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/cleancache.h>
  41 #include <asm/uaccess.h>
  42
  43 #include <linux/kthread.h>
  44 #include <linux/freezer.h>
  45
  46 #include "ext4.h"
  47 #include "ext4_extents.h"       /* Needed for trace points definition */
  48 #include "ext4_jbd2.h"
  49 #include "xattr.h"
  50 #include "acl.h"
  51 #include "mballoc.h"
  52
  53 #define CREATE_TRACE_POINTS
  54 #include <trace/events/ext4.h>
  55
  56 static struct ext4_lazy_init *ext4_li_info;
  57 static struct mutex ext4_li_mtx;
  58 static struct ratelimit_state ext4_mount_msg_ratelimit;
  59
  60 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  61                              unsigned long journal_devnum);
  62 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  63 static int ext4_commit_super(struct super_block *sb, int sync);
  64 static void ext4_mark_recovery_complete(struct super_block *sb,
  65                                         struct ext4_super_block *es);
  66 static void ext4_clear_journal_err(struct super_block *sb,
  67                                    struct ext4_super_block *es);
  68 static int ext4_sync_fs(struct super_block *sb, int wait);
  69 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  70 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  71 static int ext4_unfreeze(struct super_block *sb);
  72 static int ext4_freeze(struct super_block *sb);
  73 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  74                        const char *dev_name, void *data);
  75 static inline int ext2_feature_set_ok(struct super_block *sb);
  76 static inline int ext3_feature_set_ok(struct super_block *sb);
  77 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  78 static void ext4_destroy_lazyinit_thread(void);
  79 static void ext4_unregister_li_request(struct super_block *sb);
  80 static void ext4_clear_request_list(void);
  81
  82 /*
  83  * Lock ordering
  84  *
  85  * Note the difference between i_mmap_sem (EXT4_I(inode)->i_mmap_sem) and
  86  * i_mmap_rwsem (inode->i_mmap_rwsem)!
  87  *
  88  * page fault path:
  89  * mmap_sem -> sb_start_pagefault -> i_mmap_sem (r) -> transaction start ->
  90  *   page lock -> i_data_sem (rw)
  91  *
  92  * buffered write path:
  93  * sb_start_write -> i_mutex -> mmap_sem
  94  * sb_start_write -> i_mutex -> transaction start -> page lock ->
  95  *   i_data_sem (rw)
  96  *
  97  * truncate:
  98  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (w) -> i_mmap_sem (w) ->
  99  *   i_mmap_rwsem (w) -> page lock
 100  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (w) -> i_mmap_sem (w) ->
 101  *   transaction start -> i_data_sem (rw)
 102  *
 103  * direct IO:
 104  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (r) -> mmap_sem
 105  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (r) ->
 106  *   transaction start -> i_data_sem (rw)
 107  *
 108  * writepages:
 109  * transaction start -> page lock(s) -> i_data_sem (rw)
 110  */
 111
 112 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
 113 static struct file_system_type ext2_fs_type = {
 114         .owner          = THIS_MODULE,
 115         .name           = "ext2",
 116         .mount          = ext4_mount,
 117         .kill_sb        = kill_block_super,
 118         .fs_flags       = FS_REQUIRES_DEV,
 119 };
 120 MODULE_ALIAS_FS("ext2");
 121 MODULE_ALIAS("ext2");
 122 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
 123 #else
 124 #define IS_EXT2_SB(sb) (0)
 125 #endif
 126
 127
 128 static struct file_system_type ext3_fs_type = {
 129         .owner          = THIS_MODULE,
 130         .name           = "ext3",
 131         .mount          = ext4_mount,
 132         .kill_sb        = kill_block_super,
 133         .fs_flags       = FS_REQUIRES_DEV,
 134 };
 135 MODULE_ALIAS_FS("ext3");
 136 MODULE_ALIAS("ext3");
 137 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 138
 139 static int ext4_verify_csum_type(struct super_block *sb,
 140                                  struct ext4_super_block *es)
 141 {
 142         if (!ext4_has_feature_metadata_csum(sb))
 143                 return 1;
 144
 145         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 146 }
 147
 148 static __le32 ext4_superblock_csum(struct super_block *sb,
 149                                    struct ext4_super_block *es)
 150 {
 151         struct ext4_sb_info *sbi = EXT4_SB(sb);
 152         int offset = offsetof(struct ext4_super_block, s_checksum);
 153         __u32 csum;
 154
 155         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 156
 157         return cpu_to_le32(csum);
 158 }
 159
 160 static int ext4_superblock_csum_verify(struct super_block *sb,
 161                                        struct ext4_super_block *es)
 162 {
 163         if (!ext4_has_metadata_csum(sb))
 164                 return 1;
 165
 166         return es->s_checksum == ext4_superblock_csum(sb, es);
 167 }
 168
 169 void ext4_superblock_csum_set(struct super_block *sb)
 170 {
 171         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 172
 173         if (!ext4_has_metadata_csum(sb))
 174                 return;
 175
 176         es->s_checksum = ext4_superblock_csum(sb, es);
 177 }
 178
 179 void *ext4_kvmalloc(size_t size, gfp_t flags)
 180 {
 181         void *ret;
 182
 183         ret = kmalloc(size, flags | __GFP_NOWARN);
 184         if (!ret)
 185                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 186         return ret;
 187 }
 188
 189 void *ext4_kvzalloc(size_t size, gfp_t flags)
 190 {
 191         void *ret;
 192
 193         ret = kzalloc(size, flags | __GFP_NOWARN);
 194         if (!ret)
 195                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 196         return ret;
 197 }
 198
 199 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 200                                struct ext4_group_desc *bg)
 201 {
 202         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 203                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 204                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 205 }
 206
 207 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 208                                struct ext4_group_desc *bg)
 209 {
 210         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 211                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 212                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 213 }
 214
 215 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 216                               struct ext4_group_desc *bg)
 217 {
 218         return le32_to_cpu(bg->bg_inode_table_lo) |
 219                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 220                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 221 }
 222
 223 __u32 ext4_free_group_clusters(struct super_block *sb,
 224                                struct ext4_group_desc *bg)
 225 {
 226         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 227                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 228                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 229 }
 230
 231 __u32 ext4_free_inodes_count(struct super_block *sb,
 232                               struct ext4_group_desc *bg)
 233 {
 234         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 235                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 236                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 237 }
 238
 239 __u32 ext4_used_dirs_count(struct super_block *sb,
 240                               struct ext4_group_desc *bg)
 241 {
 242         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 243                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 244                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 245 }
 246
 247 __u32 ext4_itable_unused_count(struct super_block *sb,
 248                               struct ext4_group_desc *bg)
 249 {
 250         return le16_to_cpu(bg->bg_itable_unused_lo) |
 251                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 252                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 253 }
 254
 255 void ext4_block_bitmap_set(struct super_block *sb,
 256                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 257 {
 258         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 259         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 260                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 261 }
 262
 263 void ext4_inode_bitmap_set(struct super_block *sb,
 264                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 265 {
 266         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 267         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 268                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 269 }
 270
 271 void ext4_inode_table_set(struct super_block *sb,
 272                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 273 {
 274         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 275         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 276                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 277 }
 278
 279 void ext4_free_group_clusters_set(struct super_block *sb,
 280                                   struct ext4_group_desc *bg, __u32 count)
 281 {
 282         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 283         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 284                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 285 }
 286
 287 void ext4_free_inodes_set(struct super_block *sb,
 288                           struct ext4_group_desc *bg, __u32 count)
 289 {
 290         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 291         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 292                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 293 }
 294
 295 void ext4_used_dirs_set(struct super_block *sb,
 296                           struct ext4_group_desc *bg, __u32 count)
 297 {
 298         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 299         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 300                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 301 }
 302
 303 void ext4_itable_unused_set(struct super_block *sb,
 304                           struct ext4_group_desc *bg, __u32 count)
 305 {
 306         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 307         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 308                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 309 }
 310
 311
 312 static void __save_error_info(struct super_block *sb, const char *func,
 313                             unsigned int line)
 314 {
 315         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 316
 317         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 318         if (bdev_read_only(sb->s_bdev))
 319                 return;
 320         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 321         es->s_last_error_time = cpu_to_le32(get_seconds());
 322         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 323         es->s_last_error_line = cpu_to_le32(line);
 324         if (!es->s_first_error_time) {
 325                 es->s_first_error_time = es->s_last_error_time;
 326                 strncpy(es->s_first_error_func, func,
 327                         sizeof(es->s_first_error_func));
 328                 es->s_first_error_line = cpu_to_le32(line);
 329                 es->s_first_error_ino = es->s_last_error_ino;
 330                 es->s_first_error_block = es->s_last_error_block;
 331         }
 332         /*
 333          * Start the daily error reporting function if it hasn't been
 334          * started already
 335          */
 336         if (!es->s_error_count)
 337                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 338         le32_add_cpu(&es->s_error_count, 1);
 339 }
 340
 341 static void save_error_info(struct super_block *sb, const char *func,
 342                             unsigned int line)
 343 {
 344         __save_error_info(sb, func, line);
 345         ext4_commit_super(sb, 1);
 346 }
 347
 348 /*
 349  * The del_gendisk() function uninitializes the disk-specific data
 350  * structures, including the bdi structure, without telling anyone
 351  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 352  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 353  * This is a kludge to prevent these oops until we can put in a proper
 354  * hook in del_gendisk() to inform the VFS and file system layers.
 355  */
 356 static int block_device_ejected(struct super_block *sb)
 357 {
 358         struct inode *bd_inode = sb->s_bdev->bd_inode;
 359         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 360
 361         return bdi->dev == NULL;
 362 }
 363
 364 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 365 {
 366         struct super_block              *sb = journal->j_private;
 367         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 368         int                             error = is_journal_aborted(journal);
 369         struct ext4_journal_cb_entry    *jce;
 370
 371         BUG_ON(txn->t_state == T_FINISHED);
 372         spin_lock(&sbi->s_md_lock);
 373         while (!list_empty(&txn->t_private_list)) {
 374                 jce = list_entry(txn->t_private_list.next,
 375                                  struct ext4_journal_cb_entry, jce_list);
 376                 list_del_init(&jce->jce_list);
 377                 spin_unlock(&sbi->s_md_lock);
 378                 jce->jce_func(sb, jce, error);
 379                 spin_lock(&sbi->s_md_lock);
 380         }
 381         spin_unlock(&sbi->s_md_lock);
 382 }
 383
 384 /* Deal with the reporting of failure conditions on a filesystem such as
 385  * inconsistencies detected or read IO failures.
 386  *
 387  * On ext2, we can store the error state of the filesystem in the
 388  * superblock.  That is not possible on ext4, because we may have other
 389  * write ordering constraints on the superblock which prevent us from
 390  * writing it out straight away; and given that the journal is about to
 391  * be aborted, we can't rely on the current, or future, transactions to
 392  * write out the superblock safely.
 393  *
 394  * We'll just use the jbd2_journal_abort() error code to record an error in
 395  * the journal instead.  On recovery, the journal will complain about
 396  * that error until we've noted it down and cleared it.
 397  */
 398
 399 static void ext4_handle_error(struct super_block *sb)
 400 {
 401         if (sb->s_flags & MS_RDONLY)
 402                 return;
 403
 404         if (!test_opt(sb, ERRORS_CONT)) {
 405                 journal_t *journal = EXT4_SB(sb)->s_journal;
 406
 407                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 408                 if (journal)
 409                         jbd2_journal_abort(journal, -EIO);
 410         }
 411         if (test_opt(sb, ERRORS_RO)) {
 412                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 413                 /*
 414                  * Make sure updated value of ->s_mount_flags will be visible
 415                  * before ->s_flags update
 416                  */
 417                 smp_wmb();
 418                 sb->s_flags |= MS_RDONLY;
 419         }
 420         if (test_opt(sb, ERRORS_PANIC)) {
 421                 if (EXT4_SB(sb)->s_journal &&
 422                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 423                         return;
 424                 panic("EXT4-fs (device %s): panic forced after error\n",
 425                         sb->s_id);
 426         }
 427 }
 428
 429 #define ext4_error_ratelimit(sb)                                        \
 430                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 431                              "EXT4-fs error")
 432
 433 void __ext4_error(struct super_block *sb, const char *function,
 434                   unsigned int line, const char *fmt, ...)
 435 {
 436         struct va_format vaf;
 437         va_list args;
 438
 439         if (ext4_error_ratelimit(sb)) {
 440                 va_start(args, fmt);
 441                 vaf.fmt = fmt;
 442                 vaf.va = &args;
 443                 printk(KERN_CRIT
 444                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 445                        sb->s_id, function, line, current->comm, &vaf);
 446                 va_end(args);
 447         }
 448         save_error_info(sb, function, line);
 449         ext4_handle_error(sb);
 450 }
 451
 452 void __ext4_error_inode(struct inode *inode, const char *function,
 453                         unsigned int line, ext4_fsblk_t block,
 454                         const char *fmt, ...)
 455 {
 456         va_list args;
 457         struct va_format vaf;
 458         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 459
 460         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 461         es->s_last_error_block = cpu_to_le64(block);
 462         if (ext4_error_ratelimit(inode->i_sb)) {
 463                 va_start(args, fmt);
 464                 vaf.fmt = fmt;
 465                 vaf.va = &args;
 466                 if (block)
 467                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 468                                "inode #%lu: block %llu: comm %s: %pV\n",
 469                                inode->i_sb->s_id, function, line, inode->i_ino,
 470                                block, current->comm, &vaf);
 471                 else
 472                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 473                                "inode #%lu: comm %s: %pV\n",
 474                                inode->i_sb->s_id, function, line, inode->i_ino,
 475                                current->comm, &vaf);
 476                 va_end(args);
 477         }
 478         save_error_info(inode->i_sb, function, line);
 479         ext4_handle_error(inode->i_sb);
 480 }
 481
 482 void __ext4_error_file(struct file *file, const char *function,
 483                        unsigned int line, ext4_fsblk_t block,
 484                        const char *fmt, ...)
 485 {
 486         va_list args;
 487         struct va_format vaf;
 488         struct ext4_super_block *es;
 489         struct inode *inode = file_inode(file);
 490         char pathname[80], *path;
 491
 492         es = EXT4_SB(inode->i_sb)->s_es;
 493         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 494         if (ext4_error_ratelimit(inode->i_sb)) {
 495                 path = file_path(file, pathname, sizeof(pathname));
 496                 if (IS_ERR(path))
 497                         path = "(unknown)";
 498                 va_start(args, fmt);
 499                 vaf.fmt = fmt;
 500                 vaf.va = &args;
 501                 if (block)
 502                         printk(KERN_CRIT
 503                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 504                                "block %llu: comm %s: path %s: %pV\n",
 505                                inode->i_sb->s_id, function, line, inode->i_ino,
 506                                block, current->comm, path, &vaf);
 507                 else
 508                         printk(KERN_CRIT
 509                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 510                                "comm %s: path %s: %pV\n",
 511                                inode->i_sb->s_id, function, line, inode->i_ino,
 512                                current->comm, path, &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 const char *ext4_decode_error(struct super_block *sb, int errno,
 520                               char nbuf[16])
 521 {
 522         char *errstr = NULL;
 523
 524         switch (errno) {
 525         case -EFSCORRUPTED:
 526                 errstr = "Corrupt filesystem";
 527                 break;
 528         case -EFSBADCRC:
 529                 errstr = "Filesystem failed CRC";
 530                 break;
 531         case -EIO:
 532                 errstr = "IO failure";
 533                 break;
 534         case -ENOMEM:
 535                 errstr = "Out of memory";
 536                 break;
 537         case -EROFS:
 538                 if (!sb || (EXT4_SB(sb)->s_journal &&
 539                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 540                         errstr = "Journal has aborted";
 541                 else
 542                         errstr = "Readonly filesystem";
 543                 break;
 544         default:
 545                 /* If the caller passed in an extra buffer for unknown
 546                  * errors, textualise them now.  Else we just return
 547                  * NULL. */
 548                 if (nbuf) {
 549                         /* Check for truncated error codes... */
 550                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 551                                 errstr = nbuf;
 552                 }
 553                 break;
 554         }
 555
 556         return errstr;
 557 }
 558
 559 /* __ext4_std_error decodes expected errors from journaling functions
 560  * automatically and invokes the appropriate error response.  */
 561
 562 void __ext4_std_error(struct super_block *sb, const char *function,
 563                       unsigned int line, int errno)
 564 {
 565         char nbuf[16];
 566         const char *errstr;
 567
 568         /* Special case: if the error is EROFS, and we're not already
 569          * inside a transaction, then there's really no point in logging
 570          * an error. */
 571         if (errno == -EROFS && journal_current_handle() == NULL &&
 572             (sb->s_flags & MS_RDONLY))
 573                 return;
 574
 575         if (ext4_error_ratelimit(sb)) {
 576                 errstr = ext4_decode_error(sb, errno, nbuf);
 577                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 578                        sb->s_id, function, line, errstr);
 579         }
 580
 581         save_error_info(sb, function, line);
 582         ext4_handle_error(sb);
 583 }
 584
 585 /*
 586  * ext4_abort is a much stronger failure handler than ext4_error.  The
 587  * abort function may be used to deal with unrecoverable failures such
 588  * as journal IO errors or ENOMEM at a critical moment in log management.
