         }
2548
2549         if (s_flags & MS_RDONLY) {
2550                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2551                 sb->s_flags &= ~MS_RDONLY;
2552         }
2553 #ifdef CONFIG_QUOTA
2554         /* Needed for iput() to work correctly and not trash data */
2555         sb->s_flags |= MS_ACTIVE;
2556
2557         /*
2558          * Turn on quotas which were not enabled for read-only mounts if
2559          * filesystem has quota feature, so that they are updated correctly.
2560          */
2561         if (ext4_has_feature_quota(sb) && (s_flags & MS_RDONLY)) {
2562                 int ret = ext4_enable_quotas(sb);
2563
2564                 if (!ret)
2565                         quota_update = 1;
2566                 else
2567                         ext4_msg(sb, KERN_ERR,
2568                                 "Cannot turn on quotas: error %d", ret);
2569         }
2570
2571         /* Turn on journaled quotas used for old sytle */
2572         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2573                 if (EXT4_SB(sb)->s_qf_names[i]) {
2574                         int ret = ext4_quota_on_mount(sb, i);
2575
2576                         if (!ret)
2577                                 quota_update = 1;
2578                         else
2579                                 ext4_msg(sb, KERN_ERR,
2580                                         "Cannot turn on journaled "
2581                                         "quota: type %d: error %d", i, ret);
2582                 }
2583         }
2584 #endif
2585
2586         while (es->s_last_orphan) {
2587                 struct inode *inode;
2588
2589                 /*
2590                  * We may have encountered an error during cleanup; if
2591                  * so, skip the rest.
2592                  */
2593                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2594                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2595                         es->s_last_orphan = 0;
2596                         break;
2597                 }
2598
2599                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2600                 if (IS_ERR(inode)) {
2601                         es->s_last_orphan = 0;
2602                         break;
2603                 }
2604
2605                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2606                 dquot_initialize(inode);
2607                 if (inode->i_nlink) {
2608                         if (test_opt(sb, DEBUG))
2609                                 ext4_msg(sb, KERN_DEBUG,
2610                                         "%s: truncating inode %lu to %lld bytes",
2611                                         __func__, inode->i_ino, inode->i_size);
2612                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2613                                   inode->i_ino, inode->i_size);
2614                         inode_lock(inode);
2615                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2616                         ret = ext4_truncate(inode);
2617                         if (ret) {
2618                                 /*
2619                                  * We need to clean up the in-core orphan list
2620                                  * manually if ext4_truncate() failed to get a
2621                                  * transaction handle.
2622                                  */
2623                                 ext4_orphan_del(NULL, inode);
2624                                 ext4_std_error(inode->i_sb, ret);
2625                         }
2626                         inode_unlock(inode);
2627                         nr_truncates++;
2628                 } else {
2629                         if (test_opt(sb, DEBUG))
2630                                 ext4_msg(sb, KERN_DEBUG,
2631                                         "%s: deleting unreferenced inode %lu",
2632                                         __func__, inode->i_ino);
2633                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2634                                   inode->i_ino);
2635                         nr_orphans++;
2636                 }
2637                 iput(inode);  /* The delete magic happens here! */
2638         }
2639
2640 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2641
2642         if (nr_orphans)
2643                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2644                        PLURAL(nr_orphans));
2645         if (nr_truncates)
2646                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2647                        PLURAL(nr_truncates));
2648 #ifdef CONFIG_QUOTA
2649         /* Turn off quotas if they were enabled for orphan cleanup */
2650         if (quota_update) {
2651                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2652                         if (sb_dqopt(sb)->files[i])
2653                                 dquot_quota_off(sb, i);
2654                 }
2655         }
2656 #endif
2657         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2658 }
2659
2660 /*
2661  * Maximal extent format file size.
2662  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2663  * extent format containers, within a sector_t, and within i_blocks
2664  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2665  * so that won't be a limiting factor.
2666  *
2667  * However there is other limiting factor. We do store extents in the form
2668  * of starting block and length, hence the resulting length of the extent
2669  * covering maximum file size must fit into on-disk format containers as
2670  * well. Given that length is always by 1 unit bigger than max unit (because
2671  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2672  *
2673  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2674  */
2675 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2676 {
2677         loff_t res;
2678         loff_t upper_limit = MAX_LFS_FILESIZE;
2679
2680         /* small i_blocks in vfs inode? */
2681         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2682                 /*
2683                  * CONFIG_LBDAF is not enabled implies the inode
2684                  * i_block represent total blocks in 512 bytes
2685                  * 32 == size of vfs inode i_blocks * 8
2686                  */
2687                 upper_limit = (1LL << 32) - 1;
2688
2689                 /* total blocks in file system block size */
2690                 upper_limit >>= (blkbits - 9);
2691                 upper_limit <<= blkbits;
2692         }
2693
2694         /*
2695          * 32-bit extent-start container, ee_block. We lower the maxbytes
2696          * by one fs block, so ee_len can cover the extent of maximum file
2697          * size
2698          */
2699         res = (1LL << 32) - 1;
2700         res <<= blkbits;
2701
2702         /* Sanity check against vm- & vfs- imposed limits */
2703         if (res > upper_limit)
2704                 res = upper_limit;
2705
2706         return res;
2707 }
2708
2709 /*
2710  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2711  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2712  * We need to be 1 filesystem block less than the 2^48 sector limit.
2713  */
2714 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2715 {
2716         loff_t res = EXT4_NDIR_BLOCKS;
2717         int meta_blocks;
2718         loff_t upper_limit;
2719         /* This is calculated to be the largest file size for a dense, block
2720          * mapped file such that the file's total number of 512-byte sectors,
2721          * including data and all indirect blocks, does not exceed (2^48 - 1).
2722          *
2723          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2724          * number of 512-byte sectors of the file.
2725          */
2726
2727         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2728                 /*
2729                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2730                  * the inode i_block field represents total file blocks in
2731                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2732                  */
2733                 upper_limit = (1LL << 32) - 1;
2734
2735                 /* total blocks in file system block size */
2736                 upper_limit >>= (bits - 9);
2737
2738         } else {
2739                 /*
2740                  * We use 48 bit ext4_inode i_blocks
2741                  * With EXT4_HUGE_FILE_FL set the i_blocks
2742                  * represent total number of blocks in
2743                  * file system block size
2744                  */
2745                 upper_limit = (1LL << 48) - 1;
2746
2747         }
2748
2749         /* indirect blocks */
2750         meta_blocks = 1;
2751         /* double indirect blocks */
2752         meta_blocks += 1 + (1LL << (bits-2));
2753         /* tripple indirect blocks */
2754         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2755
2756         upper_limit -= meta_blocks;
2757         upper_limit <<= bits;
2758
2759         res += 1LL << (bits-2);
2760         res += 1LL << (2*(bits-2));
2761         res += 1LL << (3*(bits-2));
2762         res <<= bits;
2763         if (res > upper_limit)
2764                 res = upper_limit;
2765
2766         if (res > MAX_LFS_FILESIZE)
2767                 res = MAX_LFS_FILESIZE;
2768
2769         return res;
2770 }
2771
2772 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2773                                    ext4_fsblk_t logical_sb_block, int nr)
2774 {
2775         struct ext4_sb_info *sbi = EXT4_SB(sb);
2776         ext4_group_t bg, first_meta_bg;
2777         int has_super = 0;
2778
2779         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2780
2781         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2782                 return logical_sb_block + nr + 1;
2783         bg = sbi->s_desc_per_block * nr;
2784         if (ext4_bg_has_super(sb, bg))
2785                 has_super = 1;
2786
2787         /*
2788          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2789          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2790          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2791          * compensate.
2792          */
2793         if (sb->s_blocksize == 1024 && nr == 0 &&
2794             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2795                 has_super++;
2796
2797         return (has_super + ext4_group_first_block_no(sb, bg));
2798 }
2799
2800 /**
2801  * ext4_get_stripe_size: Get the stripe size.
2802  * @sbi: In memory super block info
2803  *
2804  * If we have specified it via mount option, then
2805  * use the mount option value. If the value specified at mount time is
2806  * greater than the blocks per group use the super block value.
2807  * If the super block value is greater than blocks per group return 0.
2808  * Allocator needs it be less than blocks per group.
2809  *
2810  */
2811 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2812 {
2813         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2814         unsigned long stripe_width =
2815                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2816         int ret;
2817
2818         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2819                 ret = sbi->s_stripe;
2820         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
2821                 ret = stripe_width;
2822         else if (stride && stride <= sbi->s_blocks_per_group)
2823                 ret = stride;
2824         else
2825                 ret = 0;
2826
2827         /*
2828          * If the stripe width is 1, this makes no sense and
2829          * we set it to 0 to turn off stripe handling code.
2830          */
2831         if (ret <= 1)
2832                 ret = 0;
2833
2834         return ret;
2835 }
2836
2837 /*
2838  * Check whether this filesystem can be mounted based on
2839  * the features present and the RDONLY/RDWR mount requested.
2840  * Returns 1 if this filesystem can be mounted as requested,
2841  * 0 if it cannot be.
2842  */
2843 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2844 {
2845         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2846                 ext4_msg(sb, KERN_ERR,
2847                         "Couldn't mount because of "
2848                         "unsupported optional features (%x)",
2849                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2850                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2851                 return 0;
2852         }
2853
2854         if (readonly)
2855                 return 1;
2856
2857         if (ext4_has_feature_readonly(sb)) {
2858                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2859                 sb->s_flags |= MS_RDONLY;
2860                 return 1;
2861         }
2862
2863         /* Check that feature set is OK for a read-write mount */
2864         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2865                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2866                          "unsupported optional features (%x)",
2867                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2868                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2869                 return 0;
2870         }
2871         /*
2872          * Large file size enabled file system can only be mounted
2873          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2874          */
2875         if (ext4_has_feature_huge_file(sb)) {
2876                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2877                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2878                                  "cannot be mounted RDWR without "
2879                                  "CONFIG_LBDAF");
2880                         return 0;
2881                 }
2882         }
2883         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2884                 ext4_msg(sb, KERN_ERR,
2885                          "Can't support bigalloc feature without "
2886                          "extents feature\n");
2887                 return 0;
2888         }
2889
2890 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2)
2891         if (!readonly && (ext4_has_feature_quota(sb) ||
2892                           ext4_has_feature_project(sb))) {
2893                 ext4_msg(sb, KERN_ERR,
2894                          "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2");
2895                 return 0;
2896         }
2897 #endif  /* CONFIG_QUOTA */
2898         return 1;
2899 }
2900
2901 /*
2902  * This function is called once a day if we have errors logged
2903  * on the file system
2904  */
2905 static void print_daily_error_info(unsigned long arg)
2906 {
2907         struct super_block *sb = (struct super_block *) arg;
2908         struct ext4_sb_info *sbi;
2909         struct ext4_super_block *es;
2910
2911         sbi = EXT4_SB(sb);
2912         es = sbi->s_es;
2913
2914         if (es->s_error_count)
2915                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2916                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2917                          le32_to_cpu(es->s_error_count));
2918         if (es->s_first_error_time) {
2919                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2920                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2921                        (int) sizeof(es->s_first_error_func),
2922                        es->s_first_error_func,
2923                        le32_to_cpu(es->s_first_error_line));
2924                 if (es->s_first_error_ino)
2925                         printk(KERN_CONT ": inode %u",
2926                                le32_to_cpu(es->s_first_error_ino));
2927                 if (es->s_first_error_block)
2928                         printk(KERN_CONT ": block %llu", (unsigned long long)
2929                                le64_to_cpu(es->s_first_error_block));
2930                 printk(KERN_CONT "\n");
2931         }
2932         if (es->s_last_error_time) {
2933                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2934                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2935                        (int) sizeof(es->s_last_error_func),
2936                        es->s_last_error_func,
2937                        le32_to_cpu(es->s_last_error_line));
2938                 if (es->s_last_error_ino)
2939                         printk(KERN_CONT ": inode %u",
2940                                le32_to_cpu(es->s_last_error_ino));
2941                 if (es->s_last_error_block)
2942                         printk(KERN_CONT ": block %llu", (unsigned long long)
2943                                le64_to_cpu(es->s_last_error_block));
2944                 printk(KERN_CONT "\n");
2945         }
2946         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2947 }
2948
2949 /* Find next suitable group and run ext4_init_inode_table */
2950 static int ext4_run_li_request(struct ext4_li_request *elr)
2951 {
2952         struct ext4_group_desc *gdp = NULL;
2953         ext4_group_t group, ngroups;
2954         struct super_block *sb;
2955         int ret = 0;
2956         u64 start_time;
2957
2958         sb = elr->lr_super;
2959         ngroups = EXT4_SB(sb)->s_groups_count;
2960
2961         for (group = elr->lr_next_group; group < ngroups; group++) {
2962                 gdp = ext4_get_group_desc(sb, group, NULL);
2963                 if (!gdp) {
2964                         ret = 1;
2965                         break;
2966                 }
2967
2968                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2969                         break;
2970         }
2971
2972         if (group >= ngroups)
2973                 ret = 1;
2974
2975         if (!ret) {
2976                 start_time = ktime_get_real_ns();
2977                 ret = ext4_init_inode_table(sb, group,
2978                                             elr->lr_timeout ? 0 : 1);
2979                 if (elr->lr_timeout == 0) {
2980                         elr->lr_timeout = nsecs_to_jiffies((ktime_get_real_ns() - start_time) *
2981                                   elr->lr_sbi->s_li_wait_mult);
2982                 }
2983                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2984                 elr->lr_next_group = group + 1;
2985         }
2986         return ret;
2987 }
2988
2989 /*
2990  * Remove lr_request from the list_request and free the
2991  * request structure. Should be called with li_list_mtx held
2992  */
2993 static void ext4_remove_li_request(struct ext4_li_request *elr)
2994 {
2995         struct ext4_sb_info *sbi;
2996
2997         if (!elr)
2998                 return;
2999
3000         sbi = elr->lr_sbi;
3001
3002         list_del(&elr->lr_request);
3003         sbi->s_li_request = NULL;
3004         kfree(elr);
3005 }
3006
3007 static void ext4_unregister_li_request(struct super_block *sb)
3008 {
3009         mutex_lock(&ext4_li_mtx);
3010         if (!ext4_li_info) {
3011                 mutex_unlock(&ext4_li_mtx);
3012                 return;
3013         }
3014
3015         mutex_lock(&ext4_li_info->li_list_mtx);
3016         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
3017         mutex_unlock(&ext4_li_info->li_list_mtx);
3018         mutex_unlock(&ext4_li_mtx);
3019 }
3020
3021 static struct task_struct *ext4_lazyinit_task;
3022
3023 /*
3024  * This is the function where ext4lazyinit thread lives. It walks
3025  * through the request list searching for next scheduled filesystem.
3026  * When such a fs is found, run the lazy initialization request
3027  * (ext4_rn_li_request) and keep track of the time spend in this
3028  * function. Based on that time we compute next schedule time of
3029  * the request. When walking through the list is complete, compute
3030  * next waking time and put itself into sleep.
3031  */
3032 static int ext4_lazyinit_thread(void *arg)
3033 {
3034         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
3035         struct list_head *pos, *n;
3036         struct ext4_li_request *elr;
3037         unsigned long next_wakeup, cur;
3038
3039         BUG_ON(NULL == eli);
3040
3041 cont_thread:
3042         while (true) {
3043                 next_wakeup = MAX_JIFFY_OFFSET;
3044
3045                 mutex_lock(&eli->li_list_mtx);
3046                 if (list_empty(&eli->li_request_list)) {
3047                         mutex_unlock(&eli->li_list_mtx);
3048                         goto exit_thread;
3049                 }
3050                 list_for_each_safe(pos, n, &eli->li_request_list) {
3051                         int err = 0;
3052                         int progress = 0;
3053                         elr = list_entry(pos, struct ext4_li_request,
3054                                          lr_request);
3055
3056                         if (time_before(jiffies, elr->lr_next_sched)) {
3057                                 if (time_before(elr->lr_next_sched, next_wakeup))
3058                                         next_wakeup = elr->lr_next_sched;
3059                                 continue;
3060                         }
3061                         if (down_read_trylock(&elr->lr_super->s_umount)) {
3062                                 if (sb_start_write_trylock(elr->lr_super)) {
3063                                         progress = 1;
3064                                         /*
3065                                          * We hold sb->s_umount, sb can not
3066                                          * be removed from the list, it is
3067                                          * now safe to drop li_list_mtx
3068                                          */
3069                                         mutex_unlock(&eli->li_list_mtx);
3070                                         err = ext4_run_li_request(elr);
3071                                         sb_end_write(elr->lr_super);
3072                                         mutex_lock(&eli->li_list_mtx);
3073                                         n = pos->next;
3074                                 }
3075                                 up_read((&elr->lr_super->s_umount));
3076                         }
3077                         /* error, remove the lazy_init job */
3078                         if (err) {
3079                                 ext4_remove_li_request(elr);
3080                                 continue;
3081                         }
3082                         if (!progress) {
3083                                 elr->lr_next_sched = jiffies +
3084                                         (prandom_u32()
3085                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3086                         }
3087                         if (time_before(elr->lr_next_sched, next_wakeup))
3088                                 next_wakeup = elr->lr_next_sched;
3089                 }
3090                 mutex_unlock(&eli->li_list_mtx);
3091
3092                 try_to_freeze();
3093
3094                 cur = jiffies;
3095                 if ((time_after_eq(cur, next_wakeup)) ||
3096                     (MAX_JIFFY_OFFSET == next_wakeup)) {
3097                         cond_resched();
3098                         continue;
3099                 }
3100
3101                 schedule_timeout_interruptible(next_wakeup - cur);
3102
3103                 if (kthread_should_stop()) {
3104                         ext4_clear_request_list();
3105                         goto exit_thread;
3106                 }
3107         }
3108
3109 exit_thread:
3110         /*
3111          * It looks like the request list is empty, but we need
3112          * to check it under the li_list_mtx lock, to prevent any
3113          * additions into it, and of course we should lock ext4_li_mtx
3114          * to atomically free the list and ext4_li_info, because at
3115          * this point another ext4 filesystem could be registering
3116          * new one.
3117          */
3118         mutex_lock(&ext4_li_mtx);
3119         mutex_lock(&eli->li_list_mtx);
3120         if (!list_empty(&eli->li_request_list)) {
3121                 mutex_unlock(&eli->li_list_mtx);
3122                 mutex_unlock(&ext4_li_mtx);
3123                 goto cont_thread;
3124         }
3125         mutex_unlock(&eli->li_list_mtx);
3126         kfree(ext4_li_info);
3127         ext4_li_info = NULL;
3128         mutex_unlock(&ext4_li_mtx);
3129
3130         return 0;
3131 }
3132
3133 static void ext4_clear_request_list(void)
3134 {
3135         struct list_head *pos, *n;
3136         struct ext4_li_request *elr;
3137
3138         mutex_lock(&ext4_li_info->li_list_mtx);
3139         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3140                 elr = list_entry(pos, struct ext4_li_request,
3141                                  lr_request);
3142                 ext4_remove_li_request(elr);
3143         }
3144         mutex_unlock(&ext4_li_info->li_list_mtx);
3145 }
3146
3147 static int ext4_run_lazyinit_thread(void)
3148 {
3149         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3150                                          ext4_li_info, "ext4lazyinit");
3151         if (IS_ERR(ext4_lazyinit_task)) {
3152                 int err = PTR_ERR(ext4_lazyinit_task);
3153                 ext4_clear_request_list();
3154                 kfree(ext4_li_info);
3155                 ext4_li_info = NULL;
3156                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3157                                  "initialization thread\n",
3158                                  err);
3159                 return err;
3160         }
3161         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3162         return 0;
3163 }
3164
3165 /*
3166  * Check whether it make sense to run itable init. thread or not.
3167  * If there is at least one uninitialized inode table, return
3168  * corresponding group number, else the loop goes through all
3169  * groups and return total number of groups.
3170  */
3171 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3172 {
3173         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3174         struct ext4_group_desc *gdp = NULL;
3175
3176         if (!ext4_has_group_desc_csum(sb))
3177                 return ngroups;
3178
3179         for (group = 0; group < ngroups; group++) {
3180                 gdp = ext4_get_group_desc(sb, group, NULL);
3181                 if (!gdp)
3182                         continue;
3183
3184                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3185                         break;
3186         }
3187
3188         return group;
3189 }
3190
3191 static int ext4_li_info_new(void)
3192 {
3193         struct ext4_lazy_init *eli = NULL;
3194
3195         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3196         if (!eli)
3197                 return -ENOMEM;
3198
3199         INIT_LIST_HEAD(&eli->li_request_list);
3200         mutex_init(&eli->li_list_mtx);
3201
3202         eli->li_state |= EXT4_LAZYINIT_QUIT;
3203
3204         ext4_li_info = eli;
3205
3206         return 0;
3207 }
3208
3209 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3210                                             ext4_group_t start)
3211 {
3212         struct ext4_sb_info *sbi = EXT4_SB(sb);
3213         struct ext4_li_request *elr;
3214
3215         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3216         if (!elr)
3217                 return NULL;
3218
3219         elr->lr_super = sb;
3220         elr->lr_sbi = sbi;
3221         elr->lr_next_group = start;
3222
3223         /*
3224          * Randomize first schedule time of the request to
3225          * spread the inode table initialization requests
3226          * better.
3227          */
3228         elr->lr_next_sched = jiffies + (prandom_u32() %
3229                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3230         return elr;
3231 }
3232
3233 int ext4_register_li_request(struct super_block *sb,
3234                              ext4_group_t first_not_zeroed)
3235 {
3236         struct ext4_sb_info *sbi = EXT4_SB(sb);
3237         struct ext4_li_request *elr = NULL;
3238         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3239         int ret = 0;
3240
3241         mutex_lock(&ext4_li_mtx);
3242         if (sbi->s_li_request != NULL) {
3243                 /*
3244                  * Reset timeout so it can be computed again, because
3245                  * s_li_wait_mult might have changed.
3246                  */
3247                 sbi->s_li_request->lr_timeout = 0;
3248                 goto out;
3249         }
3250
3251         if (first_not_zeroed == ngroups || sb_rdonly(sb) ||
3252             !test_opt(sb, INIT_INODE_TABLE))
3253                 goto out;
3254
3255         elr = ext4_li_request_new(sb, first_not_zeroed);
3256         if (!elr) {
3257                 ret = -ENOMEM;
3258                 goto out;
3259         }
3260
3261         if (NULL == ext4_li_info) {
3262                 ret = ext4_li_info_new();
3263                 if (ret)
3264                         goto out;
3265         }
3266
3267         mutex_lock(&ext4_li_info->li_list_mtx);
3268         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3269         mutex_unlock(&ext4_li_info->li_list_mtx);
3270
3271         sbi->s_li_request = elr;
3272         /*
3273          * set elr to NULL here since it has been inserted to
3274          * the request_list and the removal and free of it is
3275          * handled by ext4_clear_request_list from now on.
3276          */
3277         elr = NULL;
3278
3279         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3280                 ret = ext4_run_lazyinit_thread();
3281                 if (ret)
3282                         goto out;
3283         }
3284 out:
3285         mutex_unlock(&ext4_li_mtx);
3286         if (ret)
3287                 kfree(elr);
3288         return ret;
3289 }
3290
3291 /*
3292  * We do not need to lock anything since this is called on
3293  * module unload.
3294  */
3295 static void ext4_destroy_lazyinit_thread(void)
3296 {
3297         /*
3298          * If thread exited earlier
3299          * there's nothing to be done.
3300          */
3301         if (!ext4_li_info || !ext4_lazyinit_task)
3302                 return;
3303
3304         kthread_stop(ext4_lazyinit_task);
3305 }
3306
3307 static int set_journal_csum_feature_set(struct super_block *sb)
3308 {
3309         int ret = 1;
3310         int compat, incompat;
3311         struct ext4_sb_info *sbi = EXT4_SB(sb);
3312
3313         if (ext4_has_metadata_csum(sb)) {
3314                 /* journal checksum v3 */
3315                 compat = 0;
3316                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3317         } else {
3318                 /* journal checksum v1 */
3319                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3320                 incompat = 0;
3321         }
3322
3323         jbd2_journal_clear_features(sbi->s_journal,
3324                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3325                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3326                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3327         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3328                 ret = jbd2_journal_set_features(sbi->s_journal,
3329                                 compat, 0,
3330                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3331                                 incompat);
3332         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3333                 ret = jbd2_journal_set_features(sbi->s_journal,
3334                                 compat, 0,
3335                                 incompat);
3336                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3337                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3338         } else {
3339                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3340                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3341         }
3342
3343         return ret;
3344 }
3345
3346 /*
3347  * Note: calculating the overhead so we can be compatible with
3348  * historical BSD practice is quite difficult in the face of
3349  * clusters/bigalloc.  This is because multiple metadata blocks from
3350  * different block group can end up in the same allocation cluster.
3351  * Calculating the exact overhead in the face of clustered allocation
3352  * requires either O(all block bitmaps) in memory or O(number of block
3353  * groups**2) in time.  We will still calculate the superblock for
3354  * older file systems --- and if we come across with a bigalloc file
3355  * system with zero in s_overhead_clusters the estimate will be close to
3356  * correct especially for very large cluster sizes --- but for newer
3357  * file systems, it's better to calculate this figure once at mkfs
3358  * time, and store it in the superblock.  If the superblock value is
3359  * present (even for non-bigalloc file systems), we will use it.
3360  */
3361 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3362                           char *buf)
3363 {
3364         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3365         struct ext4_group_desc  *gdp;
3366         ext4_fsblk_t            first_block, last_block, b;
3367         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3368         int                     s, j, count = 0;
3369         int                     has_super = ext4_bg_has_super(sb, grp);
3370
3371         if (!ext4_has_feature_bigalloc(sb))
3372                 return (has_super + ext4_bg_num_gdb(sb, grp) +
3373                         (has_super ? le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0) +
3374                         sbi->s_itb_per_group + 2);
3375
3376         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3377                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3378         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3379         for (i = 0; i < ngroups; i++) {
3380                 gdp = ext4_get_group_desc(sb, i, NULL);
3381                 b = ext4_block_bitmap(sb, gdp);
3382                 if (b >= first_block && b <= last_block) {
3383                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3384                         count++;
3385                 }
3386                 b = ext4_inode_bitmap(sb, gdp);
3387                 if (b >= first_block && b <= last_block) {
3388                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3389                         count++;
3390                 }
3391                 b = ext4_inode_table(sb, gdp);
3392                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3393                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3394                                 int c = EXT4_B2C(sbi, b - first_block);
3395                                 ext4_set_bit(c, buf);
3396                                 count++;
3397                         }
3398                 if (i != grp)
3399                         continue;
3400                 s = 0;
3401                 if (ext4_bg_has_super(sb, grp)) {
3402                         ext4_set_bit(s++, buf);
3403                         count++;
3404                 }
3405                 j = ext4_bg_num_gdb(sb, grp);
3406                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3407                         ext4_error(sb, "Invalid number of block group "
3408                                    "descriptor blocks: %d", j);
3409                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3410                 }
3411                 count += j;
3412                 for (; j > 0; j--)
3413                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3414         }
3415         if (!count)
3416                 return 0;
3417         return EXT4_CLUSTERS_PER_GROUP(sb) -
3418                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3419 }
3420
3421 /*
3422  * Compute the overhead and stash it in sbi->s_overhead
3423  */
3424 int ext4_calculate_overhead(struct super_block *sb)
3425 {
3426         struct ext4_sb_info *sbi = EXT4_SB(sb);
3427         struct ext4_super_block *es = sbi->s_es;
3428         struct inode *j_inode;
3429         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3430         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3431         ext4_fsblk_t overhead = 0;
3432         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3433
3434         if (!buf)
3435                 return -ENOMEM;
3436
3437         /*
3438          * Compute the overhead (FS structures).  This is constant
3439          * for a given filesystem unless the number of block groups
3440          * changes so we cache the previous value until it does.