 589  *
 590  * We unconditionally force the filesystem into an ABORT|READONLY state,
 591  * unless the error response on the fs has been set to panic in which
 592  * case we take the easy way out and panic immediately.
 593  */
 594
 595 void __ext4_abort(struct super_block *sb, const char *function,
 596                 unsigned int line, const char *fmt, ...)
 597 {
 598         va_list args;
 599
 600         save_error_info(sb, function, line);
 601         va_start(args, fmt);
 602         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 603                function, line);
 604         vprintk(fmt, args);
 605         printk("\n");
 606         va_end(args);
 607
 608         if ((sb->s_flags & MS_RDONLY) == 0) {
 609                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 610                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 611                 /*
 612                  * Make sure updated value of ->s_mount_flags will be visible
 613                  * before ->s_flags update
 614                  */
 615                 smp_wmb();
 616                 sb->s_flags |= MS_RDONLY;
 617                 if (EXT4_SB(sb)->s_journal)
 618                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 619                 save_error_info(sb, function, line);
 620         }
 621         if (test_opt(sb, ERRORS_PANIC)) {
 622                 if (EXT4_SB(sb)->s_journal &&
 623                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 624                         return;
 625                 panic("EXT4-fs panic from previous error\n");
 626         }
 627 }
 628
 629 void __ext4_msg(struct super_block *sb,
 630                 const char *prefix, const char *fmt, ...)
 631 {
 632         struct va_format vaf;
 633         va_list args;
 634
 635         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 636                 return;
 637
 638         va_start(args, fmt);
 639         vaf.fmt = fmt;
 640         vaf.va = &args;
 641         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 642         va_end(args);
 643 }
 644
 645 #define ext4_warning_ratelimit(sb)                                      \
 646                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
 647                              "EXT4-fs warning")
 648
 649 void __ext4_warning(struct super_block *sb, const char *function,
 650                     unsigned int line, const char *fmt, ...)
 651 {
 652         struct va_format vaf;
 653         va_list args;
 654
 655         if (!ext4_warning_ratelimit(sb))
 656                 return;
 657
 658         va_start(args, fmt);
 659         vaf.fmt = fmt;
 660         vaf.va = &args;
 661         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 662                sb->s_id, function, line, &vaf);
 663         va_end(args);
 664 }
 665
 666 void __ext4_warning_inode(const struct inode *inode, const char *function,
 667                           unsigned int line, const char *fmt, ...)
 668 {
 669         struct va_format vaf;
 670         va_list args;
 671
 672         if (!ext4_warning_ratelimit(inode->i_sb))
 673                 return;
 674
 675         va_start(args, fmt);
 676         vaf.fmt = fmt;
 677         vaf.va = &args;
 678         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 679                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 680                function, line, inode->i_ino, current->comm, &vaf);
 681         va_end(args);
 682 }
 683
 684 void __ext4_grp_locked_error(const char *function, unsigned int line,
 685                              struct super_block *sb, ext4_group_t grp,
 686                              unsigned long ino, ext4_fsblk_t block,
 687                              const char *fmt, ...)
 688 __releases(bitlock)
 689 __acquires(bitlock)
 690 {
 691         struct va_format vaf;
 692         va_list args;
 693         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 694
 695         es->s_last_error_ino = cpu_to_le32(ino);
 696         es->s_last_error_block = cpu_to_le64(block);
 697         __save_error_info(sb, function, line);
 698
 699         if (ext4_error_ratelimit(sb)) {
 700                 va_start(args, fmt);
 701                 vaf.fmt = fmt;
 702                 vaf.va = &args;
 703                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 704                        sb->s_id, function, line, grp);
 705                 if (ino)
 706                         printk(KERN_CONT "inode %lu: ", ino);
 707                 if (block)
 708                         printk(KERN_CONT "block %llu:",
 709                                (unsigned long long) block);
 710                 printk(KERN_CONT "%pV\n", &vaf);
 711                 va_end(args);
 712         }
 713
 714         if (test_opt(sb, ERRORS_CONT)) {
 715                 ext4_commit_super(sb, 0);
 716                 return;
 717         }
 718
 719         ext4_unlock_group(sb, grp);
 720         ext4_handle_error(sb);
 721         /*
 722          * We only get here in the ERRORS_RO case; relocking the group
 723          * may be dangerous, but nothing bad will happen since the
 724          * filesystem will have already been marked read/only and the
 725          * journal has been aborted.  We return 1 as a hint to callers
 726          * who might what to use the return value from
 727          * ext4_grp_locked_error() to distinguish between the
 728          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 729          * aggressively from the ext4 function in question, with a
 730          * more appropriate error code.
 731          */
 732         ext4_lock_group(sb, grp);
 733         return;
 734 }
 735
 736 void ext4_update_dynamic_rev(struct super_block *sb)
 737 {
 738         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 739
 740         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 741                 return;
 742
 743         ext4_warning(sb,
 744                      "updating to rev %d because of new feature flag, "
 745                      "running e2fsck is recommended",
 746                      EXT4_DYNAMIC_REV);
 747
 748         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 749         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 750         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 751         /* leave es->s_feature_*compat flags alone */
 752         /* es->s_uuid will be set by e2fsck if empty */
 753
 754         /*
 755          * The rest of the superblock fields should be zero, and if not it
 756          * means they are likely already in use, so leave them alone.  We
 757          * can leave it up to e2fsck to clean up any inconsistencies there.
 758          */
 759 }
 760
 761 /*
 762  * Open the external journal device
 763  */
 764 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 765 {
 766         struct block_device *bdev;
 767         char b[BDEVNAME_SIZE];
 768
 769         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 770         if (IS_ERR(bdev))
 771                 goto fail;
 772         return bdev;
 773
 774 fail:
 775         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 776                         __bdevname(dev, b), PTR_ERR(bdev));
 777         return NULL;
 778 }
 779
 780 /*
 781  * Release the journal device
 782  */
 783 static void ext4_blkdev_put(struct block_device *bdev)
 784 {
 785         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 786 }
 787
 788 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 789 {
 790         struct block_device *bdev;
 791         bdev = sbi->journal_bdev;
 792         if (bdev) {
 793                 ext4_blkdev_put(bdev);
 794                 sbi->journal_bdev = NULL;
 795         }
 796 }
 797
 798 static inline struct inode *orphan_list_entry(struct list_head *l)
 799 {
 800         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 801 }
 802
 803 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 804 {
 805         struct list_head *l;
 806
 807         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 808                  le32_to_cpu(sbi->s_es->s_last_orphan));
 809
 810         printk(KERN_ERR "sb_info orphan list:\n");
 811         list_for_each(l, &sbi->s_orphan) {
 812                 struct inode *inode = orphan_list_entry(l);
 813                 printk(KERN_ERR "  "
 814                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 815                        inode->i_sb->s_id, inode->i_ino, inode,
 816                        inode->i_mode, inode->i_nlink,
 817                        NEXT_ORPHAN(inode));
 818         }
 819 }
 820
 821 static void ext4_put_super(struct super_block *sb)
 822 {
 823         struct ext4_sb_info *sbi = EXT4_SB(sb);
 824         struct ext4_super_block *es = sbi->s_es;
 825         int i, err;
 826
 827         ext4_unregister_li_request(sb);
 828         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 829
 830         flush_workqueue(sbi->rsv_conversion_wq);
 831         destroy_workqueue(sbi->rsv_conversion_wq);
 832
 833         if (sbi->s_journal) {
 834                 err = jbd2_journal_destroy(sbi->s_journal);
 835                 sbi->s_journal = NULL;
 836                 if (err < 0)
 837                         ext4_abort(sb, "Couldn't clean up the journal");
 838         }
 839
 840         ext4_unregister_sysfs(sb);
 841         ext4_es_unregister_shrinker(sbi);
 842         del_timer_sync(&sbi->s_err_report);
 843         ext4_release_system_zone(sb);
 844         ext4_mb_release(sb);
 845         ext4_ext_release(sb);
 846
 847         if (!(sb->s_flags & MS_RDONLY)) {
 848                 ext4_clear_feature_journal_needs_recovery(sb);
 849                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 850         }
 851         if (!(sb->s_flags & MS_RDONLY))
 852                 ext4_commit_super(sb, 1);
 853
 854         for (i = 0; i < sbi->s_gdb_count; i++)
 855                 brelse(sbi->s_group_desc[i]);
 856         kvfree(sbi->s_group_desc);
 857         kvfree(sbi->s_flex_groups);
 858         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 859         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 860         percpu_counter_destroy(&sbi->s_dirs_counter);
 861         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 862         brelse(sbi->s_sbh);
 863 #ifdef CONFIG_QUOTA
 864         for (i = 0; i < EXT4_MAXQUOTAS; i++)
 865                 kfree(sbi->s_qf_names[i]);
 866 #endif
 867
 868         /* Debugging code just in case the in-memory inode orphan list
 869          * isn't empty.  The on-disk one can be non-empty if we've
 870          * detected an error and taken the fs readonly, but the
 871          * in-memory list had better be clean by this point. */
 872         if (!list_empty(&sbi->s_orphan))
 873                 dump_orphan_list(sb, sbi);
 874         J_ASSERT(list_empty(&sbi->s_orphan));
 875
 876         sync_blockdev(sb->s_bdev);
 877         invalidate_bdev(sb->s_bdev);
 878         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 879                 /*
 880                  * Invalidate the journal device's buffers.  We don't want them
 881                  * floating about in memory - the physical journal device may
 882                  * hotswapped, and it breaks the `ro-after' testing code.
 883                  */
 884                 sync_blockdev(sbi->journal_bdev);
 885                 invalidate_bdev(sbi->journal_bdev);
 886                 ext4_blkdev_remove(sbi);
 887         }
 888         if (sbi->s_mb_cache) {
 889                 ext4_xattr_destroy_cache(sbi->s_mb_cache);
 890                 sbi->s_mb_cache = NULL;
 891         }
 892         if (sbi->s_mmp_tsk)
 893                 kthread_stop(sbi->s_mmp_tsk);
 894         sb->s_fs_info = NULL;
 895         /*
 896          * Now that we are completely done shutting down the
 897          * superblock, we need to actually destroy the kobject.
 898          */
 899         kobject_put(&sbi->s_kobj);
 900         wait_for_completion(&sbi->s_kobj_unregister);
 901         if (sbi->s_chksum_driver)
 902                 crypto_free_shash(sbi->s_chksum_driver);
 903         kfree(sbi->s_blockgroup_lock);
 904         kfree(sbi);
 905 }
 906
 907 static struct kmem_cache *ext4_inode_cachep;
 908
 909 /*
 910  * Called inside transaction, so use GFP_NOFS
 911  */
 912 static struct inode *ext4_alloc_inode(struct super_block *sb)
 913 {
 914         struct ext4_inode_info *ei;
 915
 916         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 917         if (!ei)
 918                 return NULL;
 919
 920         ei->vfs_inode.i_version = 1;
 921         spin_lock_init(&ei->i_raw_lock);
 922         INIT_LIST_HEAD(&ei->i_prealloc_list);
 923         spin_lock_init(&ei->i_prealloc_lock);
 924         ext4_es_init_tree(&ei->i_es_tree);
 925         rwlock_init(&ei->i_es_lock);
 926         INIT_LIST_HEAD(&ei->i_es_list);
 927         ei->i_es_all_nr = 0;
 928         ei->i_es_shk_nr = 0;
 929         ei->i_es_shrink_lblk = 0;
 930         ei->i_reserved_data_blocks = 0;
 931         ei->i_reserved_meta_blocks = 0;
 932         ei->i_allocated_meta_blocks = 0;
 933         ei->i_da_metadata_calc_len = 0;
 934         ei->i_da_metadata_calc_last_lblock = 0;
 935         spin_lock_init(&(ei->i_block_reservation_lock));
 936 #ifdef CONFIG_QUOTA
 937         ei->i_reserved_quota = 0;
 938         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
 939 #endif
 940         ei->jinode = NULL;
 941         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
 942         spin_lock_init(&ei->i_completed_io_lock);
 943         ei->i_sync_tid = 0;
 944         ei->i_datasync_tid = 0;
 945         atomic_set(&ei->i_unwritten, 0);
 946         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
 947 #ifdef CONFIG_EXT4_FS_ENCRYPTION
 948         ei->i_crypt_info = NULL;
 949 #endif
 950         return &ei->vfs_inode;
 951 }
 952
 953 static int ext4_drop_inode(struct inode *inode)
 954 {
 955         int drop = generic_drop_inode(inode);
 956
 957         trace_ext4_drop_inode(inode, drop);
 958         return drop;
 959 }
 960
 961 static void ext4_i_callback(struct rcu_head *head)
 962 {
 963         struct inode *inode = container_of(head, struct inode, i_rcu);
 964         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 965 }
 966
 967 static void ext4_destroy_inode(struct inode *inode)
 968 {
 969         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 970                 ext4_msg(inode->i_sb, KERN_ERR,
 971                          "Inode %lu (%p): orphan list check failed!",
 972                          inode->i_ino, EXT4_I(inode));
 973                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 974                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 975                                 true);
 976                 dump_stack();
 977         }
 978         call_rcu(&inode->i_rcu, ext4_i_callback);
 979 }
 980
 981 static void init_once(void *foo)
 982 {
 983         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 984
 985         INIT_LIST_HEAD(&ei->i_orphan);
 986         init_rwsem(&ei->xattr_sem);
 987         init_rwsem(&ei->i_data_sem);
 988         init_rwsem(&ei->i_mmap_sem);
 989         inode_init_once(&ei->vfs_inode);
 990 }
 991
 992 static int __init init_inodecache(void)
 993 {
 994         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 995                                              sizeof(struct ext4_inode_info),
 996                                              0, (SLAB_RECLAIM_ACCOUNT|
 997                                                 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
 998                                              init_once);
 999         if (ext4_inode_cachep == NULL)
1000                 return -ENOMEM;
1001         return 0;
1002 }
1003
1004 static void destroy_inodecache(void)
1005 {
1006         /*
1007          * Make sure all delayed rcu free inodes are flushed before we
1008          * destroy cache.
1009          */
1010         rcu_barrier();
1011         kmem_cache_destroy(ext4_inode_cachep);
1012 }
1013
1014 void ext4_clear_inode(struct inode *inode)
1015 {
1016         invalidate_inode_buffers(inode);
1017         clear_inode(inode);
1018         dquot_drop(inode);
1019         ext4_discard_preallocations(inode);
1020         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1021         if (EXT4_I(inode)->jinode) {
1022                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1023                                                EXT4_I(inode)->jinode);
1024                 jbd2_free_inode(EXT4_I(inode)->jinode);
1025                 EXT4_I(inode)->jinode = NULL;
1026         }
1027 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1028         if (EXT4_I(inode)->i_crypt_info)
1029                 ext4_free_encryption_info(inode, EXT4_I(inode)->i_crypt_info);
1030 #endif
1031 }
1032
1033 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1034                                         u64 ino, u32 generation)
1035 {
1036         struct inode *inode;
1037
1038         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1039                 return ERR_PTR(-ESTALE);
1040         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1041                 return ERR_PTR(-ESTALE);
1042
1043         /* iget isn't really right if the inode is currently unallocated!!
1044          *
1045          * ext4_read_inode will return a bad_inode if the inode had been
1046          * deleted, so we should be safe.
1047          *
1048          * Currently we don't know the generation for parent directory, so
1049          * a generation of 0 means "accept any"
1050          */
1051         inode = ext4_iget_normal(sb, ino);
1052         if (IS_ERR(inode))
1053                 return ERR_CAST(inode);
1054         if (generation && inode->i_generation != generation) {
1055                 iput(inode);
1056                 return ERR_PTR(-ESTALE);
1057         }
1058
1059         return inode;
1060 }
1061
1062 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1063                                         int fh_len, int fh_type)
1064 {
1065         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1066                                     ext4_nfs_get_inode);
1067 }
1068
1069 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1070                                         int fh_len, int fh_type)
1071 {
1072         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1073                                     ext4_nfs_get_inode);
1074 }
1075
1076 /*
1077  * Try to release metadata pages (indirect blocks, directories) which are
1078  * mapped via the block device.  Since these pages could have journal heads
1079  * which would prevent try_to_free_buffers() from freeing them, we must use
1080  * jbd2 layer's try_to_free_buffers() function to release them.