3441          */
3442
3443         /*
3444          * All of the blocks before first_data_block are overhead
3445          */
3446         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3447
3448         /*
3449          * Add the overhead found in each block group
3450          */
3451         for (i = 0; i < ngroups; i++) {
3452                 int blks;
3453
3454                 blks = count_overhead(sb, i, buf);
3455                 overhead += blks;
3456                 if (blks)
3457                         memset(buf, 0, PAGE_SIZE);
3458                 cond_resched();
3459         }
3460
3461         /*
3462          * Add the internal journal blocks whether the journal has been
3463          * loaded or not
3464          */
3465         if (sbi->s_journal && !sbi->journal_bdev)
3466                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3467         else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) {
3468                 /* j_inum for internal journal is non-zero */
3469                 j_inode = ext4_get_journal_inode(sb, j_inum);
3470                 if (j_inode) {
3471                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3472                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
3473                         iput(j_inode);
3474                 } else {
3475                         ext4_msg(sb, KERN_ERR, "can't get journal size");
3476                 }
3477         }
3478         sbi->s_overhead = overhead;
3479         smp_wmb();
3480         free_page((unsigned long) buf);
3481         return 0;
3482 }
3483
3484 static void ext4_set_resv_clusters(struct super_block *sb)
3485 {
3486         ext4_fsblk_t resv_clusters;
3487         struct ext4_sb_info *sbi = EXT4_SB(sb);
3488
3489         /*
3490          * There's no need to reserve anything when we aren't using extents.
3491          * The space estimates are exact, there are no unwritten extents,
3492          * hole punching doesn't need new metadata... This is needed especially
3493          * to keep ext2/3 backward compatibility.
3494          */
3495         if (!ext4_has_feature_extents(sb))
3496                 return;
3497         /*
3498          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3499          * This should cover the situations where we can not afford to run
3500          * out of space like for example punch hole, or converting
3501          * unwritten extents in delalloc path. In most cases such
3502          * allocation would require 1, or 2 blocks, higher numbers are
3503          * very rare.
3504          */
3505         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3506                          sbi->s_cluster_bits);
3507
3508         do_div(resv_clusters, 50);
3509         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3510
3511         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3512 }
3513
3514 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3515 {
3516         struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev);
3517         char *orig_data = kstrdup(data, GFP_KERNEL);
3518         struct buffer_head *bh, **group_desc;
3519         struct ext4_super_block *es = NULL;
3520         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3521         struct flex_groups **flex_groups;
3522         ext4_fsblk_t block;
3523         ext4_fsblk_t sb_block = get_sb_block(&data);
3524         ext4_fsblk_t logical_sb_block;
3525         unsigned long offset = 0;
3526         unsigned long journal_devnum = 0;
3527         unsigned long def_mount_opts;
3528         struct inode *root;
3529         const char *descr;
3530         int ret = -ENOMEM;
3531         int blocksize, clustersize;
3532         unsigned int db_count;
3533         unsigned int i;
3534         int needs_recovery, has_huge_files, has_bigalloc;
3535         __u64 blocks_count;
3536         int err = 0;
3537         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3538         ext4_group_t first_not_zeroed;
3539
3540         if ((data && !orig_data) || !sbi)
3541                 goto out_free_base;
3542
3543         sbi->s_daxdev = dax_dev;
3544         sbi->s_blockgroup_lock =
3545                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3546         if (!sbi->s_blockgroup_lock)
3547                 goto out_free_base;
3548
3549         sb->s_fs_info = sbi;
3550         sbi->s_sb = sb;
3551         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3552         sbi->s_sb_block = sb_block;
3553         if (sb->s_bdev->bd_part)
3554                 sbi->s_sectors_written_start =
3555                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3556
3557         /* Cleanup superblock name */
3558         strreplace(sb->s_id, '/', '!');
3559
3560         /* -EINVAL is default */
3561         ret = -EINVAL;
3562         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3563         if (!blocksize) {
3564                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3565                 goto out_fail;
3566         }
3567
3568         /*
3569          * The ext4 superblock will not be buffer aligned for other than 1kB
3570          * block sizes.  We need to calculate the offset from buffer start.
3571          */
3572         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3573                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3574                 offset = do_div(logical_sb_block, blocksize);
3575         } else {
3576                 logical_sb_block = sb_block;
3577         }
3578
3579         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3580                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3581                 goto out_fail;
3582         }
3583         /*
3584          * Note: s_es must be initialized as soon as possible because
3585          *       some ext4 macro-instructions depend on its value
3586          */
3587         es = (struct ext4_super_block *) (bh->b_data + offset);
3588         sbi->s_es = es;
3589         sb->s_magic = le16_to_cpu(es->s_magic);
3590         if (sb->s_magic != EXT4_SUPER_MAGIC)
3591                 goto cantfind_ext4;
3592         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3593
3594         /* Warn if metadata_csum and gdt_csum are both set. */
3595         if (ext4_has_feature_metadata_csum(sb) &&
3596             ext4_has_feature_gdt_csum(sb))
3597                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3598                              "redundant flags; please run fsck.");
3599
3600         /* Check for a known checksum algorithm */
3601         if (!ext4_verify_csum_type(sb, es)) {
3602                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3603                          "unknown checksum algorithm.");
3604                 silent = 1;
3605                 goto cantfind_ext4;
3606         }
3607
3608         /* Load the checksum driver */
3609         sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3610         if (IS_ERR(sbi->s_chksum_driver)) {
3611                 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3612                 ret = PTR_ERR(sbi->s_chksum_driver);
3613                 sbi->s_chksum_driver = NULL;
3614                 goto failed_mount;
3615         }
3616
3617         /* Check superblock checksum */
3618         if (!ext4_superblock_csum_verify(sb, es)) {
3619                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3620                          "invalid superblock checksum.  Run e2fsck?");
3621                 silent = 1;
3622                 ret = -EFSBADCRC;
3623                 goto cantfind_ext4;
3624         }
3625
3626         /* Precompute checksum seed for all metadata */
3627         if (ext4_has_feature_csum_seed(sb))
3628                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3629         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
3630                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3631                                                sizeof(es->s_uuid));
3632
3633         /* Set defaults before we parse the mount options */
3634         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3635         set_opt(sb, INIT_INODE_TABLE);
3636         if (def_mount_opts & EXT4_DEFM_DEBUG)
3637                 set_opt(sb, DEBUG);
3638         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3639                 set_opt(sb, GRPID);
3640         if (def_mount_opts & EXT4_DEFM_UID16)
3641                 set_opt(sb, NO_UID32);
3642         /* xattr user namespace & acls are now defaulted on */
3643         set_opt(sb, XATTR_USER);
3644 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3645         set_opt(sb, POSIX_ACL);
3646 #endif
3647         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3648         if (ext4_has_metadata_csum(sb))
3649                 set_opt(sb, JOURNAL_CHECKSUM);
3650
3651         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3652                 set_opt(sb, JOURNAL_DATA);
3653         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3654                 set_opt(sb, ORDERED_DATA);
3655         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3656                 set_opt(sb, WRITEBACK_DATA);
3657
3658         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3659                 set_opt(sb, ERRORS_PANIC);
3660         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3661                 set_opt(sb, ERRORS_CONT);
3662         else
3663                 set_opt(sb, ERRORS_RO);
3664         /* block_validity enabled by default; disable with noblock_validity */
3665         set_opt(sb, BLOCK_VALIDITY);
3666         if (def_mount_opts & EXT4_DEFM_DISCARD)
3667                 set_opt(sb, DISCARD);
3668
3669         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3670         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3671         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3672         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3673         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3674
3675         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3676                 set_opt(sb, BARRIER);
3677
3678         /*
3679          * enable delayed allocation by default
3680          * Use -o nodelalloc to turn it off
3681          */
3682         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3683             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3684                 set_opt(sb, DELALLOC);
3685
3686         /*
3687          * set default s_li_wait_mult for lazyinit, for the case there is
3688          * no mount option specified.
3689          */
3690         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3691
3692         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3693         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3694             blocksize > EXT4_MAX_BLOCK_SIZE) {
3695                 ext4_msg(sb, KERN_ERR,
3696                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3697                          blocksize, le32_to_cpu(es->s_log_block_size));
3698                 goto failed_mount;
3699         }
3700
3701         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3702                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3703                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3704         } else {
3705                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3706                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3707                 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) {
3708                         ext4_msg(sb, KERN_ERR, "invalid first ino: %u",
3709                                  sbi->s_first_ino);
3710                         goto failed_mount;
3711                 }
3712                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3713                     (!is_power_of_2(sbi->s_inode_size)) ||
3714                     (sbi->s_inode_size > blocksize)) {
3715                         ext4_msg(sb, KERN_ERR,
3716                                "unsupported inode size: %d",
3717                                sbi->s_inode_size);
3718                         ext4_msg(sb, KERN_ERR, "blocksize: %d", blocksize);
3719                         goto failed_mount;
3720                 }
3721                 /*
3722                  * i_atime_extra is the last extra field available for
3723                  * [acm]times in struct ext4_inode. Checking for that
3724                  * field should suffice to ensure we have extra space
3725                  * for all three.
3726                  */
3727                 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) +
3728                         sizeof(((struct ext4_inode *)0)->i_atime_extra)) {
3729                         sb->s_time_gran = 1;
3730                 } else {
3731                         sb->s_time_gran = NSEC_PER_SEC;
3732                 }
3733         }
3734         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3735                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3736                         EXT4_GOOD_OLD_INODE_SIZE;
3737                 if (ext4_has_feature_extra_isize(sb)) {
3738                         unsigned v, max = (sbi->s_inode_size -
3739                                            EXT4_GOOD_OLD_INODE_SIZE);
3740
3741                         v = le16_to_cpu(es->s_want_extra_isize);
3742                         if (v > max) {
3743                                 ext4_msg(sb, KERN_ERR,
3744                                          "bad s_want_extra_isize: %d", v);
3745                                 goto failed_mount;
3746                         }
3747                         if (sbi->s_want_extra_isize < v)
3748                                 sbi->s_want_extra_isize = v;
3749
3750                         v = le16_to_cpu(es->s_min_extra_isize);
3751                         if (v > max) {
3752                                 ext4_msg(sb, KERN_ERR,
3753                                          "bad s_min_extra_isize: %d", v);
3754                                 goto failed_mount;
3755                         }
3756                         if (sbi->s_want_extra_isize < v)
3757                                 sbi->s_want_extra_isize = v;
3758                 }
3759         }
3760
3761         if (sbi->s_es->s_mount_opts[0]) {
3762                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3763                                               sizeof(sbi->s_es->s_mount_opts),
3764                                               GFP_KERNEL);
3765                 if (!s_mount_opts)
3766                         goto failed_mount;
3767                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3768                                    &journal_ioprio, 0)) {
3769                         ext4_msg(sb, KERN_WARNING,
3770                                  "failed to parse options in superblock: %s",
3771                                  s_mount_opts);
3772                 }
3773                 kfree(s_mount_opts);
3774         }
3775         sbi->s_def_mount_opt = sbi->s_mount_opt;
3776         if (!parse_options((char *) data, sb, &journal_devnum,
3777                            &journal_ioprio, 0))
3778                 goto failed_mount;
3779
3780         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3781                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3782                             "with data=journal disables delayed "
3783                             "allocation and O_DIRECT support!\n");
3784                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3785                         ext4_msg(sb, KERN_ERR, "can't mount with "
3786                                  "both data=journal and delalloc");
3787                         goto failed_mount;
3788                 }
3789                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3790                         ext4_msg(sb, KERN_ERR, "can't mount with "
3791                                  "both data=journal and dioread_nolock");
3792                         goto failed_mount;
3793                 }
3794                 if (test_opt(sb, DAX)) {
3795                         ext4_msg(sb, KERN_ERR, "can't mount with "
3796                                  "both data=journal and dax");
3797                         goto failed_mount;
3798                 }
3799                 if (ext4_has_feature_encrypt(sb)) {
3800                         ext4_msg(sb, KERN_WARNING,
3801                                  "encrypted files will use data=ordered "
3802                                  "instead of data journaling mode");
3803                 }
3804                 if (test_opt(sb, DELALLOC))
3805                         clear_opt(sb, DELALLOC);
3806         } else {
3807                 sb->s_iflags |= SB_I_CGROUPWB;
3808         }
3809
3810         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3811                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3812
3813         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3814             (ext4_has_compat_features(sb) ||
3815              ext4_has_ro_compat_features(sb) ||
3816              ext4_has_incompat_features(sb)))
3817                 ext4_msg(sb, KERN_WARNING,
3818                        "feature flags set on rev 0 fs, "
3819                        "running e2fsck is recommended");
3820
3821         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3822                 set_opt2(sb, HURD_COMPAT);
3823                 if (ext4_has_feature_64bit(sb)) {
3824                         ext4_msg(sb, KERN_ERR,
3825                                  "The Hurd can't support 64-bit file systems");
3826                         goto failed_mount;
3827                 }
3828
3829                 /*
3830                  * ea_inode feature uses l_i_version field which is not
3831                  * available in HURD_COMPAT mode.
3832                  */
3833                 if (ext4_has_feature_ea_inode(sb)) {
3834                         ext4_msg(sb, KERN_ERR,
3835                                  "ea_inode feature is not supported for Hurd");
3836                         goto failed_mount;
3837                 }
3838         }
3839
3840         if (IS_EXT2_SB(sb)) {
3841                 if (ext2_feature_set_ok(sb))
3842                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3843                                  "using the ext4 subsystem");
3844                 else {
3845                         /*
3846                          * If we're probing be silent, if this looks like
3847                          * it's actually an ext[34] filesystem.
3848                          */
3849                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
3850                                 goto failed_mount;
3851                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3852                                  "to feature incompatibilities");
3853                         goto failed_mount;
3854                 }
3855         }
3856
3857         if (IS_EXT3_SB(sb)) {
3858                 if (ext3_feature_set_ok(sb))
3859                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3860                                  "using the ext4 subsystem");
3861                 else {
3862                         /*
3863                          * If we're probing be silent, if this looks like
3864                          * it's actually an ext4 filesystem.
3865                          */
3866                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
3867                                 goto failed_mount;
3868                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3869                                  "to feature incompatibilities");
3870                         goto failed_mount;
3871                 }
3872         }
3873
3874         /*
3875          * Check feature flags regardless of the revision level, since we
3876          * previously didn't change the revision level when setting the flags,
3877          * so there is a chance incompat flags are set on a rev 0 filesystem.
3878          */
3879         if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
3880                 goto failed_mount;
3881
3882         if (le32_to_cpu(es->s_log_block_size) >
3883             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3884                 ext4_msg(sb, KERN_ERR,
3885                          "Invalid log block size: %u",
3886                          le32_to_cpu(es->s_log_block_size));
3887                 goto failed_mount;
3888         }
3889         if (le32_to_cpu(es->s_log_cluster_size) >
3890             (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3891                 ext4_msg(sb, KERN_ERR,
3892                          "Invalid log cluster size: %u",
3893                          le32_to_cpu(es->s_log_cluster_size));
3894                 goto failed_mount;
3895         }
3896
3897         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
3898                 ext4_msg(sb, KERN_ERR,
3899                          "Number of reserved GDT blocks insanely large: %d",
3900                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
3901                 goto failed_mount;
3902         }
3903
3904         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3905                 if (ext4_has_feature_inline_data(sb)) {
3906                         ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
3907                                         " that may contain inline data");
3908                         goto failed_mount;
3909                 }
3910                 if (!bdev_dax_supported(sb->s_bdev, blocksize))
3911                         goto failed_mount;
3912         }
3913
3914         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3915                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3916                          es->s_encryption_level);
3917                 goto failed_mount;
3918         }
3919
3920         if (sb->s_blocksize != blocksize) {
3921                 /* Validate the filesystem blocksize */
3922                 if (!sb_set_blocksize(sb, blocksize)) {
3923                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3924                                         blocksize);
3925                         goto failed_mount;
3926                 }
3927
3928                 brelse(bh);
3929                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3930                 offset = do_div(logical_sb_block, blocksize);
3931                 bh = sb_bread_unmovable(sb, logical_sb_block);
3932                 if (!bh) {
3933                         ext4_msg(sb, KERN_ERR,
3934                                "Can't read superblock on 2nd try");
3935                         goto failed_mount;
3936                 }
3937                 es = (struct ext4_super_block *)(bh->b_data + offset);
3938                 sbi->s_es = es;
3939                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3940                         ext4_msg(sb, KERN_ERR,
3941                                "Magic mismatch, very weird!");
3942                         goto failed_mount;
3943                 }
3944         }
3945
3946         has_huge_files = ext4_has_feature_huge_file(sb);
3947         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3948                                                       has_huge_files);
3949         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3950
3951         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3952         if (ext4_has_feature_64bit(sb)) {
3953                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3954                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3955                     !is_power_of_2(sbi->s_desc_size)) {
3956                         ext4_msg(sb, KERN_ERR,
3957                                "unsupported descriptor size %lu",
3958                                sbi->s_desc_size);
3959                         goto failed_mount;
3960                 }
3961         } else
3962                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3963
3964         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3965         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3966
3967         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3968         if (sbi->s_inodes_per_block == 0)
3969                 goto cantfind_ext4;
3970         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
3971             sbi->s_inodes_per_group > blocksize * 8) {
3972                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
3973                          sbi->s_inodes_per_group);
3974                 goto failed_mount;
3975         }
3976         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3977                                         sbi->s_inodes_per_block;
3978         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3979         sbi->s_sbh = bh;
3980         sbi->s_mount_state = le16_to_cpu(es->s_state);
3981         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3982         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3983
3984         for (i = 0; i < 4; i++)
3985                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3986         sbi->s_def_hash_version = es->s_def_hash_version;
3987         if (ext4_has_feature_dir_index(sb)) {
3988                 i = le32_to_cpu(es->s_flags);
3989                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3990                         sbi->s_hash_unsigned = 3;
3991                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3992 #ifdef __CHAR_UNSIGNED__
3993                         if (!sb_rdonly(sb))
3994                                 es->s_flags |=
3995                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3996                         sbi->s_hash_unsigned = 3;
3997 #else
3998                         if (!sb_rdonly(sb))
3999                                 es->s_flags |=
4000                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
4001 #endif
4002                 }
4003         }
4004
4005         /* Handle clustersize */
4006         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
4007         has_bigalloc = ext4_has_feature_bigalloc(sb);
4008         if (has_bigalloc) {
4009                 if (clustersize < blocksize) {
4010                         ext4_msg(sb, KERN_ERR,
4011                                  "cluster size (%d) smaller than "
4012                                  "block size (%d)", clustersize, blocksize);
4013                         goto failed_mount;
4014                 }
4015                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
4016                         le32_to_cpu(es->s_log_block_size);
4017                 sbi->s_clusters_per_group =
4018                         le32_to_cpu(es->s_clusters_per_group);
4019                 if (sbi->s_clusters_per_group > blocksize * 8) {
4020                         ext4_msg(sb, KERN_ERR,
4021                                  "#clusters per group too big: %lu",
4022                                  sbi->s_clusters_per_group);
4023                         goto failed_mount;
4024                 }
4025                 if (sbi->s_blocks_per_group !=
4026                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
4027                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
4028                                  "clusters per group (%lu) inconsistent",
4029                                  sbi->s_blocks_per_group,
4030                                  sbi->s_clusters_per_group);
4031                         goto failed_mount;
4032                 }
4033         } else {
4034                 if (clustersize != blocksize) {
4035                         ext4_msg(sb, KERN_ERR,
4036                                  "fragment/cluster size (%d) != "
4037                                  "block size (%d)", clustersize, blocksize);
4038                         goto failed_mount;
4039                 }
4040                 if (sbi->s_blocks_per_group > blocksize * 8) {
4041                         ext4_msg(sb, KERN_ERR,
4042                                  "#blocks per group too big: %lu",
4043                                  sbi->s_blocks_per_group);
4044                         goto failed_mount;
4045                 }
4046                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
4047                 sbi->s_cluster_bits = 0;
4048         }
4049         sbi->s_cluster_ratio = clustersize / blocksize;
4050
4051         /* Do we have standard group size of clustersize * 8 blocks ? */
4052         if (sbi->s_blocks_per_group == clustersize << 3)
4053                 set_opt2(sb, STD_GROUP_SIZE);
4054
4055         /*
4056          * Test whether we have more sectors than will fit in sector_t,
4057          * and whether the max offset is addressable by the page cache.
4058          */
4059         err = generic_check_addressable(sb->s_blocksize_bits,
4060                                         ext4_blocks_count(es));
4061         if (err) {
4062                 ext4_msg(sb, KERN_ERR, "filesystem"
4063                          " too large to mount safely on this system");
4064                 if (sizeof(sector_t) < 8)
4065                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
4066                 goto failed_mount;
4067         }
4068
4069         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
4070                 goto cantfind_ext4;
4071
4072         /* check blocks count against device size */
4073         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
4074         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
4075                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
4076                        "exceeds size of device (%llu blocks)",
4077                        ext4_blocks_count(es), blocks_count);
4078                 goto failed_mount;
4079         }
4080
4081         /*
4082          * It makes no sense for the first data block to be beyond the end
4083          * of the filesystem.
4084          */
4085         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
4086                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4087                          "block %u is beyond end of filesystem (%llu)",
4088                          le32_to_cpu(es->s_first_data_block),
4089                          ext4_blocks_count(es));
4090                 goto failed_mount;
4091         }
4092         if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) &&
4093             (sbi->s_cluster_ratio == 1)) {
4094                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4095                          "block is 0 with a 1k block and cluster size");
4096                 goto failed_mount;
4097         }
4098
4099         blocks_count = (ext4_blocks_count(es) -
4100                         le32_to_cpu(es->s_first_data_block) +
4101                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
4102         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
4103         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
4104                 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu "
4105                        "(block count %llu, first data block %u, "
4106                        "blocks per group %lu)", blocks_count,
4107                        ext4_blocks_count(es),
4108                        le32_to_cpu(es->s_first_data_block),
4109                        EXT4_BLOCKS_PER_GROUP(sb));
4110                 goto failed_mount;
4111         }
4112         sbi->s_groups_count = blocks_count;
4113         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
4114                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
4115         if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) !=
4116             le32_to_cpu(es->s_inodes_count)) {
4117                 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu",
4118                          le32_to_cpu(es->s_inodes_count),
4119                          ((u64)sbi->s_groups_count * sbi->s_inodes_per_group));
4120                 ret = -EINVAL;
4121                 goto failed_mount;
4122         }
4123         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
4124                    EXT4_DESC_PER_BLOCK(sb);
4125         if (ext4_has_feature_meta_bg(sb)) {
4126                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
4127                         ext4_msg(sb, KERN_WARNING,
4128                                  "first meta block group too large: %u "
4129                                  "(group descriptor block count %u)",
4130                                  le32_to_cpu(es->s_first_meta_bg), db_count);
4131                         goto failed_mount;
4132                 }
4133         }
4134         rcu_assign_pointer(sbi->s_group_desc,
4135                            kvmalloc_array(db_count,
4136                                           sizeof(struct buffer_head *),
4137                                           GFP_KERNEL));
4138         if (sbi->s_group_desc == NULL) {
4139                 ext4_msg(sb, KERN_ERR, "not enough memory");
4140                 ret = -ENOMEM;
4141                 goto failed_mount;
4142         }
4143
4144         bgl_lock_init(sbi->s_blockgroup_lock);
4145
4146         /* Pre-read the descriptors into the buffer cache */
4147         for (i = 0; i < db_count; i++) {
4148                 block = descriptor_loc(sb, logical_sb_block, i);
4149                 sb_breadahead_unmovable(sb, block);
4150         }
4151
4152         for (i = 0; i < db_count; i++) {
4153                 struct buffer_head *bh;
4154
4155                 block = descriptor_loc(sb, logical_sb_block, i);
4156                 bh = sb_bread_unmovable(sb, block);
4157                 if (!bh) {
4158                         ext4_msg(sb, KERN_ERR,
4159                                "can't read group descriptor %d", i);
4160                         db_count = i;
4161                         goto failed_mount2;
4162                 }
4163                 rcu_read_lock();
4164                 rcu_dereference(sbi->s_group_desc)[i] = bh;
4165                 rcu_read_unlock();
4166         }
4167         sbi->s_gdb_count = db_count;
4168         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
4169                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
4170                 ret = -EFSCORRUPTED;
4171                 goto failed_mount2;
4172         }
4173
4174         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
4175         spin_lock_init(&sbi->s_next_gen_lock);
4176
4177         setup_timer(&sbi->s_err_report, print_daily_error_info,
4178                 (unsigned long) sb);
4179
4180         /* Register extent status tree shrinker */
4181         if (ext4_es_register_shrinker(sbi))
4182                 goto failed_mount3;
4183
4184         sbi->s_stripe = ext4_get_stripe_size(sbi);
4185         sbi->s_extent_max_zeroout_kb = 32;
4186
4187         /*
4188          * set up enough so that it can read an inode
4189          */
4190         sb->s_op = &ext4_sops;
4191         sb->s_export_op = &ext4_export_ops;
4192         sb->s_xattr = ext4_xattr_handlers;
4193         sb->s_cop = &ext4_cryptops;
4194 #ifdef CONFIG_QUOTA
4195         sb->dq_op = &ext4_quota_operations;
4196         if (ext4_has_feature_quota(sb))
4197                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4198         else
4199                 sb->s_qcop = &ext4_qctl_operations;
4200         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
4201 #endif
4202         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
4203
4204         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
4205         mutex_init(&sbi->s_orphan_lock);
4206
4207         sb->s_root = NULL;
4208
4209         needs_recovery = (es->s_last_orphan != 0 ||
4210                           ext4_has_feature_journal_needs_recovery(sb));
4211
4212         if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb))
4213                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
4214                         goto failed_mount3a;
4215
4216         /*
4217          * The first inode we look at is the journal inode.  Don't try
4218          * root first: it may be modified in the journal!
4219          */
4220         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
4221                 err = ext4_load_journal(sb, es, journal_devnum);
4222                 if (err)
4223                         goto failed_mount3a;
4224         } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
4225                    ext4_has_feature_journal_needs_recovery(sb)) {
4226                 ext4_msg(sb, KERN_ERR, "required journal recovery "
4227                        "suppressed and not mounted read-only");
4228                 goto failed_mount_wq;
4229         } else {
4230                 /* Nojournal mode, all journal mount options are illegal */
4231                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
4232                         ext4_msg(sb, KERN_ERR, "can't mount with "
4233                                  "journal_checksum, fs mounted w/o journal");
4234                         goto failed_mount_wq;
4235                 }
4236                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4237                         ext4_msg(sb, KERN_ERR, "can't mount with "
4238                                  "journal_async_commit, fs mounted w/o journal");
4239                         goto failed_mount_wq;
4240                 }
4241                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
4242                         ext4_msg(sb, KERN_ERR, "can't mount with "
4243                                  "commit=%lu, fs mounted w/o journal",
4244                                  sbi->s_commit_interval / HZ);
4245                         goto failed_mount_wq;
4246                 }
4247                 if (EXT4_MOUNT_DATA_FLAGS &
4248                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
4249                         ext4_msg(sb, KERN_ERR, "can't mount with "
4250                                  "data=, fs mounted w/o journal");
4251                         goto failed_mount_wq;
4252                 }
4253                 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM;
4254                 clear_opt(sb, JOURNAL_CHECKSUM);
4255                 clear_opt(sb, DATA_FLAGS);
4256                 sbi->s_journal = NULL;
4257                 needs_recovery = 0;
4258                 goto no_journal;
4259         }
4260
4261         if (ext4_has_feature_64bit(sb) &&
4262             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4263                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4264                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4265                 goto failed_mount_wq;
4266         }
4267
4268         if (!set_journal_csum_feature_set(sb)) {
4269                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4270                          "feature set");
4271                 goto failed_mount_wq;
4272         }
4273
4274         /* We have now updated the journal if required, so we can
4275          * validate the data journaling mode. */
4276         switch (test_opt(sb, DATA_FLAGS)) {
4277         case 0:
4278                 /* No mode set, assume a default based on the journal
4279                  * capabilities: ORDERED_DATA if the journal can
4280                  * cope, else JOURNAL_DATA
4281                  */
4282                 if (jbd2_journal_check_available_features
4283                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
4284                         set_opt(sb, ORDERED_DATA);
4285                 else
4286                         set_opt(sb, JOURNAL_DATA);
4287                 break;
4288
4289         case EXT4_MOUNT_ORDERED_DATA:
4290         case EXT4_MOUNT_WRITEBACK_DATA:
4291                 if (!jbd2_journal_check_available_features
4292                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4293                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4294                                "requested data journaling mode");
4295                         goto failed_mount_wq;
4296                 }
4297         default:
4298                 break;
4299         }
4300
4301         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4302             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4303                 ext4_msg(sb, KERN_ERR, "can't mount with "
4304                         "journal_async_commit in data=ordered mode");
4305                 goto failed_mount_wq;
4306         }
4307
4308         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4309
4310         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4311
4312 no_journal:
4313         if (!test_opt(sb, NO_MBCACHE)) {
4314                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
4315                 if (!sbi->s_ea_block_cache) {
4316                         ext4_msg(sb, KERN_ERR,
4317                                  "Failed to create ea_block_cache");
4318                         goto failed_mount_wq;
4319                 }
4320
4321                 if (ext4_has_feature_ea_inode(sb)) {
4322                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
4323                         if (!sbi->s_ea_inode_cache) {
4324                                 ext4_msg(sb, KERN_ERR,
4325                                          "Failed to create ea_inode_cache");
4326                                 goto failed_mount_wq;
4327                         }
4328                 }
4329         }
4330
4331         if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
4332             (blocksize != PAGE_SIZE)) {
4333                 ext4_msg(sb, KERN_ERR,
4334                          "Unsupported blocksize for fs encryption");
4335                 goto failed_mount_wq;
4336         }
4337
4338         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) &&
4339             !ext4_has_feature_encrypt(sb)) {
4340                 ext4_set_feature_encrypt(sb);
4341                 ext4_commit_super(sb, 1);
4342         }
4343
4344         /*
4345          * Get the # of file system overhead blocks from the
4346          * superblock if present.