1081  */
1082 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1083                                  gfp_t wait)
1084 {
1085         journal_t *journal = EXT4_SB(sb)->s_journal;
1086
1087         WARN_ON(PageChecked(page));
1088         if (!page_has_buffers(page))
1089                 return 0;
1090         if (journal)
1091                 return jbd2_journal_try_to_free_buffers(journal, page,
1092                                                 wait & ~__GFP_DIRECT_RECLAIM);
1093         return try_to_free_buffers(page);
1094 }
1095
1096 #ifdef CONFIG_QUOTA
1097 static char *quotatypes[] = INITQFNAMES;
1098 #define QTYPE2NAME(t) (quotatypes[t])
1099
1100 static int ext4_write_dquot(struct dquot *dquot);
1101 static int ext4_acquire_dquot(struct dquot *dquot);
1102 static int ext4_release_dquot(struct dquot *dquot);
1103 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1104 static int ext4_write_info(struct super_block *sb, int type);
1105 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1106                          struct path *path);
1107 static int ext4_quota_off(struct super_block *sb, int type);
1108 static int ext4_quota_on_mount(struct super_block *sb, int type);
1109 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1110                                size_t len, loff_t off);
1111 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1112                                 const char *data, size_t len, loff_t off);
1113 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1114                              unsigned int flags);
1115 static int ext4_enable_quotas(struct super_block *sb);
1116
1117 static struct dquot **ext4_get_dquots(struct inode *inode)
1118 {
1119         return EXT4_I(inode)->i_dquot;
1120 }
1121
1122 static const struct dquot_operations ext4_quota_operations = {
1123         .get_reserved_space = ext4_get_reserved_space,
1124         .write_dquot    = ext4_write_dquot,
1125         .acquire_dquot  = ext4_acquire_dquot,
1126         .release_dquot  = ext4_release_dquot,
1127         .mark_dirty     = ext4_mark_dquot_dirty,
1128         .write_info     = ext4_write_info,
1129         .alloc_dquot    = dquot_alloc,
1130         .destroy_dquot  = dquot_destroy,
1131         .get_projid     = ext4_get_projid,
1132 };
1133
1134 static const struct quotactl_ops ext4_qctl_operations = {
1135         .quota_on       = ext4_quota_on,
1136         .quota_off      = ext4_quota_off,
1137         .quota_sync     = dquot_quota_sync,
1138         .get_state      = dquot_get_state,
1139         .set_info       = dquot_set_dqinfo,
1140         .get_dqblk      = dquot_get_dqblk,
1141         .set_dqblk      = dquot_set_dqblk
1142 };
1143 #endif
1144
1145 static const struct super_operations ext4_sops = {
1146         .alloc_inode    = ext4_alloc_inode,
1147         .destroy_inode  = ext4_destroy_inode,
1148         .write_inode    = ext4_write_inode,
1149         .dirty_inode    = ext4_dirty_inode,
1150         .drop_inode     = ext4_drop_inode,
1151         .evict_inode    = ext4_evict_inode,
1152         .put_super      = ext4_put_super,
1153         .sync_fs        = ext4_sync_fs,
1154         .freeze_fs      = ext4_freeze,
1155         .unfreeze_fs    = ext4_unfreeze,
1156         .statfs         = ext4_statfs,
1157         .remount_fs     = ext4_remount,
1158         .show_options   = ext4_show_options,
1159 #ifdef CONFIG_QUOTA
1160         .quota_read     = ext4_quota_read,
1161         .quota_write    = ext4_quota_write,
1162         .get_dquots     = ext4_get_dquots,
1163 #endif
1164         .bdev_try_to_free_page = bdev_try_to_free_page,
1165 };
1166
1167 static const struct export_operations ext4_export_ops = {
1168         .fh_to_dentry = ext4_fh_to_dentry,
1169         .fh_to_parent = ext4_fh_to_parent,
1170         .get_parent = ext4_get_parent,
1171 };
1172
1173 enum {
1174         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1175         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1176         Opt_nouid32, Opt_debug, Opt_removed,
1177         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1178         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1179         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1180         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1181         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1182         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1183         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1184         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1185         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1186         Opt_usrquota, Opt_grpquota, Opt_i_version, Opt_dax,
1187         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1188         Opt_lazytime, Opt_nolazytime,
1189         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1190         Opt_inode_readahead_blks, Opt_journal_ioprio,
1191         Opt_dioread_nolock, Opt_dioread_lock,
1192         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1193         Opt_max_dir_size_kb, Opt_nojournal_checksum,
1194 };
1195
1196 static const match_table_t tokens = {
1197         {Opt_bsd_df, "bsddf"},
1198         {Opt_minix_df, "minixdf"},
1199         {Opt_grpid, "grpid"},
1200         {Opt_grpid, "bsdgroups"},
1201         {Opt_nogrpid, "nogrpid"},
1202         {Opt_nogrpid, "sysvgroups"},
1203         {Opt_resgid, "resgid=%u"},
1204         {Opt_resuid, "resuid=%u"},
1205         {Opt_sb, "sb=%u"},
1206         {Opt_err_cont, "errors=continue"},
1207         {Opt_err_panic, "errors=panic"},
1208         {Opt_err_ro, "errors=remount-ro"},
1209         {Opt_nouid32, "nouid32"},
1210         {Opt_debug, "debug"},
1211         {Opt_removed, "oldalloc"},
1212         {Opt_removed, "orlov"},
1213         {Opt_user_xattr, "user_xattr"},
1214         {Opt_nouser_xattr, "nouser_xattr"},
1215         {Opt_acl, "acl"},
1216         {Opt_noacl, "noacl"},
1217         {Opt_noload, "norecovery"},
1218         {Opt_noload, "noload"},
1219         {Opt_removed, "nobh"},
1220         {Opt_removed, "bh"},
1221         {Opt_commit, "commit=%u"},
1222         {Opt_min_batch_time, "min_batch_time=%u"},
1223         {Opt_max_batch_time, "max_batch_time=%u"},
1224         {Opt_journal_dev, "journal_dev=%u"},
1225         {Opt_journal_path, "journal_path=%s"},
1226         {Opt_journal_checksum, "journal_checksum"},
1227         {Opt_nojournal_checksum, "nojournal_checksum"},
1228         {Opt_journal_async_commit, "journal_async_commit"},
1229         {Opt_abort, "abort"},
1230         {Opt_data_journal, "data=journal"},
1231         {Opt_data_ordered, "data=ordered"},
1232         {Opt_data_writeback, "data=writeback"},
1233         {Opt_data_err_abort, "data_err=abort"},
1234         {Opt_data_err_ignore, "data_err=ignore"},
1235         {Opt_offusrjquota, "usrjquota="},
1236         {Opt_usrjquota, "usrjquota=%s"},
1237         {Opt_offgrpjquota, "grpjquota="},
1238         {Opt_grpjquota, "grpjquota=%s"},
1239         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1240         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1241         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1242         {Opt_grpquota, "grpquota"},
1243         {Opt_noquota, "noquota"},
1244         {Opt_quota, "quota"},
1245         {Opt_usrquota, "usrquota"},
1246         {Opt_barrier, "barrier=%u"},
1247         {Opt_barrier, "barrier"},
1248         {Opt_nobarrier, "nobarrier"},
1249         {Opt_i_version, "i_version"},
1250         {Opt_dax, "dax"},
1251         {Opt_stripe, "stripe=%u"},
1252         {Opt_delalloc, "delalloc"},
1253         {Opt_lazytime, "lazytime"},
1254         {Opt_nolazytime, "nolazytime"},
1255         {Opt_nodelalloc, "nodelalloc"},
1256         {Opt_removed, "mblk_io_submit"},
1257         {Opt_removed, "nomblk_io_submit"},
1258         {Opt_block_validity, "block_validity"},
1259         {Opt_noblock_validity, "noblock_validity"},
1260         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1261         {Opt_journal_ioprio, "journal_ioprio=%u"},
1262         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1263         {Opt_auto_da_alloc, "auto_da_alloc"},
1264         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1265         {Opt_dioread_nolock, "dioread_nolock"},
1266         {Opt_dioread_lock, "dioread_lock"},
1267         {Opt_discard, "discard"},
1268         {Opt_nodiscard, "nodiscard"},
1269         {Opt_init_itable, "init_itable=%u"},
1270         {Opt_init_itable, "init_itable"},
1271         {Opt_noinit_itable, "noinit_itable"},
1272         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1273         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1274         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1275         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1276         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1277         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1278         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1279         {Opt_err, NULL},
1280 };
1281
1282 static ext4_fsblk_t get_sb_block(void **data)
1283 {
1284         ext4_fsblk_t    sb_block;
1285         char            *options = (char *) *data;
1286
1287         if (!options || strncmp(options, "sb=", 3) != 0)
1288                 return 1;       /* Default location */
1289
1290         options += 3;
1291         /* TODO: use simple_strtoll with >32bit ext4 */
1292         sb_block = simple_strtoul(options, &options, 0);
1293         if (*options && *options != ',') {
1294                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1295                        (char *) *data);
1296                 return 1;
1297         }
1298         if (*options == ',')
1299                 options++;
1300         *data = (void *) options;
1301
1302         return sb_block;
1303 }
1304
1305 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1306 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1307         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1308
1309 #ifdef CONFIG_QUOTA
1310 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1311 {
1312         struct ext4_sb_info *sbi = EXT4_SB(sb);
1313         char *qname;
1314         int ret = -1;
1315
1316         if (sb_any_quota_loaded(sb) &&
1317                 !sbi->s_qf_names[qtype]) {
1318                 ext4_msg(sb, KERN_ERR,
1319                         "Cannot change journaled "
1320                         "quota options when quota turned on");
1321                 return -1;
1322         }
1323         if (ext4_has_feature_quota(sb)) {
1324                 ext4_msg(sb, KERN_ERR, "Cannot set journaled quota options "
1325                          "when QUOTA feature is enabled");
1326                 return -1;
1327         }
1328         qname = match_strdup(args);
1329         if (!qname) {
1330                 ext4_msg(sb, KERN_ERR,
1331                         "Not enough memory for storing quotafile name");
1332                 return -1;
1333         }
1334         if (sbi->s_qf_names[qtype]) {
1335                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1336                         ret = 1;
1337                 else
1338                         ext4_msg(sb, KERN_ERR,
1339                                  "%s quota file already specified",
1340                                  QTYPE2NAME(qtype));
1341                 goto errout;
1342         }
1343         if (strchr(qname, '/')) {
1344                 ext4_msg(sb, KERN_ERR,
1345                         "quotafile must be on filesystem root");
1346                 goto errout;
1347         }
1348         sbi->s_qf_names[qtype] = qname;
1349         set_opt(sb, QUOTA);
1350         return 1;
1351 errout:
1352         kfree(qname);
1353         return ret;
1354 }
1355
1356 static int clear_qf_name(struct super_block *sb, int qtype)
1357 {
1358
1359         struct ext4_sb_info *sbi = EXT4_SB(sb);
1360
1361         if (sb_any_quota_loaded(sb) &&
1362                 sbi->s_qf_names[qtype]) {
1363                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1364                         " when quota turned on");
1365                 return -1;
1366         }
1367         kfree(sbi->s_qf_names[qtype]);
1368         sbi->s_qf_names[qtype] = NULL;
1369         return 1;
1370 }
1371 #endif
1372
1373 #define MOPT_SET        0x0001
1374 #define MOPT_CLEAR      0x0002
1375 #define MOPT_NOSUPPORT  0x0004
1376 #define MOPT_EXPLICIT   0x0008
1377 #define MOPT_CLEAR_ERR  0x0010
1378 #define MOPT_GTE0       0x0020
1379 #ifdef CONFIG_QUOTA
1380 #define MOPT_Q          0
1381 #define MOPT_QFMT       0x0040
1382 #else
1383 #define MOPT_Q          MOPT_NOSUPPORT
1384 #define MOPT_QFMT       MOPT_NOSUPPORT
1385 #endif
1386 #define MOPT_DATAJ      0x0080
1387 #define MOPT_NO_EXT2    0x0100
1388 #define MOPT_NO_EXT3    0x0200
1389 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1390 #define MOPT_STRING     0x0400
1391
1392 static const struct mount_opts {
1393         int     token;
1394         int     mount_opt;
1395         int     flags;
1396 } ext4_mount_opts[] = {
1397         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1398         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1399         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1400         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1401         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1402         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1403         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1404          MOPT_EXT4_ONLY | MOPT_SET},
1405         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1406          MOPT_EXT4_ONLY | MOPT_CLEAR},
1407         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1408         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1409         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1410          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1411         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1412          MOPT_EXT4_ONLY | MOPT_CLEAR},
1413         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1414          MOPT_EXT4_ONLY | MOPT_CLEAR},
1415         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1416          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1417         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1418                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1419          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1420         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1421         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1422         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1423         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1424         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1425          MOPT_NO_EXT2},
1426         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1427          MOPT_NO_EXT2},
1428         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1429         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1430         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1431         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1432         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1433         {Opt_commit, 0, MOPT_GTE0},
1434         {Opt_max_batch_time, 0, MOPT_GTE0},
1435         {Opt_min_batch_time, 0, MOPT_GTE0},
1436         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1437         {Opt_init_itable, 0, MOPT_GTE0},
1438         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1439         {Opt_stripe, 0, MOPT_GTE0},
1440         {Opt_resuid, 0, MOPT_GTE0},
1441         {Opt_resgid, 0, MOPT_GTE0},
1442         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1443         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1444         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1445         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1446         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1447         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1448          MOPT_NO_EXT2 | MOPT_DATAJ},
1449         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1450         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1451 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1452         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1453         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1454 #else
1455         {Opt_acl, 0, MOPT_NOSUPPORT},
1456         {Opt_noacl, 0, MOPT_NOSUPPORT},
1457 #endif
1458         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1459         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1460         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1461         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1462                                                         MOPT_SET | MOPT_Q},
1463         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1464                                                         MOPT_SET | MOPT_Q},
1465         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1466                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1467         {Opt_usrjquota, 0, MOPT_Q},
1468         {Opt_grpjquota, 0, MOPT_Q},
1469         {Opt_offusrjquota, 0, MOPT_Q},
1470         {Opt_offgrpjquota, 0, MOPT_Q},
1471         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1472         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1473         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1474         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1475         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1476         {Opt_err, 0, 0}
1477 };
1478
1479 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1480                             substring_t *args, unsigned long *journal_devnum,
1481                             unsigned int *journal_ioprio, int is_remount)
1482 {
1483         struct ext4_sb_info *sbi = EXT4_SB(sb);
1484         const struct mount_opts *m;
1485         kuid_t uid;
1486         kgid_t gid;
1487         int arg = 0;
1488
1489 #ifdef CONFIG_QUOTA
1490         if (token == Opt_usrjquota)
1491                 return set_qf_name(sb, USRQUOTA, &args[0]);
1492         else if (token == Opt_grpjquota)
1493                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1494         else if (token == Opt_offusrjquota)
1495                 return clear_qf_name(sb, USRQUOTA);
1496         else if (token == Opt_offgrpjquota)
1497                 return clear_qf_name(sb, GRPQUOTA);
1498 #endif
1499         switch (token) {
1500         case Opt_noacl:
1501         case Opt_nouser_xattr:
1502                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1503                 break;
1504         case Opt_sb:
1505                 return 1;       /* handled by get_sb_block() */
1506         case Opt_removed:
1507                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1508                 return 1;
1509         case Opt_abort:
1510                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1511                 return 1;
1512         case Opt_i_version:
1513                 sb->s_flags |= MS_I_VERSION;
1514                 return 1;
1515         case Opt_lazytime:
1516                 sb->s_flags |= MS_LAZYTIME;
1517                 return 1;
1518         case Opt_nolazytime:
1519                 sb->s_flags &= ~MS_LAZYTIME;
1520                 return 1;
1521         }
1522
1523         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1524                 if (token == m->token)
1525                         break;
1526
1527         if (m->token == Opt_err) {
1528                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1529                          "or missing value", opt);
1530                 return -1;
1531         }
1532
1533         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1534                 ext4_msg(sb, KERN_ERR,
1535                          "Mount option \"%s\" incompatible with ext2", opt);
1536                 return -1;
1537         }
1538         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1539                 ext4_msg(sb, KERN_ERR,
1540                          "Mount option \"%s\" incompatible with ext3", opt);
1541                 return -1;
1542         }
1543
1544         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1545                 return -1;
1546         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1547                 return -1;
1548         if (m->flags & MOPT_EXPLICIT) {
1549                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1550                         set_opt2(sb, EXPLICIT_DELALLOC);
1551                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1552                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1553                 } else
1554                         return -1;
1555         }
1556         if (m->flags & MOPT_CLEAR_ERR)
1557                 clear_opt(sb, ERRORS_MASK);
1558         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1559                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1560                          "options when quota turned on");
1561                 return -1;
1562         }
1563
1564         if (m->flags & MOPT_NOSUPPORT) {
1565                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1566         } else if (token == Opt_commit) {
1567                 if (arg == 0)
1568                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1569                 sbi->s_commit_interval = HZ * arg;
1570         } else if (token == Opt_max_batch_time) {
1571                 sbi->s_max_batch_time = arg;
1572         } else if (token == Opt_min_batch_time) {
1573                 sbi->s_min_batch_time = arg;
1574         } else if (token == Opt_inode_readahead_blks) {
1575                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1576                         ext4_msg(sb, KERN_ERR,
1577                                  "EXT4-fs: inode_readahead_blks must be "
1578                                  "0 or a power of 2 smaller than 2^31");
1579                         return -1;
1580                 }
1581                 sbi->s_inode_readahead_blks = arg;
1582         } else if (token == Opt_init_itable) {
1583                 set_opt(sb, INIT_INODE_TABLE);
1584                 if (!args->from)
1585                         arg = EXT4_DEF_LI_WAIT_MULT;
1586                 sbi->s_li_wait_mult = arg;
1587         } else if (token == Opt_max_dir_size_kb) {
1588                 sbi->s_max_dir_size_kb = arg;
1589         } else if (token == Opt_stripe) {
1590                 sbi->s_stripe = arg;
1591         } else if (token == Opt_resuid) {
1592                 uid = make_kuid(current_user_ns(), arg);
1593                 if (!uid_valid(uid)) {
1594                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1595                         return -1;
1596                 }
1597                 sbi->s_resuid = uid;
1598         } else if (token == Opt_resgid) {
1599                 gid = make_kgid(current_user_ns(), arg);
1600                 if (!gid_valid(gid)) {
1601                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1602                         return -1;
1603                 }
1604                 sbi->s_resgid = gid;
1605         } else if (token == Opt_journal_dev) {
1606                 if (is_remount) {
1607                         ext4_msg(sb, KERN_ERR,
1608                                  "Cannot specify journal on remount");
1609                         return -1;
1610                 }
1611                 *journal_devnum = arg;
1612         } else if (token == Opt_journal_path) {
1613                 char *journal_path;
1614                 struct inode *journal_inode;
1615                 struct path path;
1616                 int error;
1617
1618                 if (is_remount) {
1619                         ext4_msg(sb, KERN_ERR,
1620                                  "Cannot specify journal on remount");
1621                         return -1;
1622                 }
1623                 journal_path = match_strdup(&args[0]);
1624                 if (!journal_path) {
1625                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1626                                 "journal device string");
1627                         return -1;
1628                 }
1629
1630                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1631                 if (error) {
1632                         ext4_msg(sb, KERN_ERR, "error: could not find "
1633                                 "journal device path: error %d", error);
1634                         kfree(journal_path);
1635                         return -1;
1636                 }
1637
1638                 journal_inode = d_inode(path.dentry);
1639                 if (!S_ISBLK(journal_inode->i_mode)) {