4347          */
4348         sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4349         /* ignore the precalculated value if it is ridiculous */
4350         if (sbi->s_overhead > ext4_blocks_count(es))
4351                 sbi->s_overhead = 0;
4352         /*
4353          * If the bigalloc feature is not enabled recalculating the
4354          * overhead doesn't take long, so we might as well just redo
4355          * it to make sure we are using the correct value.
4356          */
4357         if (!ext4_has_feature_bigalloc(sb))
4358                 sbi->s_overhead = 0;
4359         if (sbi->s_overhead == 0) {
4360                 err = ext4_calculate_overhead(sb);
4361                 if (err)
4362                         goto failed_mount_wq;
4363         }
4364
4365         /*
4366          * The maximum number of concurrent works can be high and
4367          * concurrency isn't really necessary.  Limit it to 1.
4368          */
4369         EXT4_SB(sb)->rsv_conversion_wq =
4370                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4371         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4372                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4373                 ret = -ENOMEM;
4374                 goto failed_mount4;
4375         }
4376
4377         /*
4378          * The jbd2_journal_load will have done any necessary log recovery,
4379          * so we can safely mount the rest of the filesystem now.
4380          */
4381
4382         root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL);
4383         if (IS_ERR(root)) {
4384                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4385                 ret = PTR_ERR(root);
4386                 root = NULL;
4387                 goto failed_mount4;
4388         }
4389         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4390                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4391                 iput(root);
4392                 goto failed_mount4;
4393         }
4394         sb->s_root = d_make_root(root);
4395         if (!sb->s_root) {
4396                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4397                 ret = -ENOMEM;
4398                 goto failed_mount4;
4399         }
4400
4401         if (ext4_setup_super(sb, es, sb_rdonly(sb)))
4402                 sb->s_flags |= MS_RDONLY;
4403
4404         ext4_set_resv_clusters(sb);
4405
4406         err = ext4_setup_system_zone(sb);
4407         if (err) {
4408                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4409                          "zone (%d)", err);
4410                 goto failed_mount4a;
4411         }
4412
4413         ext4_ext_init(sb);
4414         err = ext4_mb_init(sb);
4415         if (err) {
4416                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4417                          err);
4418                 goto failed_mount5;
4419         }
4420
4421         block = ext4_count_free_clusters(sb);
4422         ext4_free_blocks_count_set(sbi->s_es,
4423                                    EXT4_C2B(sbi, block));
4424         ext4_superblock_csum_set(sb);
4425         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4426                                   GFP_KERNEL);
4427         if (!err) {
4428                 unsigned long freei = ext4_count_free_inodes(sb);
4429                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4430                 ext4_superblock_csum_set(sb);
4431                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4432                                           GFP_KERNEL);
4433         }
4434         if (!err)
4435                 err = percpu_counter_init(&sbi->s_dirs_counter,
4436                                           ext4_count_dirs(sb), GFP_KERNEL);
4437         if (!err)
4438                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4439                                           GFP_KERNEL);
4440         if (!err)
4441                 err = percpu_init_rwsem(&sbi->s_writepages_rwsem);
4442
4443         if (err) {
4444                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4445                 goto failed_mount6;
4446         }
4447
4448         if (ext4_has_feature_flex_bg(sb))
4449                 if (!ext4_fill_flex_info(sb)) {
4450                         ext4_msg(sb, KERN_ERR,
4451                                "unable to initialize "
4452                                "flex_bg meta info!");
4453                         goto failed_mount6;
4454                 }
4455
4456         err = ext4_register_li_request(sb, first_not_zeroed);
4457         if (err)
4458                 goto failed_mount6;
4459
4460         err = ext4_register_sysfs(sb);
4461         if (err)
4462                 goto failed_mount7;
4463
4464 #ifdef CONFIG_QUOTA
4465         /* Enable quota usage during mount. */
4466         if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
4467                 err = ext4_enable_quotas(sb);
4468                 if (err)
4469                         goto failed_mount8;
4470         }
4471 #endif  /* CONFIG_QUOTA */
4472
4473         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4474         ext4_orphan_cleanup(sb, es);
4475         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4476         if (needs_recovery) {
4477                 ext4_msg(sb, KERN_INFO, "recovery complete");
4478                 err = ext4_mark_recovery_complete(sb, es);
4479                 if (err)
4480                         goto failed_mount8;
4481         }
4482         if (EXT4_SB(sb)->s_journal) {
4483                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4484                         descr = " journalled data mode";
4485                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4486                         descr = " ordered data mode";
4487                 else
4488                         descr = " writeback data mode";
4489         } else
4490                 descr = "out journal";
4491
4492         if (test_opt(sb, DISCARD)) {
4493                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4494                 if (!blk_queue_discard(q))
4495                         ext4_msg(sb, KERN_WARNING,
4496                                  "mounting with \"discard\" option, but "
4497                                  "the device does not support discard");
4498         }
4499
4500         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4501                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4502                          "Opts: %.*s%s%s", descr,
4503                          (int) sizeof(sbi->s_es->s_mount_opts),
4504                          sbi->s_es->s_mount_opts,
4505                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4506
4507         if (es->s_error_count)
4508                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4509
4510         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4511         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4512         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4513         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4514
4515         kfree(orig_data);
4516         return 0;
4517
4518 cantfind_ext4:
4519         if (!silent)
4520                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4521         goto failed_mount;
4522
4523 failed_mount8:
4524         ext4_unregister_sysfs(sb);
4525         kobject_put(&sbi->s_kobj);
4526 failed_mount7:
4527         ext4_unregister_li_request(sb);
4528 failed_mount6:
4529         ext4_mb_release(sb);
4530         rcu_read_lock();
4531         flex_groups = rcu_dereference(sbi->s_flex_groups);
4532         if (flex_groups) {
4533                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
4534                         kvfree(flex_groups[i]);
4535                 kvfree(flex_groups);
4536         }
4537         rcu_read_unlock();
4538         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4539         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4540         percpu_counter_destroy(&sbi->s_dirs_counter);
4541         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4542         percpu_free_rwsem(&sbi->s_writepages_rwsem);
4543 failed_mount5:
4544         ext4_ext_release(sb);
4545         ext4_release_system_zone(sb);
4546 failed_mount4a:
4547         dput(sb->s_root);
4548         sb->s_root = NULL;
4549 failed_mount4:
4550         ext4_msg(sb, KERN_ERR, "mount failed");
4551         if (EXT4_SB(sb)->rsv_conversion_wq)
4552                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4553 failed_mount_wq:
4554         if (sbi->s_ea_inode_cache) {
4555                 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
4556                 sbi->s_ea_inode_cache = NULL;
4557         }
4558         if (sbi->s_ea_block_cache) {
4559                 ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
4560                 sbi->s_ea_block_cache = NULL;
4561         }
4562         if (sbi->s_journal) {
4563                 jbd2_journal_destroy(sbi->s_journal);
4564                 sbi->s_journal = NULL;
4565         }
4566 failed_mount3a:
4567         ext4_es_unregister_shrinker(sbi);
4568 failed_mount3:
4569         del_timer_sync(&sbi->s_err_report);
4570         if (sbi->s_mmp_tsk)
4571                 kthread_stop(sbi->s_mmp_tsk);
4572 failed_mount2:
4573         rcu_read_lock();
4574         group_desc = rcu_dereference(sbi->s_group_desc);
4575         for (i = 0; i < db_count; i++)
4576                 brelse(group_desc[i]);
4577         kvfree(group_desc);
4578         rcu_read_unlock();
4579 failed_mount:
4580         if (sbi->s_chksum_driver)
4581                 crypto_free_shash(sbi->s_chksum_driver);
4582 #ifdef CONFIG_QUOTA
4583         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4584                 kfree(sbi->s_qf_names[i]);
4585 #endif
4586         ext4_blkdev_remove(sbi);
4587         brelse(bh);
4588 out_fail:
4589         sb->s_fs_info = NULL;
4590         kfree(sbi->s_blockgroup_lock);
4591 out_free_base:
4592         kfree(sbi);
4593         kfree(orig_data);
4594         fs_put_dax(dax_dev);
4595         return err ? err : ret;
4596 }
4597
4598 /*
4599  * Setup any per-fs journal parameters now.  We'll do this both on
4600  * initial mount, once the journal has been initialised but before we've
4601  * done any recovery; and again on any subsequent remount.
4602  */
4603 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4604 {
4605         struct ext4_sb_info *sbi = EXT4_SB(sb);
4606
4607         journal->j_commit_interval = sbi->s_commit_interval;
4608         journal->j_min_batch_time = sbi->s_min_batch_time;
4609         journal->j_max_batch_time = sbi->s_max_batch_time;
4610
4611         write_lock(&journal->j_state_lock);
4612         if (test_opt(sb, BARRIER))
4613                 journal->j_flags |= JBD2_BARRIER;
4614         else
4615                 journal->j_flags &= ~JBD2_BARRIER;
4616         if (test_opt(sb, DATA_ERR_ABORT))
4617                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4618         else
4619                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4620         write_unlock(&journal->j_state_lock);
4621 }
4622
4623 static struct inode *ext4_get_journal_inode(struct super_block *sb,
4624                                              unsigned int journal_inum)
4625 {
4626         struct inode *journal_inode;
4627
4628         /*
4629          * Test for the existence of a valid inode on disk.  Bad things
4630          * happen if we iget() an unused inode, as the subsequent iput()
4631          * will try to delete it.
4632          */
4633         journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL);
4634         if (IS_ERR(journal_inode)) {
4635                 ext4_msg(sb, KERN_ERR, "no journal found");
4636                 return NULL;
4637         }
4638         if (!journal_inode->i_nlink) {
4639                 make_bad_inode(journal_inode);
4640                 iput(journal_inode);
4641                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4642                 return NULL;
4643         }
4644
4645         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4646                   journal_inode, journal_inode->i_size);
4647         if (!S_ISREG(journal_inode->i_mode)) {
4648                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4649                 iput(journal_inode);
4650                 return NULL;
4651         }
4652         return journal_inode;
4653 }
4654
4655 static journal_t *ext4_get_journal(struct super_block *sb,
4656                                    unsigned int journal_inum)
4657 {
4658         struct inode *journal_inode;
4659         journal_t *journal;
4660
4661         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
4662                 return NULL;
4663
4664         journal_inode = ext4_get_journal_inode(sb, journal_inum);
4665         if (!journal_inode)
4666                 return NULL;
4667
4668         journal = jbd2_journal_init_inode(journal_inode);
4669         if (!journal) {
4670                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4671                 iput(journal_inode);
4672                 return NULL;
4673         }
4674         journal->j_private = sb;
4675         ext4_init_journal_params(sb, journal);
4676         return journal;
4677 }
4678
4679 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4680                                        dev_t j_dev)
4681 {
4682         struct buffer_head *bh;
4683         journal_t *journal;
4684         ext4_fsblk_t start;
4685         ext4_fsblk_t len;
4686         int hblock, blocksize;
4687         ext4_fsblk_t sb_block;
4688         unsigned long offset;
4689         struct ext4_super_block *es;
4690         struct block_device *bdev;
4691
4692         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
4693                 return NULL;
4694
4695         bdev = ext4_blkdev_get(j_dev, sb);
4696         if (bdev == NULL)
4697                 return NULL;
4698
4699         blocksize = sb->s_blocksize;
4700         hblock = bdev_logical_block_size(bdev);
4701         if (blocksize < hblock) {
4702                 ext4_msg(sb, KERN_ERR,
4703                         "blocksize too small for journal device");
4704                 goto out_bdev;
4705         }
4706
4707         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4708         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4709         set_blocksize(bdev, blocksize);
4710         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4711                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4712                        "external journal");
4713                 goto out_bdev;
4714         }
4715
4716         es = (struct ext4_super_block *) (bh->b_data + offset);
4717         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4718             !(le32_to_cpu(es->s_feature_incompat) &
4719               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4720                 ext4_msg(sb, KERN_ERR, "external journal has "
4721                                         "bad superblock");
4722                 brelse(bh);
4723                 goto out_bdev;
4724         }
4725
4726         if ((le32_to_cpu(es->s_feature_ro_compat) &
4727              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4728             es->s_checksum != ext4_superblock_csum(sb, es)) {
4729                 ext4_msg(sb, KERN_ERR, "external journal has "
4730                                        "corrupt superblock");
4731                 brelse(bh);
4732                 goto out_bdev;
4733         }
4734
4735         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4736                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4737                 brelse(bh);
4738                 goto out_bdev;
4739         }
4740
4741         len = ext4_blocks_count(es);
4742         start = sb_block + 1;
4743         brelse(bh);     /* we're done with the superblock */
4744
4745         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4746                                         start, len, blocksize);
4747         if (!journal) {
4748                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4749                 goto out_bdev;
4750         }
4751         journal->j_private = sb;
4752         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4753         wait_on_buffer(journal->j_sb_buffer);
4754         if (!buffer_uptodate(journal->j_sb_buffer)) {
4755                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4756                 goto out_journal;
4757         }
4758         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4759                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4760                                         "user (unsupported) - %d",
4761                         be32_to_cpu(journal->j_superblock->s_nr_users));
4762                 goto out_journal;
4763         }
4764         EXT4_SB(sb)->journal_bdev = bdev;
4765         ext4_init_journal_params(sb, journal);
4766         return journal;
4767
4768 out_journal:
4769         jbd2_journal_destroy(journal);
4770 out_bdev:
4771         ext4_blkdev_put(bdev);
4772         return NULL;
4773 }
4774
4775 static int ext4_load_journal(struct super_block *sb,
4776                              struct ext4_super_block *es,
4777                              unsigned long journal_devnum)
4778 {
4779         journal_t *journal;
4780         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4781         dev_t journal_dev;
4782         int err = 0;
4783         int really_read_only;
4784
4785         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
4786                 return -EFSCORRUPTED;
4787
4788         if (journal_devnum &&
4789             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4790                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4791                         "numbers have changed");
4792                 journal_dev = new_decode_dev(journal_devnum);
4793         } else
4794                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4795
4796         really_read_only = bdev_read_only(sb->s_bdev);
4797
4798         /*
4799          * Are we loading a blank journal or performing recovery after a
4800          * crash?  For recovery, we need to check in advance whether we
4801          * can get read-write access to the device.
4802          */
4803         if (ext4_has_feature_journal_needs_recovery(sb)) {
4804                 if (sb_rdonly(sb)) {
4805                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4806                                         "required on readonly filesystem");
4807                         if (really_read_only) {
4808                                 ext4_msg(sb, KERN_ERR, "write access "
4809                                         "unavailable, cannot proceed");
4810                                 return -EROFS;
4811                         }
4812                         ext4_msg(sb, KERN_INFO, "write access will "
4813                                "be enabled during recovery");
4814                 }
4815         }
4816
4817         if (journal_inum && journal_dev) {
4818                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4819                        "and inode journals!");
4820                 return -EINVAL;
4821         }
4822
4823         if (journal_inum) {
4824                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4825                         return -EINVAL;
4826         } else {
4827                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4828                         return -EINVAL;
4829         }
4830
4831         if (!(journal->j_flags & JBD2_BARRIER))
4832                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4833
4834         if (!ext4_has_feature_journal_needs_recovery(sb))
4835                 err = jbd2_journal_wipe(journal, !really_read_only);
4836         if (!err) {
4837                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4838                 if (save)
4839                         memcpy(save, ((char *) es) +
4840                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4841                 err = jbd2_journal_load(journal);
4842                 if (save)
4843                         memcpy(((char *) es) + EXT4_S_ERR_START,
4844                                save, EXT4_S_ERR_LEN);
4845                 kfree(save);
4846         }
4847
4848         if (err) {
4849                 ext4_msg(sb, KERN_ERR, "error loading journal");
4850                 jbd2_journal_destroy(journal);
4851                 return err;
4852         }
4853
4854         EXT4_SB(sb)->s_journal = journal;
4855         err = ext4_clear_journal_err(sb, es);
4856         if (err) {
4857                 EXT4_SB(sb)->s_journal = NULL;
4858                 jbd2_journal_destroy(journal);
4859                 return err;
4860         }
4861
4862         if (!really_read_only && journal_devnum &&
4863             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4864                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4865
4866                 /* Make sure we flush the recovery flag to disk. */
4867                 ext4_commit_super(sb, 1);
4868         }
4869
4870         return 0;
4871 }
4872
4873 static int ext4_commit_super(struct super_block *sb, int sync)
4874 {
4875         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4876         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4877         int error = 0;
4878
4879         if (!sbh)
4880                 return -EINVAL;
4881         if (block_device_ejected(sb))
4882                 return -ENODEV;
4883
4884         /*
4885          * If the file system is mounted read-only, don't update the
4886          * superblock write time.  This avoids updating the superblock
4887          * write time when we are mounting the root file system
4888          * read/only but we need to replay the journal; at that point,
4889          * for people who are east of GMT and who make their clock
4890          * tick in localtime for Windows bug-for-bug compatibility,
4891          * the clock is set in the future, and this will cause e2fsck
4892          * to complain and force a full file system check.
4893          */
4894         if (!(sb->s_flags & MS_RDONLY))
4895                 es->s_wtime = cpu_to_le32(get_seconds());
4896         if (sb->s_bdev->bd_part)
4897                 es->s_kbytes_written =
4898                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4899                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4900                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4901         else
4902                 es->s_kbytes_written =
4903                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4904         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4905                 ext4_free_blocks_count_set(es,
4906                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4907                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4908         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4909                 es->s_free_inodes_count =
4910                         cpu_to_le32(percpu_counter_sum_positive(
4911                                 &EXT4_SB(sb)->s_freeinodes_counter));
4912         BUFFER_TRACE(sbh, "marking dirty");
4913         ext4_superblock_csum_set(sb);
4914         if (sync)
4915                 lock_buffer(sbh);
4916         if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
4917                 /*
4918                  * Oh, dear.  A previous attempt to write the
4919                  * superblock failed.  This could happen because the
4920                  * USB device was yanked out.  Or it could happen to
4921                  * be a transient write error and maybe the block will
4922                  * be remapped.  Nothing we can do but to retry the
4923                  * write and hope for the best.
4924                  */
4925                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4926                        "superblock detected");
4927                 clear_buffer_write_io_error(sbh);
4928                 set_buffer_uptodate(sbh);
4929         }
4930         mark_buffer_dirty(sbh);
4931         if (sync) {
4932                 unlock_buffer(sbh);
4933                 error = __sync_dirty_buffer(sbh,
4934                         REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
4935                 if (error)
4936                         return error;
4937
4938                 error = buffer_write_io_error(sbh);
4939                 if (error) {
4940                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4941                                "superblock");
4942                         clear_buffer_write_io_error(sbh);
4943                         set_buffer_uptodate(sbh);
4944                 }
4945         }
4946         return error;
4947 }
4948
4949 /*
4950  * Have we just finished recovery?  If so, and if we are mounting (or
4951  * remounting) the filesystem readonly, then we will end up with a
4952  * consistent fs on disk.  Record that fact.
4953  */
4954 static int ext4_mark_recovery_complete(struct super_block *sb,
4955                                        struct ext4_super_block *es)
4956 {
4957         int err;
4958         journal_t *journal = EXT4_SB(sb)->s_journal;
4959
4960         if (!ext4_has_feature_journal(sb)) {
4961                 if (journal != NULL) {
4962                         ext4_error(sb, "Journal got removed while the fs was "
4963                                    "mounted!");
4964                         return -EFSCORRUPTED;
4965                 }
4966                 return 0;
4967         }
4968         jbd2_journal_lock_updates(journal);
4969         err = jbd2_journal_flush(journal);
4970         if (err < 0)
4971                 goto out;
4972
4973         if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) {
4974                 ext4_clear_feature_journal_needs_recovery(sb);
4975                 ext4_commit_super(sb, 1);
4976         }
4977 out:
4978         jbd2_journal_unlock_updates(journal);
4979         return err;
4980 }
4981
4982 /*
4983  * If we are mounting (or read-write remounting) a filesystem whose journal
4984  * has recorded an error from a previous lifetime, move that error to the
4985  * main filesystem now.
4986  */
4987 static int ext4_clear_journal_err(struct super_block *sb,
4988                                    struct ext4_super_block *es)
4989 {
4990         journal_t *journal;
4991         int j_errno;
4992         const char *errstr;
4993
4994         if (!ext4_has_feature_journal(sb)) {
4995                 ext4_error(sb, "Journal got removed while the fs was mounted!");
4996                 return -EFSCORRUPTED;
4997         }
4998
4999         journal = EXT4_SB(sb)->s_journal;
5000
5001         /*
5002          * Now check for any error status which may have been recorded in the
5003          * journal by a prior ext4_error() or ext4_abort()
5004          */
5005
5006         j_errno = jbd2_journal_errno(journal);
5007         if (j_errno) {
5008                 char nbuf[16];
5009
5010                 errstr = ext4_decode_error(sb, j_errno, nbuf);
5011                 ext4_warning(sb, "Filesystem error recorded "
5012                              "from previous mount: %s", errstr);
5013                 ext4_warning(sb, "Marking fs in need of filesystem check.");
5014
5015                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
5016                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
5017                 ext4_commit_super(sb, 1);
5018
5019                 jbd2_journal_clear_err(journal);
5020                 jbd2_journal_update_sb_errno(journal);
5021         }
5022         return 0;
5023 }
5024
5025 /*
5026  * Force the running and committing transactions to commit,
5027  * and wait on the commit.
5028  */
5029 int ext4_force_commit(struct super_block *sb)
5030 {
5031         journal_t *journal;
5032
5033         if (sb_rdonly(sb))
5034                 return 0;
5035
5036         journal = EXT4_SB(sb)->s_journal;
5037         return ext4_journal_force_commit(journal);
5038 }
5039
5040 static int ext4_sync_fs(struct super_block *sb, int wait)
5041 {
5042         int ret = 0;
5043         tid_t target;
5044         bool needs_barrier = false;
5045         struct ext4_sb_info *sbi = EXT4_SB(sb);
5046
5047         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
5048                 return 0;
5049
5050         trace_ext4_sync_fs(sb, wait);
5051         flush_workqueue(sbi->rsv_conversion_wq);
5052         /*
5053          * Writeback quota in non-journalled quota case - journalled quota has
5054          * no dirty dquots
5055          */
5056         dquot_writeback_dquots(sb, -1);
5057         /*
5058          * Data writeback is possible w/o journal transaction, so barrier must
5059          * being sent at the end of the function. But we can skip it if
5060          * transaction_commit will do it for us.
5061          */
5062         if (sbi->s_journal) {
5063                 target = jbd2_get_latest_transaction(sbi->s_journal);
5064                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
5065                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
5066                         needs_barrier = true;
5067
5068                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
5069                         if (wait)
5070                                 ret = jbd2_log_wait_commit(sbi->s_journal,
5071                                                            target);
5072                 }
5073         } else if (wait && test_opt(sb, BARRIER))
5074                 needs_barrier = true;
5075         if (needs_barrier) {
5076                 int err;
5077                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
5078                 if (!ret)
5079                         ret = err;
5080         }
5081
5082         return ret;
5083 }
5084
5085 /*
5086  * LVM calls this function before a (read-only) snapshot is created.  This
5087  * gives us a chance to flush the journal completely and mark the fs clean.
5088  *
5089  * Note that only this function cannot bring a filesystem to be in a clean
5090  * state independently. It relies on upper layer to stop all data & metadata
5091  * modifications.
5092  */
5093 static int ext4_freeze(struct super_block *sb)
5094 {
5095         int error = 0;
5096         journal_t *journal;
5097
5098         if (sb_rdonly(sb))
5099                 return 0;
5100
5101         journal = EXT4_SB(sb)->s_journal;
5102
5103         if (journal) {
5104                 /* Now we set up the journal barrier. */
5105                 jbd2_journal_lock_updates(journal);
5106
5107                 /*
5108                  * Don't clear the needs_recovery flag if we failed to
5109                  * flush the journal.
5110                  */
5111                 error = jbd2_journal_flush(journal);
5112                 if (error < 0)
5113                         goto out;
5114
5115                 /* Journal blocked and flushed, clear needs_recovery flag. */
5116                 ext4_clear_feature_journal_needs_recovery(sb);
5117         }
5118
5119         error = ext4_commit_super(sb, 1);
5120 out:
5121         if (journal)
5122                 /* we rely on upper layer to stop further updates */
5123                 jbd2_journal_unlock_updates(journal);
5124         return error;
5125 }
5126
5127 /*
5128  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
5129  * flag here, even though the filesystem is not technically dirty yet.