1640                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1641                                 "is not a block device", journal_path);
1642                         path_put(&path);
1643                         kfree(journal_path);
1644                         return -1;
1645                 }
1646
1647                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1648                 path_put(&path);
1649                 kfree(journal_path);
1650         } else if (token == Opt_journal_ioprio) {
1651                 if (arg > 7) {
1652                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1653                                  " (must be 0-7)");
1654                         return -1;
1655                 }
1656                 *journal_ioprio =
1657                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1658         } else if (token == Opt_test_dummy_encryption) {
1659 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1660                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1661                 ext4_msg(sb, KERN_WARNING,
1662                          "Test dummy encryption mode enabled");
1663 #else
1664                 ext4_msg(sb, KERN_WARNING,
1665                          "Test dummy encryption mount option ignored");
1666 #endif
1667         } else if (m->flags & MOPT_DATAJ) {
1668                 if (is_remount) {
1669                         if (!sbi->s_journal)
1670                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1671                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1672                                 ext4_msg(sb, KERN_ERR,
1673                                          "Cannot change data mode on remount");
1674                                 return -1;
1675                         }
1676                 } else {
1677                         clear_opt(sb, DATA_FLAGS);
1678                         sbi->s_mount_opt |= m->mount_opt;
1679                 }
1680 #ifdef CONFIG_QUOTA
1681         } else if (m->flags & MOPT_QFMT) {
1682                 if (sb_any_quota_loaded(sb) &&
1683                     sbi->s_jquota_fmt != m->mount_opt) {
1684                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1685                                  "quota options when quota turned on");
1686                         return -1;
1687                 }
1688                 if (ext4_has_feature_quota(sb)) {
1689                         ext4_msg(sb, KERN_ERR,
1690                                  "Cannot set journaled quota options "
1691                                  "when QUOTA feature is enabled");
1692                         return -1;
1693                 }
1694                 sbi->s_jquota_fmt = m->mount_opt;
1695 #endif
1696         } else if (token == Opt_dax) {
1697 #ifdef CONFIG_FS_DAX
1698                 ext4_msg(sb, KERN_WARNING,
1699                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1700                         sbi->s_mount_opt |= m->mount_opt;
1701 #else
1702                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1703                 return -1;
1704 #endif
1705         } else if (token == Opt_data_err_abort) {
1706                 sbi->s_mount_opt |= m->mount_opt;
1707         } else if (token == Opt_data_err_ignore) {
1708                 sbi->s_mount_opt &= ~m->mount_opt;
1709         } else {
1710                 if (!args->from)
1711                         arg = 1;
1712                 if (m->flags & MOPT_CLEAR)
1713                         arg = !arg;
1714                 else if (unlikely(!(m->flags & MOPT_SET))) {
1715                         ext4_msg(sb, KERN_WARNING,
1716                                  "buggy handling of option %s", opt);
1717                         WARN_ON(1);
1718                         return -1;
1719                 }
1720                 if (arg != 0)
1721                         sbi->s_mount_opt |= m->mount_opt;
1722                 else
1723                         sbi->s_mount_opt &= ~m->mount_opt;
1724         }
1725         return 1;
1726 }
1727
1728 static int parse_options(char *options, struct super_block *sb,
1729                          unsigned long *journal_devnum,
1730                          unsigned int *journal_ioprio,
1731                          int is_remount)
1732 {
1733         struct ext4_sb_info *sbi = EXT4_SB(sb);
1734         char *p;
1735         substring_t args[MAX_OPT_ARGS];
1736         int token;
1737
1738         if (!options)
1739                 return 1;
1740
1741         while ((p = strsep(&options, ",")) != NULL) {
1742                 if (!*p)
1743                         continue;
1744                 /*
1745                  * Initialize args struct so we know whether arg was
1746                  * found; some options take optional arguments.
1747                  */
1748                 args[0].to = args[0].from = NULL;
1749                 token = match_token(p, tokens, args);
1750                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1751                                      journal_ioprio, is_remount) < 0)
1752                         return 0;
1753         }
1754 #ifdef CONFIG_QUOTA
1755         if (ext4_has_feature_quota(sb) &&
1756             (test_opt(sb, USRQUOTA) || test_opt(sb, GRPQUOTA))) {
1757                 ext4_msg(sb, KERN_ERR, "Cannot set quota options when QUOTA "
1758                          "feature is enabled");
1759                 return 0;
1760         }
1761         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1762                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1763                         clear_opt(sb, USRQUOTA);
1764
1765                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1766                         clear_opt(sb, GRPQUOTA);
1767
1768                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1769                         ext4_msg(sb, KERN_ERR, "old and new quota "
1770                                         "format mixing");
1771                         return 0;
1772                 }
1773
1774                 if (!sbi->s_jquota_fmt) {
1775                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1776                                         "not specified");
1777                         return 0;
1778                 }
1779         }
1780 #endif
1781         if (test_opt(sb, DIOREAD_NOLOCK)) {
1782                 int blocksize =
1783                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1784
1785                 if (blocksize < PAGE_CACHE_SIZE) {
1786                         ext4_msg(sb, KERN_ERR, "can't mount with "
1787                                  "dioread_nolock if block size != PAGE_SIZE");
1788                         return 0;
1789                 }
1790         }
1791         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
1792             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
1793                 ext4_msg(sb, KERN_ERR, "can't mount with journal_async_commit "
1794                          "in data=ordered mode");
1795                 return 0;
1796         }
1797         return 1;
1798 }
1799
1800 static inline void ext4_show_quota_options(struct seq_file *seq,
1801                                            struct super_block *sb)
1802 {
1803 #if defined(CONFIG_QUOTA)
1804         struct ext4_sb_info *sbi = EXT4_SB(sb);
1805
1806         if (sbi->s_jquota_fmt) {
1807                 char *fmtname = "";
1808
1809                 switch (sbi->s_jquota_fmt) {
1810                 case QFMT_VFS_OLD:
1811                         fmtname = "vfsold";
1812                         break;
1813                 case QFMT_VFS_V0:
1814                         fmtname = "vfsv0";
1815                         break;
1816                 case QFMT_VFS_V1:
1817                         fmtname = "vfsv1";
1818                         break;
1819                 }
1820                 seq_printf(seq, ",jqfmt=%s", fmtname);
1821         }
1822
1823         if (sbi->s_qf_names[USRQUOTA])
1824                 seq_show_option(seq, "usrjquota", sbi->s_qf_names[USRQUOTA]);
1825
1826         if (sbi->s_qf_names[GRPQUOTA])
1827                 seq_show_option(seq, "grpjquota", sbi->s_qf_names[GRPQUOTA]);
1828 #endif
1829 }
1830
1831 static const char *token2str(int token)
1832 {
1833         const struct match_token *t;
1834
1835         for (t = tokens; t->token != Opt_err; t++)
1836                 if (t->token == token && !strchr(t->pattern, '='))
1837                         break;
1838         return t->pattern;
1839 }
1840
1841 /*
1842  * Show an option if
1843  *  - it's set to a non-default value OR
1844  *  - if the per-sb default is different from the global default
1845  */
1846 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1847                               int nodefs)
1848 {
1849         struct ext4_sb_info *sbi = EXT4_SB(sb);
1850         struct ext4_super_block *es = sbi->s_es;
1851         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1852         const struct mount_opts *m;
1853         char sep = nodefs ? '\n' : ',';
1854
1855 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1856 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1857
1858         if (sbi->s_sb_block != 1)
1859                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1860
1861         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1862                 int want_set = m->flags & MOPT_SET;
1863                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1864                     (m->flags & MOPT_CLEAR_ERR))
1865                         continue;
1866                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1867                         continue; /* skip if same as the default */
1868                 if ((want_set &&
1869                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1870                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1871                         continue; /* select Opt_noFoo vs Opt_Foo */
1872                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1873         }
1874
1875         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1876             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1877                 SEQ_OPTS_PRINT("resuid=%u",
1878                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1879         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1880             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1881                 SEQ_OPTS_PRINT("resgid=%u",
1882                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1883         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1884         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1885                 SEQ_OPTS_PUTS("errors=remount-ro");
1886         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1887                 SEQ_OPTS_PUTS("errors=continue");
1888         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1889                 SEQ_OPTS_PUTS("errors=panic");
1890         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1891                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1892         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1893                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1894         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1895                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1896         if (sb->s_flags & MS_I_VERSION)
1897                 SEQ_OPTS_PUTS("i_version");
1898         if (nodefs || sbi->s_stripe)
1899                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1900         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1901                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1902                         SEQ_OPTS_PUTS("data=journal");
1903                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1904                         SEQ_OPTS_PUTS("data=ordered");
1905                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1906                         SEQ_OPTS_PUTS("data=writeback");
1907         }
1908         if (nodefs ||
1909             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1910                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1911                                sbi->s_inode_readahead_blks);
1912
1913         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1914                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1915                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1916         if (nodefs || sbi->s_max_dir_size_kb)
1917                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1918         if (test_opt(sb, DATA_ERR_ABORT))
1919                 SEQ_OPTS_PUTS("data_err=abort");
1920
1921         ext4_show_quota_options(seq, sb);
1922         return 0;
1923 }
1924
1925 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1926 {
1927         return _ext4_show_options(seq, root->d_sb, 0);
1928 }
1929
1930 int ext4_seq_options_show(struct seq_file *seq, void *offset)
1931 {
1932         struct super_block *sb = seq->private;
1933         int rc;
1934
1935         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1936         rc = _ext4_show_options(seq, sb, 1);
1937         seq_puts(seq, "\n");
1938         return rc;
1939 }
1940
1941 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1942                             int read_only)
1943 {
1944         struct ext4_sb_info *sbi = EXT4_SB(sb);
1945         int res = 0;
1946
1947         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1948                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1949                          "forcing read-only mode");
1950                 res = MS_RDONLY;
1951         }
1952         if (read_only)
1953                 goto done;
1954         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1955                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1956                          "running e2fsck is recommended");
1957         else if (sbi->s_mount_state & EXT4_ERROR_FS)
1958                 ext4_msg(sb, KERN_WARNING,
1959                          "warning: mounting fs with errors, "
1960                          "running e2fsck is recommended");
1961         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1962                  le16_to_cpu(es->s_mnt_count) >=
1963                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1964                 ext4_msg(sb, KERN_WARNING,
1965                          "warning: maximal mount count reached, "
1966                          "running e2fsck is recommended");
1967         else if (le32_to_cpu(es->s_checkinterval) &&
1968                 (le32_to_cpu(es->s_lastcheck) +
1969                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1970                 ext4_msg(sb, KERN_WARNING,
1971                          "warning: checktime reached, "
1972                          "running e2fsck is recommended");
1973         if (!sbi->s_journal)
1974                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1975         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1976                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1977         le16_add_cpu(&es->s_mnt_count, 1);
1978         es->s_mtime = cpu_to_le32(get_seconds());
1979         ext4_update_dynamic_rev(sb);
1980         if (sbi->s_journal)
1981                 ext4_set_feature_journal_needs_recovery(sb);
1982
1983         ext4_commit_super(sb, 1);
1984 done:
1985         if (test_opt(sb, DEBUG))
1986                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1987                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1988                         sb->s_blocksize,
1989                         sbi->s_groups_count,
1990                         EXT4_BLOCKS_PER_GROUP(sb),
1991                         EXT4_INODES_PER_GROUP(sb),
1992                         sbi->s_mount_opt, sbi->s_mount_opt2);
1993
1994         cleancache_init_fs(sb);
1995         return res;
1996 }
1997
1998 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1999 {
2000         struct ext4_sb_info *sbi = EXT4_SB(sb);
2001         struct flex_groups *new_groups;
2002         int size;
2003
2004         if (!sbi->s_log_groups_per_flex)
2005                 return 0;
2006
2007         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2008         if (size <= sbi->s_flex_groups_allocated)
2009                 return 0;
2010
2011         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
2012         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
2013         if (!new_groups) {
2014                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
2015                          size / (int) sizeof(struct flex_groups));
2016                 return -ENOMEM;
2017         }
2018
2019         if (sbi->s_flex_groups) {
2020                 memcpy(new_groups, sbi->s_flex_groups,
2021                        (sbi->s_flex_groups_allocated *
2022                         sizeof(struct flex_groups)));
2023                 kvfree(sbi->s_flex_groups);
2024         }
2025         sbi->s_flex_groups = new_groups;
2026         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
2027         return 0;
2028 }
2029
2030 static int ext4_fill_flex_info(struct super_block *sb)
2031 {
2032         struct ext4_sb_info *sbi = EXT4_SB(sb);
2033         struct ext4_group_desc *gdp = NULL;
2034         ext4_group_t flex_group;
2035         int i, err;
2036
2037         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2038         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2039                 sbi->s_log_groups_per_flex = 0;
2040                 return 1;
2041         }
2042
2043         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2044         if (err)
2045                 goto failed;
2046
2047         for (i = 0; i < sbi->s_groups_count; i++) {
2048                 gdp = ext4_get_group_desc(sb, i, NULL);
2049
2050                 flex_group = ext4_flex_group(sbi, i);
2051                 atomic_add(ext4_free_inodes_count(sb, gdp),
2052                            &sbi->s_flex_groups[flex_group].free_inodes);
2053                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2054                              &sbi->s_flex_groups[flex_group].free_clusters);
2055                 atomic_add(ext4_used_dirs_count(sb, gdp),
2056                            &sbi->s_flex_groups[flex_group].used_dirs);
2057         }
2058
2059         return 1;
2060 failed:
2061         return 0;
2062 }
2063
2064 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2065                                    struct ext4_group_desc *gdp)
2066 {
2067         int offset;
2068         __u16 crc = 0;
2069         __le32 le_group = cpu_to_le32(block_group);
2070         struct ext4_sb_info *sbi = EXT4_SB(sb);
2071
2072         if (ext4_has_metadata_csum(sbi->s_sb)) {
2073                 /* Use new metadata_csum algorithm */
2074                 __le16 save_csum;
2075                 __u32 csum32;
2076
2077                 save_csum = gdp->bg_checksum;
2078                 gdp->bg_checksum = 0;
2079                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2080                                      sizeof(le_group));
2081                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
2082                                      sbi->s_desc_size);
2083                 gdp->bg_checksum = save_csum;
2084
2085                 crc = csum32 & 0xFFFF;
2086                 goto out;
2087         }
2088
2089         /* old crc16 code */
2090         if (!ext4_has_feature_gdt_csum(sb))
2091                 return 0;
2092
2093         offset = offsetof(struct ext4_group_desc, bg_checksum);
2094
2095         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2096         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2097         crc = crc16(crc, (__u8 *)gdp, offset);
2098         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2099         /* for checksum of struct ext4_group_desc do the rest...*/
2100         if (ext4_has_feature_64bit(sb) &&
2101             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2102                 crc = crc16(crc, (__u8 *)gdp + offset,
2103                             le16_to_cpu(sbi->s_es->s_desc_size) -
2104                                 offset);
2105
2106 out:
2107         return cpu_to_le16(crc);
2108 }
2109
2110 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2111                                 struct ext4_group_desc *gdp)
2112 {
2113         if (ext4_has_group_desc_csum(sb) &&
2114             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2115                 return 0;
2116
2117         return 1;
2118 }
2119
2120 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2121                               struct ext4_group_desc *gdp)
2122 {
2123         if (!ext4_has_group_desc_csum(sb))
2124                 return;
2125         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2126 }
2127
2128 /* Called at mount-time, super-block is locked */
2129 static int ext4_check_descriptors(struct super_block *sb,
2130                                   ext4_group_t *first_not_zeroed)
2131 {
2132         struct ext4_sb_info *sbi = EXT4_SB(sb);
2133         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2134         ext4_fsblk_t last_block;
2135         ext4_fsblk_t block_bitmap;
2136         ext4_fsblk_t inode_bitmap;
2137         ext4_fsblk_t inode_table;
2138         int flexbg_flag = 0;
2139         ext4_group_t i, grp = sbi->s_groups_count;
2140
2141         if (ext4_has_feature_flex_bg(sb))
2142                 flexbg_flag = 1;
2143
2144         ext4_debug("Checking group descriptors");
2145
2146         for (i = 0; i < sbi->s_groups_count; i++) {
2147                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2148
2149                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2150                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2151                 else
2152                         last_block = first_block +
2153                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2154
2155                 if ((grp == sbi->s_groups_count) &&
2156                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2157                         grp = i;
2158
2159                 block_bitmap = ext4_block_bitmap(sb, gdp);
2160                 if (block_bitmap < first_block || block_bitmap > last_block) {
2161                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2162                                "Block bitmap for group %u not in group "
2163                                "(block %llu)!", i, block_bitmap);
2164                         return 0;
2165                 }
2166                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2167                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2168                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2169                                "Inode bitmap for group %u not in group "
2170                                "(block %llu)!", i, inode_bitmap);
2171                         return 0;
2172                 }
2173                 inode_table = ext4_inode_table(sb, gdp);
2174                 if (inode_table < first_block ||
2175                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2176                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2177                                "Inode table for group %u not in group "
2178                                "(block %llu)!", i, inode_table);
2179                         return 0;
2180                 }
2181                 ext4_lock_group(sb, i);
2182                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2183                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2184                                  "Checksum for group %u failed (%u!=%u)",
2185                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2186                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2187                         if (!(sb->s_flags & MS_RDONLY)) {
2188                                 ext4_unlock_group(sb, i);
2189                                 return 0;
2190                         }
2191                 }
2192                 ext4_unlock_group(sb, i);
2193                 if (!flexbg_flag)
2194                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2195         }
2196         if (NULL != first_not_zeroed)
2197                 *first_not_zeroed = grp;
2198         return 1;
2199 }
2200
2201 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2202  * the superblock) which were deleted from all directories, but held open by
2203  * a process at the time of a crash.  We walk the list and try to delete these
2204  * inodes at recovery time (only with a read-write filesystem).