5130  */
5131 static int ext4_unfreeze(struct super_block *sb)
5132 {
5133         if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
5134                 return 0;
5135
5136         if (EXT4_SB(sb)->s_journal) {
5137                 /* Reset the needs_recovery flag before the fs is unlocked. */
5138                 ext4_set_feature_journal_needs_recovery(sb);
5139         }
5140
5141         ext4_commit_super(sb, 1);
5142         return 0;
5143 }
5144
5145 /*
5146  * Structure to save mount options for ext4_remount's benefit
5147  */
5148 struct ext4_mount_options {
5149         unsigned long s_mount_opt;
5150         unsigned long s_mount_opt2;
5151         kuid_t s_resuid;
5152         kgid_t s_resgid;
5153         unsigned long s_commit_interval;
5154         u32 s_min_batch_time, s_max_batch_time;
5155 #ifdef CONFIG_QUOTA
5156         int s_jquota_fmt;
5157         char *s_qf_names[EXT4_MAXQUOTAS];
5158 #endif
5159 };
5160
5161 static int ext4_remount(struct super_block *sb, int *flags, char *data)
5162 {
5163         struct ext4_super_block *es;
5164         struct ext4_sb_info *sbi = EXT4_SB(sb);
5165         unsigned long old_sb_flags;
5166         struct ext4_mount_options old_opts;
5167         int enable_quota = 0;
5168         ext4_group_t g;
5169         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
5170         int err = 0;
5171 #ifdef CONFIG_QUOTA
5172         int i, j;
5173         char *to_free[EXT4_MAXQUOTAS];
5174 #endif
5175         char *orig_data = kstrdup(data, GFP_KERNEL);
5176
5177         /* Store the original options */
5178         old_sb_flags = sb->s_flags;
5179         old_opts.s_mount_opt = sbi->s_mount_opt;
5180         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
5181         old_opts.s_resuid = sbi->s_resuid;
5182         old_opts.s_resgid = sbi->s_resgid;
5183         old_opts.s_commit_interval = sbi->s_commit_interval;
5184         old_opts.s_min_batch_time = sbi->s_min_batch_time;
5185         old_opts.s_max_batch_time = sbi->s_max_batch_time;
5186 #ifdef CONFIG_QUOTA
5187         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
5188         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5189                 if (sbi->s_qf_names[i]) {
5190                         char *qf_name = get_qf_name(sb, sbi, i);
5191
5192                         old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL);
5193                         if (!old_opts.s_qf_names[i]) {
5194                                 for (j = 0; j < i; j++)
5195                                         kfree(old_opts.s_qf_names[j]);
5196                                 kfree(orig_data);
5197                                 return -ENOMEM;
5198                         }
5199                 } else
5200                         old_opts.s_qf_names[i] = NULL;
5201 #endif
5202         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
5203                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
5204
5205         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
5206                 err = -EINVAL;
5207                 goto restore_opts;
5208         }
5209
5210         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
5211             test_opt(sb, JOURNAL_CHECKSUM)) {
5212                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
5213                          "during remount not supported; ignoring");
5214                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
5215         }
5216
5217         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
5218                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
5219                         ext4_msg(sb, KERN_ERR, "can't mount with "
5220                                  "both data=journal and delalloc");
5221                         err = -EINVAL;
5222                         goto restore_opts;
5223                 }
5224                 if (test_opt(sb, DIOREAD_NOLOCK)) {
5225                         ext4_msg(sb, KERN_ERR, "can't mount with "
5226                                  "both data=journal and dioread_nolock");
5227                         err = -EINVAL;
5228                         goto restore_opts;
5229                 }
5230                 if (test_opt(sb, DAX)) {
5231                         ext4_msg(sb, KERN_ERR, "can't mount with "
5232                                  "both data=journal and dax");
5233                         err = -EINVAL;
5234                         goto restore_opts;
5235                 }
5236         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
5237                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5238                         ext4_msg(sb, KERN_ERR, "can't mount with "
5239                                 "journal_async_commit in data=ordered mode");
5240                         err = -EINVAL;
5241                         goto restore_opts;
5242                 }
5243         }
5244
5245         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
5246                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
5247                 err = -EINVAL;
5248                 goto restore_opts;
5249         }
5250
5251         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
5252                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
5253                         "dax flag with busy inodes while remounting");
5254                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
5255         }
5256
5257         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
5258                 ext4_abort(sb, "Abort forced by user");
5259
5260         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
5261                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
5262
5263         es = sbi->s_es;
5264
5265         if (sbi->s_journal) {
5266                 ext4_init_journal_params(sb, sbi->s_journal);
5267                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
5268         }
5269
5270         if (*flags & MS_LAZYTIME)
5271                 sb->s_flags |= MS_LAZYTIME;
5272
5273         if ((bool)(*flags & MS_RDONLY) != sb_rdonly(sb)) {
5274                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
5275                         err = -EROFS;
5276                         goto restore_opts;
5277                 }
5278
5279                 if (*flags & MS_RDONLY) {
5280                         err = sync_filesystem(sb);
5281                         if (err < 0)
5282                                 goto restore_opts;
5283                         err = dquot_suspend(sb, -1);
5284                         if (err < 0)
5285                                 goto restore_opts;
5286
5287                         /*
5288                          * First of all, the unconditional stuff we have to do
5289                          * to disable replay of the journal when we next remount
5290                          */
5291                         sb->s_flags |= MS_RDONLY;
5292
5293                         /*
5294                          * OK, test if we are remounting a valid rw partition
5295                          * readonly, and if so set the rdonly flag and then
5296                          * mark the partition as valid again.
5297                          */
5298                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5299                             (sbi->s_mount_state & EXT4_VALID_FS))
5300                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5301
5302                         if (sbi->s_journal) {
5303                                 /*
5304                                  * We let remount-ro finish even if marking fs
5305                                  * as clean failed...
5306                                  */
5307                                 ext4_mark_recovery_complete(sb, es);
5308                         }
5309                         if (sbi->s_mmp_tsk)
5310                                 kthread_stop(sbi->s_mmp_tsk);
5311                 } else {
5312                         /* Make sure we can mount this feature set readwrite */
5313                         if (ext4_has_feature_readonly(sb) ||
5314                             !ext4_feature_set_ok(sb, 0)) {
5315                                 err = -EROFS;
5316                                 goto restore_opts;
5317                         }
5318                         /*
5319                          * Make sure the group descriptor checksums
5320                          * are sane.  If they aren't, refuse to remount r/w.
5321                          */
5322                         for (g = 0; g < sbi->s_groups_count; g++) {
5323                                 struct ext4_group_desc *gdp =
5324                                         ext4_get_group_desc(sb, g, NULL);
5325
5326                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5327                                         ext4_msg(sb, KERN_ERR,
5328                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5329                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
5330                                                le16_to_cpu(gdp->bg_checksum));
5331                                         err = -EFSBADCRC;
5332                                         goto restore_opts;
5333                                 }
5334                         }
5335
5336                         /*
5337                          * If we have an unprocessed orphan list hanging
5338                          * around from a previously readonly bdev mount,
5339                          * require a full umount/remount for now.
5340                          */
5341                         if (es->s_last_orphan) {
5342                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5343                                        "remount RDWR because of unprocessed "
5344                                        "orphan inode list.  Please "
5345                                        "umount/remount instead");
5346                                 err = -EINVAL;
5347                                 goto restore_opts;
5348                         }
5349
5350                         /*
5351                          * Mounting a RDONLY partition read-write, so reread
5352                          * and store the current valid flag.  (It may have
5353                          * been changed by e2fsck since we originally mounted
5354                          * the partition.)
5355                          */
5356                         if (sbi->s_journal) {
5357                                 err = ext4_clear_journal_err(sb, es);
5358                                 if (err)
5359                                         goto restore_opts;
5360                         }
5361                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5362                         if (!ext4_setup_super(sb, es, 0))
5363                                 sb->s_flags &= ~MS_RDONLY;
5364                         if (ext4_has_feature_mmp(sb))
5365                                 if (ext4_multi_mount_protect(sb,
5366                                                 le64_to_cpu(es->s_mmp_block))) {
5367                                         err = -EROFS;
5368                                         goto restore_opts;
5369                                 }
5370                         enable_quota = 1;
5371                 }
5372         }
5373
5374         /*
5375          * Reinitialize lazy itable initialization thread based on
5376          * current settings
5377          */
5378         if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
5379                 ext4_unregister_li_request(sb);
5380         else {
5381                 ext4_group_t first_not_zeroed;
5382                 first_not_zeroed = ext4_has_uninit_itable(sb);
5383                 ext4_register_li_request(sb, first_not_zeroed);
5384         }
5385
5386         err = ext4_setup_system_zone(sb);
5387         if (err)
5388                 goto restore_opts;
5389
5390         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
5391                 ext4_commit_super(sb, 1);
5392
5393 #ifdef CONFIG_QUOTA
5394         /* Release old quota file names */
5395         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5396                 kfree(old_opts.s_qf_names[i]);
5397         if (enable_quota) {
5398                 if (sb_any_quota_suspended(sb))
5399                         dquot_resume(sb, -1);
5400                 else if (ext4_has_feature_quota(sb)) {
5401                         err = ext4_enable_quotas(sb);
5402                         if (err)
5403                                 goto restore_opts;
5404                 }
5405         }
5406 #endif
5407
5408         *flags = (*flags & ~MS_LAZYTIME) | (sb->s_flags & MS_LAZYTIME);
5409         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5410         kfree(orig_data);
5411         return 0;
5412
5413 restore_opts:
5414         sb->s_flags = old_sb_flags;
5415         sbi->s_mount_opt = old_opts.s_mount_opt;
5416         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5417         sbi->s_resuid = old_opts.s_resuid;
5418         sbi->s_resgid = old_opts.s_resgid;
5419         sbi->s_commit_interval = old_opts.s_commit_interval;
5420         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5421         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5422 #ifdef CONFIG_QUOTA
5423         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5424         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5425                 to_free[i] = get_qf_name(sb, sbi, i);
5426                 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]);
5427         }
5428         synchronize_rcu();
5429         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5430                 kfree(to_free[i]);
5431 #endif
5432         kfree(orig_data);
5433         return err;
5434 }
5435
5436 #ifdef CONFIG_QUOTA
5437 static int ext4_statfs_project(struct super_block *sb,
5438                                kprojid_t projid, struct kstatfs *buf)
5439 {
5440         struct kqid qid;
5441         struct dquot *dquot;
5442         u64 limit;
5443         u64 curblock;
5444
5445         qid = make_kqid_projid(projid);
5446         dquot = dqget(sb, qid);
5447         if (IS_ERR(dquot))
5448                 return PTR_ERR(dquot);
5449         spin_lock(&dquot->dq_dqb_lock);
5450
5451         limit = (dquot->dq_dqb.dqb_bsoftlimit ?
5452                  dquot->dq_dqb.dqb_bsoftlimit :
5453                  dquot->dq_dqb.dqb_bhardlimit) >> sb->s_blocksize_bits;
5454         if (limit && buf->f_blocks > limit) {
5455                 curblock = dquot->dq_dqb.dqb_curspace >> sb->s_blocksize_bits;
5456                 buf->f_blocks = limit;
5457                 buf->f_bfree = buf->f_bavail =
5458                         (buf->f_blocks > curblock) ?
5459                          (buf->f_blocks - curblock) : 0;
5460         }
5461
5462         limit = dquot->dq_dqb.dqb_isoftlimit ?
5463                 dquot->dq_dqb.dqb_isoftlimit :
5464                 dquot->dq_dqb.dqb_ihardlimit;
5465         if (limit && buf->f_files > limit) {
5466                 buf->f_files = limit;
5467                 buf->f_ffree =
5468                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
5469                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
5470         }
5471
5472         spin_unlock(&dquot->dq_dqb_lock);
5473         dqput(dquot);
5474         return 0;
5475 }
5476 #endif
5477
5478 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5479 {
5480         struct super_block *sb = dentry->d_sb;
5481         struct ext4_sb_info *sbi = EXT4_SB(sb);
5482         struct ext4_super_block *es = sbi->s_es;
5483         ext4_fsblk_t overhead = 0, resv_blocks;
5484         u64 fsid;
5485         s64 bfree;
5486         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5487
5488         if (!test_opt(sb, MINIX_DF))
5489                 overhead = sbi->s_overhead;
5490
5491         buf->f_type = EXT4_SUPER_MAGIC;
5492         buf->f_bsize = sb->s_blocksize;
5493         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5494         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5495                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5496         /* prevent underflow in case that few free space is available */
5497         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5498         buf->f_bavail = buf->f_bfree -
5499                         (ext4_r_blocks_count(es) + resv_blocks);
5500         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5501                 buf->f_bavail = 0;
5502         buf->f_files = le32_to_cpu(es->s_inodes_count);
5503         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5504         buf->f_namelen = EXT4_NAME_LEN;
5505         fsid = le64_to_cpup((void *)es->s_uuid) ^
5506                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5507         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5508         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5509
5510 #ifdef CONFIG_QUOTA
5511         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
5512             sb_has_quota_limits_enabled(sb, PRJQUOTA))
5513                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
5514 #endif
5515         return 0;
5516 }
5517
5518
5519 #ifdef CONFIG_QUOTA
5520
5521 /*
5522  * Helper functions so that transaction is started before we acquire dqio_sem
5523  * to keep correct lock ordering of transaction > dqio_sem
5524  */
5525 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5526 {
5527         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5528 }
5529
5530 static int ext4_write_dquot(struct dquot *dquot)
5531 {
5532         int ret, err;
5533         handle_t *handle;
5534         struct inode *inode;
5535
5536         inode = dquot_to_inode(dquot);
5537         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5538                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5539         if (IS_ERR(handle))
5540                 return PTR_ERR(handle);
5541         ret = dquot_commit(dquot);
5542         err = ext4_journal_stop(handle);
5543         if (!ret)
5544                 ret = err;
5545         return ret;
5546 }
5547
5548 static int ext4_acquire_dquot(struct dquot *dquot)
5549 {
5550         int ret, err;
5551         handle_t *handle;
5552
5553         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5554                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5555         if (IS_ERR(handle))
5556                 return PTR_ERR(handle);
5557         ret = dquot_acquire(dquot);
5558         err = ext4_journal_stop(handle);
5559         if (!ret)
5560                 ret = err;
5561         return ret;
5562 }
5563
5564 static int ext4_release_dquot(struct dquot *dquot)
5565 {
5566         int ret, err;
5567         handle_t *handle;
5568
5569         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5570                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5571         if (IS_ERR(handle)) {
5572                 /* Release dquot anyway to avoid endless cycle in dqput() */
5573                 dquot_release(dquot);
5574                 return PTR_ERR(handle);
5575         }
5576         ret = dquot_release(dquot);
5577         err = ext4_journal_stop(handle);
5578         if (!ret)
5579                 ret = err;
5580         return ret;
5581 }
5582
5583 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5584 {
5585         struct super_block *sb = dquot->dq_sb;
5586         struct ext4_sb_info *sbi = EXT4_SB(sb);
5587
5588         /* Are we journaling quotas? */
5589         if (ext4_has_feature_quota(sb) ||
5590             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5591                 dquot_mark_dquot_dirty(dquot);
5592                 return ext4_write_dquot(dquot);
5593         } else {
5594                 return dquot_mark_dquot_dirty(dquot);
5595         }
5596 }
5597
5598 static int ext4_write_info(struct super_block *sb, int type)
5599 {
5600         int ret, err;
5601         handle_t *handle;
5602
5603         /* Data block + inode block */
5604         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5605         if (IS_ERR(handle))
5606                 return PTR_ERR(handle);
5607         ret = dquot_commit_info(sb, type);
5608         err = ext4_journal_stop(handle);
5609         if (!ret)
5610                 ret = err;
5611         return ret;
5612 }
5613
5614 /*
5615  * Turn on quotas during mount time - we need to find
5616  * the quota file and such...
5617  */
5618 static int ext4_quota_on_mount(struct super_block *sb, int type)
5619 {
5620         return dquot_quota_on_mount(sb, get_qf_name(sb, EXT4_SB(sb), type),
5621                                         EXT4_SB(sb)->s_jquota_fmt, type);
5622 }
5623
5624 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5625 {
5626         struct ext4_inode_info *ei = EXT4_I(inode);
5627
5628         /* The first argument of lockdep_set_subclass has to be
5629          * *exactly* the same as the argument to init_rwsem() --- in
5630          * this case, in init_once() --- or lockdep gets unhappy
5631          * because the name of the lock is set using the
5632          * stringification of the argument to init_rwsem().