2205  *
2206  * In order to keep the orphan inode chain consistent during traversal (in
2207  * case of crash during recovery), we link each inode into the superblock
2208  * orphan list_head and handle it the same way as an inode deletion during
2209  * normal operation (which journals the operations for us).
2210  *
2211  * We only do an iget() and an iput() on each inode, which is very safe if we
2212  * accidentally point at an in-use or already deleted inode.  The worst that
2213  * can happen in this case is that we get a "bit already cleared" message from
2214  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2215  * e2fsck was run on this filesystem, and it must have already done the orphan
2216  * inode cleanup for us, so we can safely abort without any further action.
2217  */
2218 static void ext4_orphan_cleanup(struct super_block *sb,
2219                                 struct ext4_super_block *es)
2220 {
2221         unsigned int s_flags = sb->s_flags;
2222         int nr_orphans = 0, nr_truncates = 0;
2223 #ifdef CONFIG_QUOTA
2224         int i;
2225 #endif
2226         if (!es->s_last_orphan) {
2227                 jbd_debug(4, "no orphan inodes to clean up\n");
2228                 return;
2229         }
2230
2231         if (bdev_read_only(sb->s_bdev)) {
2232                 ext4_msg(sb, KERN_ERR, "write access "
2233                         "unavailable, skipping orphan cleanup");
2234                 return;
2235         }
2236
2237         /* Check if feature set would not allow a r/w mount */
2238         if (!ext4_feature_set_ok(sb, 0)) {
2239                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2240                          "unknown ROCOMPAT features");
2241                 return;
2242         }
2243
2244         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2245                 /* don't clear list on RO mount w/ errors */
2246                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2247                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2248                                   "clearing orphan list.\n");
2249                         es->s_last_orphan = 0;
2250                 }
2251                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2252                 return;
2253         }
2254
2255         if (s_flags & MS_RDONLY) {
2256                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2257                 sb->s_flags &= ~MS_RDONLY;
2258         }
2259 #ifdef CONFIG_QUOTA
2260         /* Needed for iput() to work correctly and not trash data */
2261         sb->s_flags |= MS_ACTIVE;
2262         /* Turn on quotas so that they are updated correctly */
2263         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2264                 if (EXT4_SB(sb)->s_qf_names[i]) {
2265                         int ret = ext4_quota_on_mount(sb, i);
2266                         if (ret < 0)
2267                                 ext4_msg(sb, KERN_ERR,
2268                                         "Cannot turn on journaled "
2269                                         "quota: error %d", ret);
2270                 }
2271         }
2272 #endif
2273
2274         while (es->s_last_orphan) {
2275                 struct inode *inode;
2276
2277                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2278                 if (IS_ERR(inode)) {
2279                         es->s_last_orphan = 0;
2280                         break;
2281                 }
2282
2283                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2284                 dquot_initialize(inode);
2285                 if (inode->i_nlink) {
2286                         if (test_opt(sb, DEBUG))
2287                                 ext4_msg(sb, KERN_DEBUG,
2288                                         "%s: truncating inode %lu to %lld bytes",
2289                                         __func__, inode->i_ino, inode->i_size);
2290                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2291                                   inode->i_ino, inode->i_size);
2292                         inode_lock(inode);
2293                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2294                         ext4_truncate(inode);
2295                         inode_unlock(inode);
2296                         nr_truncates++;
2297                 } else {
2298                         if (test_opt(sb, DEBUG))
2299                                 ext4_msg(sb, KERN_DEBUG,
2300                                         "%s: deleting unreferenced inode %lu",
2301                                         __func__, inode->i_ino);
2302                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2303                                   inode->i_ino);
2304                         nr_orphans++;
2305                 }
2306                 iput(inode);  /* The delete magic happens here! */
2307         }
2308
2309 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2310
2311         if (nr_orphans)
2312                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2313                        PLURAL(nr_orphans));
2314         if (nr_truncates)
2315                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2316                        PLURAL(nr_truncates));
2317 #ifdef CONFIG_QUOTA
2318         /* Turn quotas off */
2319         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2320                 if (sb_dqopt(sb)->files[i])
2321                         dquot_quota_off(sb, i);
2322         }
2323 #endif
2324         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2325 }
2326
2327 /*
2328  * Maximal extent format file size.
2329  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2330  * extent format containers, within a sector_t, and within i_blocks
2331  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2332  * so that won't be a limiting factor.
2333  *
2334  * However there is other limiting factor. We do store extents in the form
2335  * of starting block and length, hence the resulting length of the extent
2336  * covering maximum file size must fit into on-disk format containers as
2337  * well. Given that length is always by 1 unit bigger than max unit (because
2338  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2339  *
2340  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2341  */
2342 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2343 {
2344         loff_t res;
2345         loff_t upper_limit = MAX_LFS_FILESIZE;
2346
2347         /* small i_blocks in vfs inode? */
2348         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2349                 /*
2350                  * CONFIG_LBDAF is not enabled implies the inode
2351                  * i_block represent total blocks in 512 bytes
2352                  * 32 == size of vfs inode i_blocks * 8
2353                  */
2354                 upper_limit = (1LL << 32) - 1;
2355
2356                 /* total blocks in file system block size */
2357                 upper_limit >>= (blkbits - 9);
2358                 upper_limit <<= blkbits;
2359         }
2360
2361         /*
2362          * 32-bit extent-start container, ee_block. We lower the maxbytes
2363          * by one fs block, so ee_len can cover the extent of maximum file
2364          * size
2365          */
2366         res = (1LL << 32) - 1;
2367         res <<= blkbits;
2368
2369         /* Sanity check against vm- & vfs- imposed limits */
2370         if (res > upper_limit)
2371                 res = upper_limit;
2372
2373         return res;
2374 }
2375
2376 /*
2377  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2378  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2379  * We need to be 1 filesystem block less than the 2^48 sector limit.
2380  */
2381 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2382 {
2383         loff_t res = EXT4_NDIR_BLOCKS;
2384         int meta_blocks;
2385         loff_t upper_limit;
2386         /* This is calculated to be the largest file size for a dense, block
2387          * mapped file such that the file's total number of 512-byte sectors,
2388          * including data and all indirect blocks, does not exceed (2^48 - 1).
2389          *
2390          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2391          * number of 512-byte sectors of the file.
2392          */
2393
2394         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2395                 /*
2396                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2397                  * the inode i_block field represents total file blocks in
2398                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2399                  */
2400                 upper_limit = (1LL << 32) - 1;
2401
2402                 /* total blocks in file system block size */
2403                 upper_limit >>= (bits - 9);
2404
2405         } else {
2406                 /*
2407                  * We use 48 bit ext4_inode i_blocks
2408                  * With EXT4_HUGE_FILE_FL set the i_blocks
2409                  * represent total number of blocks in
2410                  * file system block size
2411                  */
2412                 upper_limit = (1LL << 48) - 1;
2413
2414         }
2415
2416         /* indirect blocks */
2417         meta_blocks = 1;
2418         /* double indirect blocks */
2419         meta_blocks += 1 + (1LL << (bits-2));
2420         /* tripple indirect blocks */
2421         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2422
2423         upper_limit -= meta_blocks;
2424         upper_limit <<= bits;
2425
2426         res += 1LL << (bits-2);
2427         res += 1LL << (2*(bits-2));
2428         res += 1LL << (3*(bits-2));
2429         res <<= bits;
2430         if (res > upper_limit)
2431                 res = upper_limit;
2432
2433         if (res > MAX_LFS_FILESIZE)
2434                 res = MAX_LFS_FILESIZE;
2435
2436         return res;
2437 }
2438
2439 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2440                                    ext4_fsblk_t logical_sb_block, int nr)
2441 {
2442         struct ext4_sb_info *sbi = EXT4_SB(sb);
2443         ext4_group_t bg, first_meta_bg;
2444         int has_super = 0;
2445
2446         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2447
2448         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2449                 return logical_sb_block + nr + 1;
2450         bg = sbi->s_desc_per_block * nr;
2451         if (ext4_bg_has_super(sb, bg))
2452                 has_super = 1;
2453
2454         /*
2455          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2456          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2457          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2458          * compensate.
2459          */
2460         if (sb->s_blocksize == 1024 && nr == 0 &&
2461             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2462                 has_super++;
2463
2464         return (has_super + ext4_group_first_block_no(sb, bg));
2465 }
2466
2467 /**
2468  * ext4_get_stripe_size: Get the stripe size.
2469  * @sbi: In memory super block info
2470  *
2471  * If we have specified it via mount option, then
2472  * use the mount option value. If the value specified at mount time is
2473  * greater than the blocks per group use the super block value.
2474  * If the super block value is greater than blocks per group return 0.
2475  * Allocator needs it be less than blocks per group.
2476  *
2477  */
2478 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2479 {
2480         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2481         unsigned long stripe_width =
2482                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2483         int ret;
2484
2485         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2486                 ret = sbi->s_stripe;
2487         else if (stripe_width <= sbi->s_blocks_per_group)
2488                 ret = stripe_width;
2489         else if (stride <= sbi->s_blocks_per_group)
2490                 ret = stride;
2491         else
2492                 ret = 0;
2493
2494         /*
2495          * If the stripe width is 1, this makes no sense and
2496          * we set it to 0 to turn off stripe handling code.
2497          */
2498         if (ret <= 1)
2499                 ret = 0;
2500
2501         return ret;
2502 }
2503
2504 /*
2505  * Check whether this filesystem can be mounted based on
2506  * the features present and the RDONLY/RDWR mount requested.
2507  * Returns 1 if this filesystem can be mounted as requested,
2508  * 0 if it cannot be.
2509  */
2510 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2511 {
2512         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2513                 ext4_msg(sb, KERN_ERR,
2514                         "Couldn't mount because of "
2515                         "unsupported optional features (%x)",
2516                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2517                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2518                 return 0;
2519         }
2520
2521         if (readonly)
2522                 return 1;
2523
2524         if (ext4_has_feature_readonly(sb)) {
2525                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2526                 sb->s_flags |= MS_RDONLY;
2527                 return 1;
2528         }
2529
2530         /* Check that feature set is OK for a read-write mount */
2531         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2532                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2533                          "unsupported optional features (%x)",
2534                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2535                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2536                 return 0;
2537         }
2538         /*
2539          * Large file size enabled file system can only be mounted
2540          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2541          */
2542         if (ext4_has_feature_huge_file(sb)) {
2543                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2544                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2545                                  "cannot be mounted RDWR without "
2546                                  "CONFIG_LBDAF");
2547                         return 0;
2548                 }
2549         }
2550         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2551                 ext4_msg(sb, KERN_ERR,
2552                          "Can't support bigalloc feature without "
2553                          "extents feature\n");
2554                 return 0;
2555         }
2556
2557 #ifndef CONFIG_QUOTA
2558         if (ext4_has_feature_quota(sb) && !readonly) {
2559                 ext4_msg(sb, KERN_ERR,
2560                          "Filesystem with quota feature cannot be mounted RDWR "
2561                          "without CONFIG_QUOTA");
2562                 return 0;
2563         }
2564         if (ext4_has_feature_project(sb) && !readonly) {
2565                 ext4_msg(sb, KERN_ERR,
2566                          "Filesystem with project quota feature cannot be mounted RDWR "
2567                          "without CONFIG_QUOTA");
2568                 return 0;
2569         }
2570 #endif  /* CONFIG_QUOTA */
2571         return 1;
2572 }
2573
2574 /*
2575  * This function is called once a day if we have errors logged
2576  * on the file system
2577  */
2578 static void print_daily_error_info(unsigned long arg)
2579 {
2580         struct super_block *sb = (struct super_block *) arg;
2581         struct ext4_sb_info *sbi;
2582         struct ext4_super_block *es;
2583
2584         sbi = EXT4_SB(sb);
2585         es = sbi->s_es;
2586
2587         if (es->s_error_count)
2588                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2589                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2590                          le32_to_cpu(es->s_error_count));
2591         if (es->s_first_error_time) {
2592                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2593                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2594                        (int) sizeof(es->s_first_error_func),
2595                        es->s_first_error_func,
2596                        le32_to_cpu(es->s_first_error_line));
2597                 if (es->s_first_error_ino)
2598                         printk(": inode %u",
2599                                le32_to_cpu(es->s_first_error_ino));
2600                 if (es->s_first_error_block)
2601                         printk(": block %llu", (unsigned long long)
2602                                le64_to_cpu(es->s_first_error_block));
2603                 printk("\n");
2604         }
2605         if (es->s_last_error_time) {
2606                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2607                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2608                        (int) sizeof(es->s_last_error_func),
2609                        es->s_last_error_func,
2610                        le32_to_cpu(es->s_last_error_line));
2611                 if (es->s_last_error_ino)
2612                         printk(": inode %u",
2613                                le32_to_cpu(es->s_last_error_ino));
2614                 if (es->s_last_error_block)
2615                         printk(": block %llu", (unsigned long long)
2616                                le64_to_cpu(es->s_last_error_block));
2617                 printk("\n");
2618         }
2619         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2620 }
2621
2622 /* Find next suitable group and run ext4_init_inode_table */
2623 static int ext4_run_li_request(struct ext4_li_request *elr)
2624 {
2625         struct ext4_group_desc *gdp = NULL;
2626         ext4_group_t group, ngroups;
2627         struct super_block *sb;
2628         unsigned long timeout = 0;
2629         int ret = 0;
2630
2631         sb = elr->lr_super;
2632         ngroups = EXT4_SB(sb)->s_groups_count;
2633
2634         sb_start_write(sb);
2635         for (group = elr->lr_next_group; group < ngroups; group++) {
2636                 gdp = ext4_get_group_desc(sb, group, NULL);
2637                 if (!gdp) {
2638                         ret = 1;
2639                         break;
2640                 }
2641
2642                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2643                         break;
2644         }
2645
2646         if (group >= ngroups)
2647                 ret = 1;
2648
2649         if (!ret) {
2650                 timeout = jiffies;
2651                 ret = ext4_init_inode_table(sb, group,
2652                                             elr->lr_timeout ? 0 : 1);
2653                 if (elr->lr_timeout == 0) {
2654                         timeout = (jiffies - timeout) *
2655                                   elr->lr_sbi->s_li_wait_mult;
2656                         elr->lr_timeout = timeout;
2657                 }
2658                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2659                 elr->lr_next_group = group + 1;
2660         }
2661         sb_end_write(sb);
2662
2663         return ret;
2664 }
2665
2666 /*
2667  * Remove lr_request from the list_request and free the
2668  * request structure. Should be called with li_list_mtx held
2669  */
2670 static void ext4_remove_li_request(struct ext4_li_request *elr)
2671 {
2672         struct ext4_sb_info *sbi;
2673
2674         if (!elr)
2675                 return;
2676
2677         sbi = elr->lr_sbi;
2678
2679         list_del(&elr->lr_request);
2680         sbi->s_li_request = NULL;
2681         kfree(elr);
2682 }
2683
2684 static void ext4_unregister_li_request(struct super_block *sb)
2685 {
2686         mutex_lock(&ext4_li_mtx);
2687         if (!ext4_li_info) {
2688                 mutex_unlock(&ext4_li_mtx);
2689                 return;
2690         }
2691
2692         mutex_lock(&ext4_li_info->li_list_mtx);
2693         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2694         mutex_unlock(&ext4_li_info->li_list_mtx);
2695         mutex_unlock(&ext4_li_mtx);
2696 }
2697
2698 static struct task_struct *ext4_lazyinit_task;
2699
2700 /*
2701  * This is the function where ext4lazyinit thread lives. It walks
2702  * through the request list searching for next scheduled filesystem.
2703  * When such a fs is found, run the lazy initialization request
2704  * (ext4_rn_li_request) and keep track of the time spend in this
2705  * function. Based on that time we compute next schedule time of
2706  * the request. When walking through the list is complete, compute
2707  * next waking time and put itself into sleep.
2708  */
2709 static int ext4_lazyinit_thread(void *arg)
2710 {
2711         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2712         struct list_head *pos, *n;
2713         struct ext4_li_request *elr;
2714         unsigned long next_wakeup, cur;
2715
2716         BUG_ON(NULL == eli);
2717
2718 cont_thread:
2719         while (true) {
2720                 next_wakeup = MAX_JIFFY_OFFSET;
2721
2722                 mutex_lock(&eli->li_list_mtx);
2723                 if (list_empty(&eli->li_request_list)) {
2724                         mutex_unlock(&eli->li_list_mtx);
2725                         goto exit_thread;
2726                 }
2727
2728                 list_for_each_safe(pos, n, &eli->li_request_list) {
2729                         elr = list_entry(pos, struct ext4_li_request,
2730                                          lr_request);
2731
2732                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2733                                 if (ext4_run_li_request(elr) != 0) {
2734                                         /* error, remove the lazy_init job */
2735                                         ext4_remove_li_request(elr);
2736                                         continue;
2737                                 }
2738                         }
2739
2740                         if (time_before(elr->lr_next_sched, next_wakeup))
2741                                 next_wakeup = elr->lr_next_sched;
2742                 }
2743                 mutex_unlock(&eli->li_list_mtx);
2744
2745                 try_to_freeze();
2746
2747                 cur = jiffies;
2748                 if ((time_after_eq(cur, next_wakeup)) ||
2749                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2750                         cond_resched();
2751                         continue;
2752                 }
2753
2754                 schedule_timeout_interruptible(next_wakeup - cur);
2755
2756                 if (kthread_should_stop()) {
2757                         ext4_clear_request_list();
2758                         goto exit_thread;
2759                 }
2760         }
2761
2762 exit_thread:
2763         /*
2764          * It looks like the request list is empty, but we need
2765          * to check it under the li_list_mtx lock, to prevent any
2766          * additions into it, and of course we should lock ext4_li_mtx
2767          * to atomically free the list and ext4_li_info, because at
2768          * this point another ext4 filesystem could be registering
2769          * new one.
2770          */
2771         mutex_lock(&ext4_li_mtx);
2772         mutex_lock(&eli->li_list_mtx);
2773         if (!list_empty(&eli->li_request_list)) {
2774                 mutex_unlock(&eli->li_list_mtx);
2775                 mutex_unlock(&ext4_li_mtx);
2776                 goto cont_thread;
2777         }
2778         mutex_unlock(&eli->li_list_mtx);
2779         kfree(ext4_li_info);
2780         ext4_li_info = NULL;
2781         mutex_unlock(&ext4_li_mtx);
2782
2783         return 0;
2784 }
2785
2786 static void ext4_clear_request_list(void)
2787 {
2788         struct list_head *pos, *n;
2789         struct ext4_li_request *elr;
2790
2791         mutex_lock(&ext4_li_info->li_list_mtx);
2792         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2793                 elr = list_entry(pos, struct ext4_li_request,
2794                                  lr_request);
2795                 ext4_remove_li_request(elr);
2796         }
2797         mutex_unlock(&ext4_li_info->li_list_mtx);
2798 }
2799
2800 static int ext4_run_lazyinit_thread(void)
2801 {
2802         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2803                                          ext4_li_info, "ext4lazyinit");
2804         if (IS_ERR(ext4_lazyinit_task)) {
2805                 int err = PTR_ERR(ext4_lazyinit_task);
2806                 ext4_clear_request_list();
2807                 kfree(ext4_li_info);
2808                 ext4_li_info = NULL;
2809                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2810                                  "initialization thread\n",
2811                                  err);
2812                 return err;
2813         }
2814         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2815         return 0;
2816 }
2817
2818 /*
2819  * Check whether it make sense to run itable init. thread or not.
2820  * If there is at least one uninitialized inode table, return
2821  * corresponding group number, else the loop goes through all
2822  * groups and return total number of groups.
2823  */
2824 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2825 {
2826         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2827         struct ext4_group_desc *gdp = NULL;
2828
2829         for (group = 0; group < ngroups; group++) {
2830                 gdp = ext4_get_group_desc(sb, group, NULL);
2831                 if (!gdp)
2832                         continue;
2833
2834                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2835                         break;
2836         }
2837
2838         return group;
2839 }
2840
2841 static int ext4_li_info_new(void)
2842 {
2843         struct ext4_lazy_init *eli = NULL;
2844
2845         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2846         if (!eli)
2847                 return -ENOMEM;
2848
2849         INIT_LIST_HEAD(&eli->li_request_list);
2850         mutex_init(&eli->li_list_mtx);
2851
2852         eli->li_state |= EXT4_LAZYINIT_QUIT;
2853
2854         ext4_li_info = eli;
2855
2856         return 0;
2857 }
2858
2859 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2860                                             ext4_group_t start)
2861 {
2862         struct ext4_sb_info *sbi = EXT4_SB(sb);
2863         struct ext4_li_request *elr;
2864
2865         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2866         if (!elr)
2867                 return NULL;
2868
2869         elr->lr_super = sb;
2870         elr->lr_sbi = sbi;
2871         elr->lr_next_group = start;
2872
2873         /*
2874          * Randomize first schedule time of the request to
2875          * spread the inode table initialization requests
2876          * better.
2877          */
2878         elr->lr_next_sched = jiffies + (prandom_u32() %
2879                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
2880         return elr;
2881 }
2882
2883 int ext4_register_li_request(struct super_block *sb,
2884                              ext4_group_t first_not_zeroed)
2885 {
2886         struct ext4_sb_info *sbi = EXT4_SB(sb);
2887         struct ext4_li_request *elr = NULL;
2888         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2889         int ret = 0;
2890
2891         mutex_lock(&ext4_li_mtx);
2892         if (sbi->s_li_request != NULL) {
2893                 /*
2894                  * Reset timeout so it can be computed again, because
2895                  * s_li_wait_mult might have changed.
2896                  */
2897                 sbi->s_li_request->lr_timeout = 0;
2898                 goto out;
2899         }
2900
2901         if (first_not_zeroed == ngroups ||
2902             (sb->s_flags & MS_RDONLY) ||
2903             !test_opt(sb, INIT_INODE_TABLE))
2904                 goto out;
2905
2906         elr = ext4_li_request_new(sb, first_not_zeroed);
2907         if (!elr) {
2908                 ret = -ENOMEM;
2909                 goto out;
2910         }
2911
2912         if (NULL == ext4_li_info) {
2913                 ret = ext4_li_info_new();
2914                 if (ret)
2915                         goto out;
2916         }
2917
2918         mutex_lock(&ext4_li_info->li_list_mtx);
2919         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2920         mutex_unlock(&ext4_li_info->li_list_mtx);
2921
2922         sbi->s_li_request = elr;
2923         /*
2924          * set elr to NULL here since it has been inserted to
2925          * the request_list and the removal and free of it is
2926          * handled by ext4_clear_request_list from now on.
2927          */
2928         elr = NULL;
2929
2930         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
2931                 ret = ext4_run_lazyinit_thread();
2932                 if (ret)
2933                         goto out;
2934         }
2935 out:
2936         mutex_unlock(&ext4_li_mtx);
2937         if (ret)
2938                 kfree(elr);
2939         return ret;
2940 }
2941
2942 /*
2943  * We do not need to lock anything since this is called on
2944  * module unload.
2945  */
2946 static void ext4_destroy_lazyinit_thread(void)
2947 {
2948         /*
2949          * If thread exited earlier
2950          * there's nothing to be done.
2951          */
2952         if (!ext4_li_info || !ext4_lazyinit_task)
2953                 return;
2954
2955         kthread_stop(ext4_lazyinit_task);
2956 }
2957
2958 static int set_journal_csum_feature_set(struct super_block *sb)
2959 {
2960         int ret = 1;
2961         int compat, incompat;
2962         struct ext4_sb_info *sbi = EXT4_SB(sb);
2963
2964         if (ext4_has_metadata_csum(sb)) {
2965                 /* journal checksum v3 */
2966                 compat = 0;
2967                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
2968         } else {
2969                 /* journal checksum v1 */
2970                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
2971                 incompat = 0;
2972         }
2973
2974         jbd2_journal_clear_features(sbi->s_journal,
2975                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2976                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
2977                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
2978         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2979                 ret = jbd2_journal_set_features(sbi->s_journal,
2980                                 compat, 0,
2981                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
2982                                 incompat);
2983         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2984                 ret = jbd2_journal_set_features(sbi->s_journal,
2985                                 compat, 0,
2986                                 incompat);
2987                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2988                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2989         } else {
2990                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2991                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2992         }
2993
2994         return ret;
2995 }
2996
2997 /*
2998  * Note: calculating the overhead so we can be compatible with
2999  * historical BSD practice is quite difficult in the face of
3000  * clusters/bigalloc.  This is because multiple metadata blocks from
3001  * different block group can end up in the same allocation cluster.
3002  * Calculating the exact overhead in the face of clustered allocation
3003  * requires either O(all block bitmaps) in memory or O(number of block
3004  * groups**2) in time.  We will still calculate the superblock for
3005  * older file systems --- and if we come across with a bigalloc file
3006  * system with zero in s_overhead_clusters the estimate will be close to
3007  * correct especially for very large cluster sizes --- but for newer
3008  * file systems, it's better to calculate this figure once at mkfs
3009  * time, and store it in the superblock.  If the superblock value is
3010  * present (even for non-bigalloc file systems), we will use it.
3011  */
3012 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3013                           char *buf)
3014 {
3015         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3016         struct ext4_group_desc  *gdp;
3017         ext4_fsblk_t            first_block, last_block, b;
3018         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3019         int                     s, j, count = 0;
3020
3021         if (!ext4_has_feature_bigalloc(sb))
3022                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3023                         sbi->s_itb_per_group + 2);
3024
3025         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3026                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3027         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3028         for (i = 0; i < ngroups; i++) {
3029                 gdp = ext4_get_group_desc(sb, i, NULL);
3030                 b = ext4_block_bitmap(sb, gdp);
3031                 if (b >= first_block && b <= last_block) {
3032                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3033                         count++;
3034                 }
3035                 b = ext4_inode_bitmap(sb, gdp);
3036                 if (b >= first_block && b <= last_block) {
3037                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3038                         count++;
3039                 }
3040                 b = ext4_inode_table(sb, gdp);
3041                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3042                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3043                                 int c = EXT4_B2C(sbi, b - first_block);
3044                                 ext4_set_bit(c, buf);
3045                                 count++;
3046                         }
3047                 if (i != grp)
3048                         continue;
3049                 s = 0;
3050                 if (ext4_bg_has_super(sb, grp)) {
3051                         ext4_set_bit(s++, buf);
3052                         count++;
3053                 }
3054                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3055                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3056                         count++;
3057                 }
3058         }
3059         if (!count)
3060                 return 0;
3061         return EXT4_CLUSTERS_PER_GROUP(sb) -
3062                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3063 }
3064
3065 /*
3066  * Compute the overhead and stash it in sbi->s_overhead
3067  */
3068 int ext4_calculate_overhead(struct super_block *sb)
3069 {
3070         struct ext4_sb_info *sbi = EXT4_SB(sb);
3071         struct ext4_super_block *es = sbi->s_es;
3072         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3073         ext4_fsblk_t overhead = 0;
3074         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3075
3076         if (!buf)
3077                 return -ENOMEM;
3078
3079         /*
3080          * Compute the overhead (FS structures).  This is constant
3081          * for a given filesystem unless the number of block groups
3082          * changes so we cache the previous value until it does.
3083          */
3084
3085         /*
3086          * All of the blocks before first_data_block are overhead
3087          */
3088         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3089
3090         /*
3091          * Add the overhead found in each block group
3092          */
3093         for (i = 0; i < ngroups; i++) {
3094                 int blks;
3095
3096                 blks = count_overhead(sb, i, buf);
3097                 overhead += blks;
3098                 if (blks)
3099                         memset(buf, 0, PAGE_SIZE);
3100                 cond_resched();
3101         }
3102         /* Add the internal journal blocks as well */
3103         if (sbi->s_journal && !sbi->journal_bdev)
3104                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3105
3106         sbi->s_overhead = overhead;
3107         smp_wmb();
3108         free_page((unsigned long) buf);
3109         return 0;
3110 }
3111
3112 static void ext4_set_resv_clusters(struct super_block *sb)
3113 {
3114         ext4_fsblk_t resv_clusters;
3115         struct ext4_sb_info *sbi = EXT4_SB(sb);
3116
3117         /*
3118          * There's no need to reserve anything when we aren't using extents.
3119          * The space estimates are exact, there are no unwritten extents,
3120          * hole punching doesn't need new metadata... This is needed especially
3121          * to keep ext2/3 backward compatibility.
3122          */
3123         if (!ext4_has_feature_extents(sb))
3124                 return;
3125         /*
3126          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3127          * This should cover the situations where we can not afford to run
3128          * out of space like for example punch hole, or converting
3129          * unwritten extents in delalloc path. In most cases such
3130          * allocation would require 1, or 2 blocks, higher numbers are
3131          * very rare.
3132          */
3133         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3134                          sbi->s_cluster_bits);
3135
3136         do_div(resv_clusters, 50);
3137         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3138
3139         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3140 }
3141
3142 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3143 {
3144         char *orig_data = kstrdup(data, GFP_KERNEL);
3145         struct buffer_head *bh;
3146         struct ext4_super_block *es = NULL;
3147         struct ext4_sb_info *sbi;
3148         ext4_fsblk_t block;
3149         ext4_fsblk_t sb_block = get_sb_block(&data);
3150         ext4_fsblk_t logical_sb_block;
3151         unsigned long offset = 0;
3152         unsigned long journal_devnum = 0;
3153         unsigned long def_mount_opts;
3154         struct inode *root;
3155         const char *descr;
3156         int ret = -ENOMEM;
3157         int blocksize, clustersize;
3158         unsigned int db_count;
3159         unsigned int i;
3160         int needs_recovery, has_huge_files, has_bigalloc;
3161         __u64 blocks_count;
3162         int err = 0;
3163         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3164         ext4_group_t first_not_zeroed;
3165
3166         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3167         if (!sbi)
3168                 goto out_free_orig;
3169
3170         sbi->s_blockgroup_lock =
3171                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3172         if (!sbi->s_blockgroup_lock) {
3173                 kfree(sbi);
3174                 goto out_free_orig;
3175         }
3176         sb->s_fs_info = sbi;
3177         sbi->s_sb = sb;
3178         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3179         sbi->s_sb_block = sb_block;
3180         if (sb->s_bdev->bd_part)
3181                 sbi->s_sectors_written_start =
3182                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3183
3184         /* Cleanup superblock name */
3185         strreplace(sb->s_id, '/', '!');
3186
3187         /* -EINVAL is default */
3188         ret = -EINVAL;
3189         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3190         if (!blocksize) {
3191                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3192                 goto out_fail;
3193         }
3194
3195         /*
3196          * The ext4 superblock will not be buffer aligned for other than 1kB
3197          * block sizes.  We need to calculate the offset from buffer start.