5633          */
5634         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5635         lockdep_set_subclass(&ei->i_data_sem, subclass);
5636 }
5637
5638 /*
5639  * Standard function to be called on quota_on
5640  */
5641 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5642                          const struct path *path)
5643 {
5644         int err;
5645
5646         if (!test_opt(sb, QUOTA))
5647                 return -EINVAL;
5648
5649         /* Quotafile not on the same filesystem? */
5650         if (path->dentry->d_sb != sb)
5651                 return -EXDEV;
5652
5653         /* Quota already enabled for this file? */
5654         if (IS_NOQUOTA(d_inode(path->dentry)))
5655                 return -EBUSY;
5656
5657         /* Journaling quota? */
5658         if (EXT4_SB(sb)->s_qf_names[type]) {
5659                 /* Quotafile not in fs root? */
5660                 if (path->dentry->d_parent != sb->s_root)
5661                         ext4_msg(sb, KERN_WARNING,
5662                                 "Quota file not on filesystem root. "
5663                                 "Journaled quota will not work");
5664                 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
5665         } else {
5666                 /*
5667                  * Clear the flag just in case mount options changed since
5668                  * last time.
5669                  */
5670                 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
5671         }
5672
5673         /*
5674          * When we journal data on quota file, we have to flush journal to see
5675          * all updates to the file when we bypass pagecache...
5676          */
5677         if (EXT4_SB(sb)->s_journal &&
5678             ext4_should_journal_data(d_inode(path->dentry))) {
5679                 /*
5680                  * We don't need to lock updates but journal_flush() could
5681                  * otherwise be livelocked...
5682                  */
5683                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5684                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5685                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5686                 if (err)
5687                         return err;
5688         }
5689
5690         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5691         err = dquot_quota_on(sb, type, format_id, path);
5692         if (!err) {
5693                 struct inode *inode = d_inode(path->dentry);
5694                 handle_t *handle;
5695
5696                 /*
5697                  * Set inode flags to prevent userspace from messing with quota
5698                  * files. If this fails, we return success anyway since quotas
5699                  * are already enabled and this is not a hard failure.
5700                  */
5701                 inode_lock(inode);
5702                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5703                 if (IS_ERR(handle))
5704                         goto unlock_inode;
5705                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
5706                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
5707                                 S_NOATIME | S_IMMUTABLE);
5708                 ext4_mark_inode_dirty(handle, inode);
5709                 ext4_journal_stop(handle);
5710         unlock_inode:
5711                 inode_unlock(inode);
5712                 if (err)
5713                         dquot_quota_off(sb, type);
5714         }
5715         if (err)
5716                 lockdep_set_quota_inode(path->dentry->d_inode,
5717                                              I_DATA_SEM_NORMAL);
5718         return err;
5719 }
5720
5721 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5722                              unsigned int flags)
5723 {
5724         int err;
5725         struct inode *qf_inode;
5726         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5727                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5728                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5729                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5730         };
5731
5732         BUG_ON(!ext4_has_feature_quota(sb));
5733
5734         if (!qf_inums[type])
5735                 return -EPERM;
5736
5737         qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL);
5738         if (IS_ERR(qf_inode)) {
5739                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5740                 return PTR_ERR(qf_inode);
5741         }
5742
5743         /* Don't account quota for quota files to avoid recursion */
5744         qf_inode->i_flags |= S_NOQUOTA;
5745         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
5746         err = dquot_enable(qf_inode, type, format_id, flags);
5747         if (err)
5748                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
5749         iput(qf_inode);
5750
5751         return err;
5752 }
5753
5754 /* Enable usage tracking for all quota types. */
5755 static int ext4_enable_quotas(struct super_block *sb)
5756 {
5757         int type, err = 0;
5758         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5759                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5760                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5761                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5762         };
5763         bool quota_mopt[EXT4_MAXQUOTAS] = {
5764                 test_opt(sb, USRQUOTA),
5765                 test_opt(sb, GRPQUOTA),
5766                 test_opt(sb, PRJQUOTA),
5767         };
5768
5769         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
5770         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5771                 if (qf_inums[type]) {
5772                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5773                                 DQUOT_USAGE_ENABLED |
5774                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
5775                         if (err) {
5776                                 for (type--; type >= 0; type--)
5777                                         dquot_quota_off(sb, type);
5778
5779                                 ext4_warning(sb,
5780                                         "Failed to enable quota tracking "
5781                                         "(type=%d, err=%d). Please run "
5782                                         "e2fsck to fix.", type, err);
5783                                 return err;
5784                         }
5785                 }
5786         }
5787         return 0;
5788 }
5789
5790 static int ext4_quota_off(struct super_block *sb, int type)
5791 {
5792         struct inode *inode = sb_dqopt(sb)->files[type];
5793         handle_t *handle;
5794         int err;
5795
5796         /* Force all delayed allocation blocks to be allocated.
5797          * Caller already holds s_umount sem */
5798         if (test_opt(sb, DELALLOC))
5799                 sync_filesystem(sb);
5800
5801         if (!inode || !igrab(inode))
5802                 goto out;
5803
5804         err = dquot_quota_off(sb, type);
5805         if (err || ext4_has_feature_quota(sb))
5806                 goto out_put;
5807
5808         inode_lock(inode);
5809         /*
5810          * Update modification times of quota files when userspace can
5811          * start looking at them. If we fail, we return success anyway since
5812          * this is not a hard failure and quotas are already disabled.
5813          */
5814         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5815         if (IS_ERR(handle))
5816                 goto out_unlock;
5817         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
5818         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
5819         inode->i_mtime = inode->i_ctime = current_time(inode);
5820         ext4_mark_inode_dirty(handle, inode);
5821         ext4_journal_stop(handle);
5822 out_unlock:
5823         inode_unlock(inode);
5824 out_put:
5825         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
5826         iput(inode);
5827         return err;
5828 out:
5829         return dquot_quota_off(sb, type);
5830 }
5831
5832 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5833  * acquiring the locks... As quota files are never truncated and quota code
5834  * itself serializes the operations (and no one else should touch the files)
5835  * we don't have to be afraid of races */
5836 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5837                                size_t len, loff_t off)
5838 {
5839         struct inode *inode = sb_dqopt(sb)->files[type];
5840         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5841         int offset = off & (sb->s_blocksize - 1);
5842         int tocopy;
5843         size_t toread;
5844         struct buffer_head *bh;
5845         loff_t i_size = i_size_read(inode);
5846
5847         if (off > i_size)
5848                 return 0;
5849         if (off+len > i_size)
5850                 len = i_size-off;
5851         toread = len;
5852         while (toread > 0) {
5853                 tocopy = sb->s_blocksize - offset < toread ?
5854                                 sb->s_blocksize - offset : toread;
5855                 bh = ext4_bread(NULL, inode, blk, 0);
5856                 if (IS_ERR(bh))
5857                         return PTR_ERR(bh);
5858                 if (!bh)        /* A hole? */
5859                         memset(data, 0, tocopy);
5860                 else
5861                         memcpy(data, bh->b_data+offset, tocopy);
5862                 brelse(bh);
5863                 offset = 0;
5864                 toread -= tocopy;
5865                 data += tocopy;
5866                 blk++;
5867         }
5868         return len;
5869 }
5870
5871 /* Write to quotafile (we know the transaction is already started and has
5872  * enough credits) */
5873 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5874                                 const char *data, size_t len, loff_t off)
5875 {
5876         struct inode *inode = sb_dqopt(sb)->files[type];
5877         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5878         int err, offset = off & (sb->s_blocksize - 1);
5879         int retries = 0;
5880         struct buffer_head *bh;
5881         handle_t *handle = journal_current_handle();
5882
5883         if (!handle) {
5884                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5885                         " cancelled because transaction is not started",
5886                         (unsigned long long)off, (unsigned long long)len);
5887                 return -EIO;
5888         }
5889         /*
5890          * Since we account only one data block in transaction credits,
5891          * then it is impossible to cross a block boundary.
5892          */
5893         if (sb->s_blocksize - offset < len) {
5894                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5895                         " cancelled because not block aligned",
5896                         (unsigned long long)off, (unsigned long long)len);
5897                 return -EIO;
5898         }
5899
5900         do {
5901                 bh = ext4_bread(handle, inode, blk,
5902                                 EXT4_GET_BLOCKS_CREATE |
5903                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5904         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5905                  ext4_should_retry_alloc(inode->i_sb, &retries));
5906         if (IS_ERR(bh))
5907                 return PTR_ERR(bh);
5908         if (!bh)
5909                 goto out;
5910         BUFFER_TRACE(bh, "get write access");
5911         err = ext4_journal_get_write_access(handle, bh);
5912         if (err) {
5913                 brelse(bh);
5914                 return err;
5915         }
5916         lock_buffer(bh);
5917         memcpy(bh->b_data+offset, data, len);
5918         flush_dcache_page(bh->b_page);
5919         unlock_buffer(bh);
5920         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5921         brelse(bh);
5922 out:
5923         if (inode->i_size < off + len) {
5924                 i_size_write(inode, off + len);
5925                 EXT4_I(inode)->i_disksize = inode->i_size;
5926                 ext4_mark_inode_dirty(handle, inode);
5927         }
5928         return len;
5929 }
5930
5931 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid)
5932 {
5933         const struct quota_format_ops   *ops;
5934
5935         if (!sb_has_quota_loaded(sb, qid->type))
5936                 return -ESRCH;
5937         ops = sb_dqopt(sb)->ops[qid->type];
5938         if (!ops || !ops->get_next_id)
5939                 return -ENOSYS;
5940         return dquot_get_next_id(sb, qid);
5941 }
5942 #endif
5943
5944 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5945                        const char *dev_name, void *data)
5946 {
5947         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5948 }
5949
5950 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5951 static inline void register_as_ext2(void)
5952 {
5953         int err = register_filesystem(&ext2_fs_type);
5954         if (err)
5955                 printk(KERN_WARNING
5956                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5957 }
5958
5959 static inline void unregister_as_ext2(void)
5960 {
5961         unregister_filesystem(&ext2_fs_type);
5962 }
5963
5964 static inline int ext2_feature_set_ok(struct super_block *sb)
5965 {
5966         if (ext4_has_unknown_ext2_incompat_features(sb))
5967                 return 0;
5968         if (sb_rdonly(sb))
5969                 return 1;
5970         if (ext4_has_unknown_ext2_ro_compat_features(sb))
5971                 return 0;
5972         return 1;
5973 }
5974 #else
5975 static inline void register_as_ext2(void) { }
5976 static inline void unregister_as_ext2(void) { }
5977 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5978 #endif
5979
5980 static inline void register_as_ext3(void)
5981 {
5982         int err = register_filesystem(&ext3_fs_type);
5983         if (err)
5984                 printk(KERN_WARNING
5985                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5986 }
5987
5988 static inline void unregister_as_ext3(void)
5989 {
5990         unregister_filesystem(&ext3_fs_type);
5991 }
5992
5993 static inline int ext3_feature_set_ok(struct super_block *sb)
5994 {
5995         if (ext4_has_unknown_ext3_incompat_features(sb))
5996                 return 0;
5997         if (!ext4_has_feature_journal(sb))
5998                 return 0;
5999         if (sb_rdonly(sb))
6000                 return 1;
6001         if (ext4_has_unknown_ext3_ro_compat_features(sb))
6002                 return 0;
6003         return 1;
6004 }
6005
6006 static struct file_system_type ext4_fs_type = {
6007         .owner          = THIS_MODULE,
6008         .name           = "ext4",
6009         .mount          = ext4_mount,
6010         .kill_sb        = kill_block_super,
6011         .fs_flags       = FS_REQUIRES_DEV,
6012 };
6013 MODULE_ALIAS_FS("ext4");
6014
6015 /* Shared across all ext4 file systems */
6016 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
6017
6018 static int __init ext4_init_fs(void)
6019 {
6020         int i, err;
6021
6022         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
6023         ext4_li_info = NULL;
6024         mutex_init(&ext4_li_mtx);
6025
6026         /* Build-time check for flags consistency */
6027         ext4_check_flag_values();
6028
6029         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
6030                 init_waitqueue_head(&ext4__ioend_wq[i]);
6031
6032         err = ext4_init_es();
6033         if (err)
6034                 return err;
6035
6036         err = ext4_init_pageio();
6037         if (err)
6038                 goto out5;
6039
6040         err = ext4_init_system_zone();
6041         if (err)
6042                 goto out4;
6043
6044         err = ext4_init_sysfs();
6045         if (err)
6046                 goto out3;
6047
6048         err = ext4_init_mballoc();
6049         if (err)
6050                 goto out2;
6051         err = init_inodecache();
6052         if (err)
6053                 goto out1;
6054         register_as_ext3();
6055         register_as_ext2();
6056         err = register_filesystem(&ext4_fs_type);
6057         if (err)
6058                 goto out;
6059
6060         return 0;
6061 out:
6062         unregister_as_ext2();
6063         unregister_as_ext3();
6064         destroy_inodecache();
6065 out1:
6066         ext4_exit_mballoc();
6067 out2:
6068         ext4_exit_sysfs();
6069 out3:
6070         ext4_exit_system_zone();
6071 out4:
6072         ext4_exit_pageio();
6073 out5:
6074         ext4_exit_es();
6075
6076         return err;
6077 }
6078
6079 static void __exit ext4_exit_fs(void)
6080 {
6081         ext4_destroy_lazyinit_thread();
6082         unregister_as_ext2();
6083         unregister_as_ext3();
6084         unregister_filesystem(&ext4_fs_type);
6085         destroy_inodecache();
6086         ext4_exit_mballoc();
6087         ext4_exit_sysfs();
6088         ext4_exit_system_zone();
6089         ext4_exit_pageio();
6090         ext4_exit_es();
6091 }
6092
6093 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
6094 MODULE_DESCRIPTION("Fourth Extended Filesystem");
6095 MODULE_LICENSE("GPL");
6096 MODULE_SOFTDEP("pre: crc32c");
6097 module_init(ext4_init_fs)
6098 module_exit(ext4_exit_fs)