3198          */
3199         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3200                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3201                 offset = do_div(logical_sb_block, blocksize);
3202         } else {
3203                 logical_sb_block = sb_block;
3204         }
3205
3206         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3207                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3208                 goto out_fail;
3209         }
3210         /*
3211          * Note: s_es must be initialized as soon as possible because
3212          *       some ext4 macro-instructions depend on its value
3213          */
3214         es = (struct ext4_super_block *) (bh->b_data + offset);
3215         sbi->s_es = es;
3216         sb->s_magic = le16_to_cpu(es->s_magic);
3217         if (sb->s_magic != EXT4_SUPER_MAGIC)
3218                 goto cantfind_ext4;
3219         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3220
3221         /* Warn if metadata_csum and gdt_csum are both set. */
3222         if (ext4_has_feature_metadata_csum(sb) &&
3223             ext4_has_feature_gdt_csum(sb))
3224                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3225                              "redundant flags; please run fsck.");
3226
3227         /* Check for a known checksum algorithm */
3228         if (!ext4_verify_csum_type(sb, es)) {
3229                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3230                          "unknown checksum algorithm.");
3231                 silent = 1;
3232                 goto cantfind_ext4;
3233         }
3234
3235         /* Load the checksum driver */
3236         if (ext4_has_feature_metadata_csum(sb)) {
3237                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3238                 if (IS_ERR(sbi->s_chksum_driver)) {
3239                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3240                         ret = PTR_ERR(sbi->s_chksum_driver);
3241                         sbi->s_chksum_driver = NULL;
3242                         goto failed_mount;
3243                 }
3244         }
3245
3246         /* Check superblock checksum */
3247         if (!ext4_superblock_csum_verify(sb, es)) {
3248                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3249                          "invalid superblock checksum.  Run e2fsck?");
3250                 silent = 1;
3251                 ret = -EFSBADCRC;
3252                 goto cantfind_ext4;
3253         }
3254
3255         /* Precompute checksum seed for all metadata */
3256         if (ext4_has_feature_csum_seed(sb))
3257                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3258         else if (ext4_has_metadata_csum(sb))
3259                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3260                                                sizeof(es->s_uuid));
3261
3262         /* Set defaults before we parse the mount options */
3263         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3264         set_opt(sb, INIT_INODE_TABLE);
3265         if (def_mount_opts & EXT4_DEFM_DEBUG)
3266                 set_opt(sb, DEBUG);
3267         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3268                 set_opt(sb, GRPID);
3269         if (def_mount_opts & EXT4_DEFM_UID16)
3270                 set_opt(sb, NO_UID32);
3271         /* xattr user namespace & acls are now defaulted on */
3272         set_opt(sb, XATTR_USER);
3273 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3274         set_opt(sb, POSIX_ACL);
3275 #endif
3276         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3277         if (ext4_has_metadata_csum(sb))
3278                 set_opt(sb, JOURNAL_CHECKSUM);
3279
3280         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3281                 set_opt(sb, JOURNAL_DATA);
3282         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3283                 set_opt(sb, ORDERED_DATA);
3284         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3285                 set_opt(sb, WRITEBACK_DATA);
3286
3287         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3288                 set_opt(sb, ERRORS_PANIC);
3289         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3290                 set_opt(sb, ERRORS_CONT);
3291         else
3292                 set_opt(sb, ERRORS_RO);
3293         /* block_validity enabled by default; disable with noblock_validity */
3294         set_opt(sb, BLOCK_VALIDITY);
3295         if (def_mount_opts & EXT4_DEFM_DISCARD)
3296                 set_opt(sb, DISCARD);
3297
3298         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3299         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3300         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3301         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3302         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3303
3304         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3305                 set_opt(sb, BARRIER);
3306
3307         /*
3308          * enable delayed allocation by default
3309          * Use -o nodelalloc to turn it off
3310          */
3311         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3312             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3313                 set_opt(sb, DELALLOC);
3314
3315         /*
3316          * set default s_li_wait_mult for lazyinit, for the case there is
3317          * no mount option specified.
3318          */
3319         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3320
3321         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3322                            &journal_devnum, &journal_ioprio, 0)) {
3323                 ext4_msg(sb, KERN_WARNING,
3324                          "failed to parse options in superblock: %s",
3325                          sbi->s_es->s_mount_opts);
3326         }
3327         sbi->s_def_mount_opt = sbi->s_mount_opt;
3328         if (!parse_options((char *) data, sb, &journal_devnum,
3329                            &journal_ioprio, 0))
3330                 goto failed_mount;
3331
3332         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3333                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3334                             "with data=journal disables delayed "
3335                             "allocation and O_DIRECT support!\n");
3336                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3337                         ext4_msg(sb, KERN_ERR, "can't mount with "
3338                                  "both data=journal and delalloc");
3339                         goto failed_mount;
3340                 }
3341                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3342                         ext4_msg(sb, KERN_ERR, "can't mount with "
3343                                  "both data=journal and dioread_nolock");
3344                         goto failed_mount;
3345                 }
3346                 if (test_opt(sb, DAX)) {
3347                         ext4_msg(sb, KERN_ERR, "can't mount with "
3348                                  "both data=journal and dax");
3349                         goto failed_mount;
3350                 }
3351                 if (test_opt(sb, DELALLOC))
3352                         clear_opt(sb, DELALLOC);
3353         } else {
3354                 sb->s_iflags |= SB_I_CGROUPWB;
3355         }
3356
3357         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3358                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3359
3360         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3361             (ext4_has_compat_features(sb) ||
3362              ext4_has_ro_compat_features(sb) ||
3363              ext4_has_incompat_features(sb)))
3364                 ext4_msg(sb, KERN_WARNING,
3365                        "feature flags set on rev 0 fs, "
3366                        "running e2fsck is recommended");
3367
3368         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3369                 set_opt2(sb, HURD_COMPAT);
3370                 if (ext4_has_feature_64bit(sb)) {
3371                         ext4_msg(sb, KERN_ERR,
3372                                  "The Hurd can't support 64-bit file systems");
3373                         goto failed_mount;
3374                 }
3375         }
3376
3377         if (IS_EXT2_SB(sb)) {
3378                 if (ext2_feature_set_ok(sb))
3379                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3380                                  "using the ext4 subsystem");
3381                 else {
3382                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3383                                  "to feature incompatibilities");
3384                         goto failed_mount;
3385                 }
3386         }
3387
3388         if (IS_EXT3_SB(sb)) {
3389                 if (ext3_feature_set_ok(sb))
3390                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3391                                  "using the ext4 subsystem");
3392                 else {
3393                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3394                                  "to feature incompatibilities");
3395                         goto failed_mount;
3396                 }
3397         }
3398
3399         /*
3400          * Check feature flags regardless of the revision level, since we
3401          * previously didn't change the revision level when setting the flags,
3402          * so there is a chance incompat flags are set on a rev 0 filesystem.
3403          */
3404         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3405                 goto failed_mount;
3406
3407         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3408         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3409             blocksize > EXT4_MAX_BLOCK_SIZE) {
3410                 ext4_msg(sb, KERN_ERR,
3411                        "Unsupported filesystem blocksize %d", blocksize);
3412                 goto failed_mount;
3413         }
3414
3415         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3416                 if (blocksize != PAGE_SIZE) {
3417                         ext4_msg(sb, KERN_ERR,
3418                                         "error: unsupported blocksize for dax");
3419                         goto failed_mount;
3420                 }
3421                 if (!sb->s_bdev->bd_disk->fops->direct_access) {
3422                         ext4_msg(sb, KERN_ERR,
3423                                         "error: device does not support dax");
3424                         goto failed_mount;
3425                 }
3426         }
3427
3428         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3429                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3430                          es->s_encryption_level);
3431                 goto failed_mount;
3432         }
3433
3434         if (sb->s_blocksize != blocksize) {
3435                 /* Validate the filesystem blocksize */
3436                 if (!sb_set_blocksize(sb, blocksize)) {
3437                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3438                                         blocksize);
3439                         goto failed_mount;
3440                 }
3441
3442                 brelse(bh);
3443                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3444                 offset = do_div(logical_sb_block, blocksize);
3445                 bh = sb_bread_unmovable(sb, logical_sb_block);
3446                 if (!bh) {
3447                         ext4_msg(sb, KERN_ERR,
3448                                "Can't read superblock on 2nd try");
3449                         goto failed_mount;
3450                 }
3451                 es = (struct ext4_super_block *)(bh->b_data + offset);
3452                 sbi->s_es = es;
3453                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3454                         ext4_msg(sb, KERN_ERR,
3455                                "Magic mismatch, very weird!");
3456                         goto failed_mount;
3457                 }
3458         }
3459
3460         has_huge_files = ext4_has_feature_huge_file(sb);
3461         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3462                                                       has_huge_files);
3463         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3464
3465         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3466                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3467                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3468         } else {
3469                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3470                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3471                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3472                     (!is_power_of_2(sbi->s_inode_size)) ||
3473                     (sbi->s_inode_size > blocksize)) {
3474                         ext4_msg(sb, KERN_ERR,
3475                                "unsupported inode size: %d",
3476                                sbi->s_inode_size);
3477                         goto failed_mount;
3478                 }
3479                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3480                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3481         }
3482
3483         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3484         if (ext4_has_feature_64bit(sb)) {
3485                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3486                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3487                     !is_power_of_2(sbi->s_desc_size)) {
3488                         ext4_msg(sb, KERN_ERR,
3489                                "unsupported descriptor size %lu",
3490                                sbi->s_desc_size);
3491                         goto failed_mount;
3492                 }
3493         } else
3494                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3495
3496         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3497         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3498         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3499                 goto cantfind_ext4;
3500
3501         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3502         if (sbi->s_inodes_per_block == 0)
3503                 goto cantfind_ext4;
3504         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3505                                         sbi->s_inodes_per_block;
3506         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3507         sbi->s_sbh = bh;
3508         sbi->s_mount_state = le16_to_cpu(es->s_state);
3509         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3510         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3511
3512         for (i = 0; i < 4; i++)
3513                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3514         sbi->s_def_hash_version = es->s_def_hash_version;
3515         if (ext4_has_feature_dir_index(sb)) {
3516                 i = le32_to_cpu(es->s_flags);
3517                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3518                         sbi->s_hash_unsigned = 3;
3519                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3520 #ifdef __CHAR_UNSIGNED__
3521                         if (!(sb->s_flags & MS_RDONLY))
3522                                 es->s_flags |=
3523                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3524                         sbi->s_hash_unsigned = 3;
3525 #else
3526                         if (!(sb->s_flags & MS_RDONLY))
3527                                 es->s_flags |=
3528                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3529 #endif
3530                 }
3531         }
3532
3533         /* Handle clustersize */
3534         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3535         has_bigalloc = ext4_has_feature_bigalloc(sb);
3536         if (has_bigalloc) {
3537                 if (clustersize < blocksize) {
3538                         ext4_msg(sb, KERN_ERR,
3539                                  "cluster size (%d) smaller than "
3540                                  "block size (%d)", clustersize, blocksize);
3541                         goto failed_mount;
3542                 }
3543                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3544                         le32_to_cpu(es->s_log_block_size);
3545                 sbi->s_clusters_per_group =
3546                         le32_to_cpu(es->s_clusters_per_group);
3547                 if (sbi->s_clusters_per_group > blocksize * 8) {
3548                         ext4_msg(sb, KERN_ERR,
3549                                  "#clusters per group too big: %lu",
3550                                  sbi->s_clusters_per_group);
3551                         goto failed_mount;
3552                 }
3553                 if (sbi->s_blocks_per_group !=
3554                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3555                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3556                                  "clusters per group (%lu) inconsistent",
3557                                  sbi->s_blocks_per_group,
3558                                  sbi->s_clusters_per_group);
3559                         goto failed_mount;
3560                 }
3561         } else {
3562                 if (clustersize != blocksize) {
3563                         ext4_warning(sb, "fragment/cluster size (%d) != "
3564                                      "block size (%d)", clustersize,
3565                                      blocksize);
3566                         clustersize = blocksize;
3567                 }
3568                 if (sbi->s_blocks_per_group > blocksize * 8) {
3569                         ext4_msg(sb, KERN_ERR,
3570                                  "#blocks per group too big: %lu",
3571                                  sbi->s_blocks_per_group);
3572                         goto failed_mount;
3573                 }
3574                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3575                 sbi->s_cluster_bits = 0;
3576         }
3577         sbi->s_cluster_ratio = clustersize / blocksize;
3578
3579         if (sbi->s_inodes_per_group > blocksize * 8) {
3580                 ext4_msg(sb, KERN_ERR,
3581                        "#inodes per group too big: %lu",
3582                        sbi->s_inodes_per_group);
3583                 goto failed_mount;
3584         }
3585
3586         /* Do we have standard group size of clustersize * 8 blocks ? */
3587         if (sbi->s_blocks_per_group == clustersize << 3)
3588                 set_opt2(sb, STD_GROUP_SIZE);
3589
3590         /*
3591          * Test whether we have more sectors than will fit in sector_t,
3592          * and whether the max offset is addressable by the page cache.
3593          */
3594         err = generic_check_addressable(sb->s_blocksize_bits,
3595                                         ext4_blocks_count(es));
3596         if (err) {
3597                 ext4_msg(sb, KERN_ERR, "filesystem"
3598                          " too large to mount safely on this system");
3599                 if (sizeof(sector_t) < 8)
3600                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3601                 goto failed_mount;
3602         }
3603
3604         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3605                 goto cantfind_ext4;
3606
3607         /* check blocks count against device size */
3608         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3609         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3610                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3611                        "exceeds size of device (%llu blocks)",
3612                        ext4_blocks_count(es), blocks_count);
3613                 goto failed_mount;
3614         }
3615
3616         /*
3617          * It makes no sense for the first data block to be beyond the end
3618          * of the filesystem.
3619          */
3620         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3621                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3622                          "block %u is beyond end of filesystem (%llu)",
3623                          le32_to_cpu(es->s_first_data_block),
3624                          ext4_blocks_count(es));
3625                 goto failed_mount;
3626         }
3627         blocks_count = (ext4_blocks_count(es) -
3628                         le32_to_cpu(es->s_first_data_block) +
3629                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3630         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3631         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3632                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3633                        "(block count %llu, first data block %u, "
3634                        "blocks per group %lu)", sbi->s_groups_count,
3635                        ext4_blocks_count(es),
3636                        le32_to_cpu(es->s_first_data_block),
3637                        EXT4_BLOCKS_PER_GROUP(sb));
3638                 goto failed_mount;
3639         }
3640         sbi->s_groups_count = blocks_count;
3641         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3642                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3643         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3644                    EXT4_DESC_PER_BLOCK(sb);
3645         sbi->s_group_desc = ext4_kvmalloc(db_count *
3646                                           sizeof(struct buffer_head *),
3647                                           GFP_KERNEL);
3648         if (sbi->s_group_desc == NULL) {
3649                 ext4_msg(sb, KERN_ERR, "not enough memory");
3650                 ret = -ENOMEM;
3651                 goto failed_mount;
3652         }
3653
3654         bgl_lock_init(sbi->s_blockgroup_lock);
3655
3656         for (i = 0; i < db_count; i++) {
3657                 block = descriptor_loc(sb, logical_sb_block, i);
3658                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
3659                 if (!sbi->s_group_desc[i]) {
3660                         ext4_msg(sb, KERN_ERR,
3661                                "can't read group descriptor %d", i);
3662                         db_count = i;
3663                         goto failed_mount2;
3664                 }
3665         }
3666         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3667                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3668                 ret = -EFSCORRUPTED;
3669                 goto failed_mount2;
3670         }
3671
3672         sbi->s_gdb_count = db_count;
3673         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3674         spin_lock_init(&sbi->s_next_gen_lock);
3675
3676         setup_timer(&sbi->s_err_report, print_daily_error_info,
3677                 (unsigned long) sb);
3678
3679         /* Register extent status tree shrinker */
3680         if (ext4_es_register_shrinker(sbi))
3681                 goto failed_mount3;
3682
3683         sbi->s_stripe = ext4_get_stripe_size(sbi);
3684         sbi->s_extent_max_zeroout_kb = 32;
3685
3686         /*
3687          * set up enough so that it can read an inode
3688          */
3689         sb->s_op = &ext4_sops;
3690         sb->s_export_op = &ext4_export_ops;
3691         sb->s_xattr = ext4_xattr_handlers;
3692 #ifdef CONFIG_QUOTA
3693         sb->dq_op = &ext4_quota_operations;
3694         if (ext4_has_feature_quota(sb))
3695                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
3696         else
3697                 sb->s_qcop = &ext4_qctl_operations;
3698         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
3699 #endif
3700         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3701
3702         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3703         mutex_init(&sbi->s_orphan_lock);
3704
3705         sb->s_root = NULL;
3706
3707         needs_recovery = (es->s_last_orphan != 0 ||
3708                           ext4_has_feature_journal_needs_recovery(sb));
3709
3710         if (ext4_has_feature_mmp(sb) && !(sb->s_flags & MS_RDONLY))
3711                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3712                         goto failed_mount3a;
3713
3714         /*
3715          * The first inode we look at is the journal inode.  Don't try
3716          * root first: it may be modified in the journal!
3717          */
3718         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
3719                 if (ext4_load_journal(sb, es, journal_devnum))
3720                         goto failed_mount3a;
3721         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3722                    ext4_has_feature_journal_needs_recovery(sb)) {
3723                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3724                        "suppressed and not mounted read-only");
3725                 goto failed_mount_wq;
3726         } else {
3727                 /* Nojournal mode, all journal mount options are illegal */
3728                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
3729                         ext4_msg(sb, KERN_ERR, "can't mount with "
3730                                  "journal_checksum, fs mounted w/o journal");
3731                         goto failed_mount_wq;
3732                 }
3733                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3734                         ext4_msg(sb, KERN_ERR, "can't mount with "
3735                                  "journal_async_commit, fs mounted w/o journal");
3736                         goto failed_mount_wq;
3737                 }
3738                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
3739                         ext4_msg(sb, KERN_ERR, "can't mount with "
3740                                  "commit=%lu, fs mounted w/o journal",
3741                                  sbi->s_commit_interval / HZ);
3742                         goto failed_mount_wq;
3743                 }
3744                 if (EXT4_MOUNT_DATA_FLAGS &
3745                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
3746                         ext4_msg(sb, KERN_ERR, "can't mount with "
3747                                  "data=, fs mounted w/o journal");
3748                         goto failed_mount_wq;
3749                 }
3750                 sbi->s_def_mount_opt &= EXT4_MOUNT_JOURNAL_CHECKSUM;
3751                 clear_opt(sb, JOURNAL_CHECKSUM);
3752                 clear_opt(sb, DATA_FLAGS);
3753                 sbi->s_journal = NULL;
3754                 needs_recovery = 0;
3755                 goto no_journal;
3756         }
3757
3758         if (ext4_has_feature_64bit(sb) &&
3759             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3760                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3761                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3762                 goto failed_mount_wq;
3763         }
3764
3765         if (!set_journal_csum_feature_set(sb)) {
3766                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3767                          "feature set");
3768                 goto failed_mount_wq;
3769         }
3770
3771         /* We have now updated the journal if required, so we can
3772          * validate the data journaling mode. */
3773         switch (test_opt(sb, DATA_FLAGS)) {
3774         case 0:
3775                 /* No mode set, assume a default based on the journal
3776                  * capabilities: ORDERED_DATA if the journal can
3777                  * cope, else JOURNAL_DATA
3778                  */
3779                 if (jbd2_journal_check_available_features
3780                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3781                         set_opt(sb, ORDERED_DATA);
3782                 else
3783                         set_opt(sb, JOURNAL_DATA);
3784                 break;
3785
3786         case EXT4_MOUNT_ORDERED_DATA:
3787         case EXT4_MOUNT_WRITEBACK_DATA:
3788                 if (!jbd2_journal_check_available_features
3789                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3790                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3791                                "requested data journaling mode");
3792                         goto failed_mount_wq;
3793                 }
3794         default:
3795                 break;
3796         }
3797         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3798
3799         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3800
3801 no_journal:
3802         sbi->s_mb_cache = ext4_xattr_create_cache();
3803         if (!sbi->s_mb_cache) {
3804                 ext4_msg(sb, KERN_ERR, "Failed to create an mb_cache");
3805                 goto failed_mount_wq;
3806         }
3807
3808         if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
3809             (blocksize != PAGE_CACHE_SIZE)) {
3810                 ext4_msg(sb, KERN_ERR,
3811                          "Unsupported blocksize for fs encryption");
3812                 goto failed_mount_wq;
3813         }
3814
3815         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !(sb->s_flags & MS_RDONLY) &&
3816             !ext4_has_feature_encrypt(sb)) {
3817                 ext4_set_feature_encrypt(sb);
3818                 ext4_commit_super(sb, 1);
3819         }
3820
3821         /*
3822          * Get the # of file system overhead blocks from the
3823          * superblock if present.
3824          */
3825         if (es->s_overhead_clusters)
3826                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3827         else {
3828                 err = ext4_calculate_overhead(sb);
3829                 if (err)
3830                         goto failed_mount_wq;
3831         }
3832
3833         /*
3834          * The maximum number of concurrent works can be high and
3835          * concurrency isn't really necessary.  Limit it to 1.
3836          */
3837         EXT4_SB(sb)->rsv_conversion_wq =
3838                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3839         if (!EXT4_SB(sb)->rsv_conversion_wq) {
3840                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
3841                 ret = -ENOMEM;
3842                 goto failed_mount4;
3843         }
3844
3845         /*
3846          * The jbd2_journal_load will have done any necessary log recovery,
3847          * so we can safely mount the rest of the filesystem now.
3848          */
3849
3850         root = ext4_iget(sb, EXT4_ROOT_INO);
3851         if (IS_ERR(root)) {
3852                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3853                 ret = PTR_ERR(root);
3854                 root = NULL;
3855                 goto failed_mount4;
3856         }
3857         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3858                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3859                 iput(root);
3860                 goto failed_mount4;
3861         }
3862         sb->s_root = d_make_root(root);
3863         if (!sb->s_root) {
3864                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3865                 ret = -ENOMEM;
3866                 goto failed_mount4;
3867         }
3868
3869         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3870                 sb->s_flags |= MS_RDONLY;
3871
3872         /* determine the minimum size of new large inodes, if present */
3873         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3874                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3875                                                      EXT4_GOOD_OLD_INODE_SIZE;
3876                 if (ext4_has_feature_extra_isize(sb)) {
3877                         if (sbi->s_want_extra_isize <
3878                             le16_to_cpu(es->s_want_extra_isize))
3879                                 sbi->s_want_extra_isize =
3880                                         le16_to_cpu(es->s_want_extra_isize);
3881                         if (sbi->s_want_extra_isize <
3882                             le16_to_cpu(es->s_min_extra_isize))
3883                                 sbi->s_want_extra_isize =
3884                                         le16_to_cpu(es->s_min_extra_isize);
3885                 }
3886         }
3887         /* Check if enough inode space is available */
3888         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3889                                                         sbi->s_inode_size) {
3890                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3891                                                        EXT4_GOOD_OLD_INODE_SIZE;
3892                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3893                          "available");
3894         }
3895
3896         ext4_set_resv_clusters(sb);
3897
3898         err = ext4_setup_system_zone(sb);
3899         if (err) {
3900                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3901                          "zone (%d)", err);
3902                 goto failed_mount4a;
3903         }
3904
3905         ext4_ext_init(sb);
3906         err = ext4_mb_init(sb);
3907         if (err) {
3908                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3909                          err);
3910                 goto failed_mount5;
3911         }
3912
3913         block = ext4_count_free_clusters(sb);
3914         ext4_free_blocks_count_set(sbi->s_es,
3915                                    EXT4_C2B(sbi, block));
3916         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
3917                                   GFP_KERNEL);
3918         if (!err) {
3919                 unsigned long freei = ext4_count_free_inodes(sb);
3920                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
3921                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
3922                                           GFP_KERNEL);
3923         }
3924         if (!err)
3925                 err = percpu_counter_init(&sbi->s_dirs_counter,
3926                                           ext4_count_dirs(sb), GFP_KERNEL);
3927         if (!err)
3928                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
3929                                           GFP_KERNEL);
3930         if (err) {
3931                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3932                 goto failed_mount6;
3933         }
3934
3935         if (ext4_has_feature_flex_bg(sb))
3936                 if (!ext4_fill_flex_info(sb)) {
3937                         ext4_msg(sb, KERN_ERR,
3938                                "unable to initialize "
3939                                "flex_bg meta info!");
3940                         goto failed_mount6;
3941                 }
3942
3943         err = ext4_register_li_request(sb, first_not_zeroed);
3944         if (err)
3945                 goto failed_mount6;
3946
3947         err = ext4_register_sysfs(sb);
3948         if (err)
3949                 goto failed_mount7;
3950
3951 #ifdef CONFIG_QUOTA
3952         /* Enable quota usage during mount. */
3953         if (ext4_has_feature_quota(sb) && !(sb->s_flags & MS_RDONLY)) {
3954                 err = ext4_enable_quotas(sb);
3955                 if (err)
3956                         goto failed_mount8;
3957         }
3958 #endif  /* CONFIG_QUOTA */
3959
3960         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3961         ext4_orphan_cleanup(sb, es);
3962         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3963         if (needs_recovery) {
3964                 ext4_msg(sb, KERN_INFO, "recovery complete");
3965                 ext4_mark_recovery_complete(sb, es);
3966         }
3967         if (EXT4_SB(sb)->s_journal) {
3968                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3969                         descr = " journalled data mode";
3970                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3971                         descr = " ordered data mode";
3972                 else
3973                         descr = " writeback data mode";
3974         } else
3975                 descr = "out journal";
3976
3977         if (test_opt(sb, DISCARD)) {
3978                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
3979                 if (!blk_queue_discard(q))
3980                         ext4_msg(sb, KERN_WARNING,
3981                                  "mounting with \"discard\" option, but "
3982                                  "the device does not support discard");
3983         }
3984
3985         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
3986                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3987                          "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3988                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3989
3990         if (es->s_error_count)
3991                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3992
3993         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
3994         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
3995         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
3996         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
3997
3998         kfree(orig_data);
3999         return 0;
4000
4001 cantfind_ext4:
4002         if (!silent)
4003                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4004         goto failed_mount;
4005
4006 #ifdef CONFIG_QUOTA
4007 failed_mount8:
4008         ext4_unregister_sysfs(sb);
4009 #endif
4010 failed_mount7:
4011         ext4_unregister_li_request(sb);
4012 failed_mount6:
4013         ext4_mb_release(sb);
4014         if (sbi->s_flex_groups)
4015                 kvfree(sbi->s_flex_groups);
4016         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4017         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4018         percpu_counter_destroy(&sbi->s_dirs_counter);
4019         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4020 failed_mount5:
4021         ext4_ext_release(sb);
4022         ext4_release_system_zone(sb);
4023 failed_mount4a:
4024         dput(sb->s_root);
4025         sb->s_root = NULL;
4026 failed_mount4:
4027         ext4_msg(sb, KERN_ERR, "mount failed");
4028         if (EXT4_SB(sb)->rsv_conversion_wq)
4029                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4030 failed_mount_wq:
4031         if (sbi->s_mb_cache) {
4032                 ext4_xattr_destroy_cache(sbi->s_mb_cache);
4033                 sbi->s_mb_cache = NULL;
4034         }
4035         if (sbi->s_journal) {
4036                 jbd2_journal_destroy(sbi->s_journal);
4037                 sbi->s_journal = NULL;
4038         }
4039 failed_mount3a:
4040         ext4_es_unregister_shrinker(sbi);
4041 failed_mount3:
4042         del_timer_sync(&sbi->s_err_report);
4043         if (sbi->s_mmp_tsk)
4044                 kthread_stop(sbi->s_mmp_tsk);
4045 failed_mount2:
4046         for (i = 0; i < db_count; i++)
4047                 brelse(sbi->s_group_desc[i]);
4048         kvfree(sbi->s_group_desc);
4049 failed_mount:
4050         if (sbi->s_chksum_driver)
4051                 crypto_free_shash(sbi->s_chksum_driver);
4052 #ifdef CONFIG_QUOTA
4053         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4054                 kfree(sbi->s_qf_names[i]);
4055 #endif
4056         ext4_blkdev_remove(sbi);
4057         brelse(bh);
4058 out_fail:
4059         sb->s_fs_info = NULL;
4060         kfree(sbi->s_blockgroup_lock);
4061         kfree(sbi);
4062 out_free_orig:
4063         kfree(orig_data);
4064         return err ? err : ret;
4065 }
4066
4067 /*
4068  * Setup any per-fs journal parameters now.  We'll do this both on
4069  * initial mount, once the journal has been initialised but before we've
4070  * done any recovery; and again on any subsequent remount.
4071  */
4072 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4073 {
4074         struct ext4_sb_info *sbi = EXT4_SB(sb);
4075
4076         journal->j_commit_interval = sbi->s_commit_interval;
4077         journal->j_min_batch_time = sbi->s_min_batch_time;
4078         journal->j_max_batch_time = sbi->s_max_batch_time;
4079
4080         write_lock(&journal->j_state_lock);
4081         if (test_opt(sb, BARRIER))
4082                 journal->j_flags |= JBD2_BARRIER;
4083         else
4084                 journal->j_flags &= ~JBD2_BARRIER;
4085         if (test_opt(sb, DATA_ERR_ABORT))
4086                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4087         else
4088                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4089         write_unlock(&journal->j_state_lock);
4090 }
4091
4092 static journal_t *ext4_get_journal(struct super_block *sb,
4093                                    unsigned int journal_inum)
4094 {
4095         struct inode *journal_inode;
4096         journal_t *journal;
4097
4098         BUG_ON(!ext4_has_feature_journal(sb));
4099
4100         /* First, test for the existence of a valid inode on disk.  Bad
4101          * things happen if we iget() an unused inode, as the subsequent
4102          * iput() will try to delete it. */
4103
4104         journal_inode = ext4_iget(sb, journal_inum);
4105         if (IS_ERR(journal_inode)) {
4106                 ext4_msg(sb, KERN_ERR, "no journal found");
4107                 return NULL;
4108         }
4109         if (!journal_inode->i_nlink) {
4110                 make_bad_inode(journal_inode);
4111                 iput(journal_inode);
4112                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4113                 return NULL;
4114         }
4115
4116         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4117                   journal_inode, journal_inode->i_size);
4118         if (!S_ISREG(journal_inode->i_mode)) {
4119                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4120                 iput(journal_inode);
4121                 return NULL;
4122         }
4123
4124         journal = jbd2_journal_init_inode(journal_inode);
4125         if (!journal) {
4126                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4127                 iput(journal_inode);
4128                 return NULL;
4129         }
4130         journal->j_private = sb;
4131         ext4_init_journal_params(sb, journal);
4132         return journal;
4133 }
4134
4135 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4136                                        dev_t j_dev)
4137 {
4138         struct buffer_head *bh;
4139         journal_t *journal;
4140         ext4_fsblk_t start;
4141         ext4_fsblk_t len;
4142         int hblock, blocksize;
4143         ext4_fsblk_t sb_block;
4144         unsigned long offset;
4145         struct ext4_super_block *es;
4146         struct block_device *bdev;
4147
4148         BUG_ON(!ext4_has_feature_journal(sb));
4149
4150         bdev = ext4_blkdev_get(j_dev, sb);
4151         if (bdev == NULL)
4152                 return NULL;
4153
4154         blocksize = sb->s_blocksize;
4155         hblock = bdev_logical_block_size(bdev);
4156         if (blocksize < hblock) {
4157                 ext4_msg(sb, KERN_ERR,
4158                         "blocksize too small for journal device");
4159                 goto out_bdev;
4160         }
4161
4162         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4163         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4164         set_blocksize(bdev, blocksize);
4165         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4166                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4167                        "external journal");
4168                 goto out_bdev;
4169         }
4170
4171         es = (struct ext4_super_block *) (bh->b_data + offset);
4172         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4173             !(le32_to_cpu(es->s_feature_incompat) &
4174               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4175                 ext4_msg(sb, KERN_ERR, "external journal has "
4176                                         "bad superblock");
4177                 brelse(bh);
4178                 goto out_bdev;
4179         }
4180
4181         if ((le32_to_cpu(es->s_feature_ro_compat) &
4182              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4183             es->s_checksum != ext4_superblock_csum(sb, es)) {
4184                 ext4_msg(sb, KERN_ERR, "external journal has "
4185                                        "corrupt superblock");
4186                 brelse(bh);
4187                 goto out_bdev;
4188         }
4189
4190         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4191                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4192                 brelse(bh);
4193                 goto out_bdev;
4194         }
4195
4196         len = ext4_blocks_count(es);
4197         start = sb_block + 1;
4198         brelse(bh);     /* we're done with the superblock */
4199
4200         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4201                                         start, len, blocksize);
4202         if (!journal) {
4203                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4204                 goto out_bdev;
4205         }
4206         journal->j_private = sb;
4207         ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4208         wait_on_buffer(journal->j_sb_buffer);
4209         if (!buffer_uptodate(journal->j_sb_buffer)) {
4210                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4211                 goto out_journal;
4212         }
4213         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4214                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4215                                         "user (unsupported) - %d",
4216                         be32_to_cpu(journal->j_superblock->s_nr_users));
4217                 goto out_journal;
4218         }
4219         EXT4_SB(sb)->journal_bdev = bdev;
4220         ext4_init_journal_params(sb, journal);
4221         return journal;
4222
4223 out_journal:
4224         jbd2_journal_destroy(journal);
4225 out_bdev:
4226         ext4_blkdev_put(bdev);
4227         return NULL;
4228 }
4229
4230 static int ext4_load_journal(struct super_block *sb,
4231                              struct ext4_super_block *es,
4232                              unsigned long journal_devnum)
4233 {
4234         journal_t *journal;
4235         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4236         dev_t journal_dev;
4237         int err = 0;
4238         int really_read_only;
4239
4240         BUG_ON(!ext4_has_feature_journal(sb));
4241
4242         if (journal_devnum &&
4243             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4244                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4245                         "numbers have changed");
4246                 journal_dev = new_decode_dev(journal_devnum);
4247         } else
4248                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4249
4250         really_read_only = bdev_read_only(sb->s_bdev);
4251
4252         /*
4253          * Are we loading a blank journal or performing recovery after a
4254          * crash?  For recovery, we need to check in advance whether we
4255          * can get read-write access to the device.
4256          */
4257         if (ext4_has_feature_journal_needs_recovery(sb)) {
4258                 if (sb->s_flags & MS_RDONLY) {
4259                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4260                                         "required on readonly filesystem");
4261                         if (really_read_only) {
4262                                 ext4_msg(sb, KERN_ERR, "write access "
4263                                         "unavailable, cannot proceed");
4264                                 return -EROFS;
4265                         }
4266                         ext4_msg(sb, KERN_INFO, "write access will "
4267                                "be enabled during recovery");
4268                 }
4269         }
4270
4271         if (journal_inum && journal_dev) {
4272                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4273                        "and inode journals!");
4274                 return -EINVAL;
4275         }
4276
4277         if (journal_inum) {
4278                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4279                         return -EINVAL;
4280         } else {
4281                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4282                         return -EINVAL;
4283         }
4284
4285         if (!(journal->j_flags & JBD2_BARRIER))
4286                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4287
4288         if (!ext4_has_feature_journal_needs_recovery(sb))
4289                 err = jbd2_journal_wipe(journal, !really_read_only);
4290         if (!err) {
4291                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4292                 if (save)
4293                         memcpy(save, ((char *) es) +
4294                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4295                 err = jbd2_journal_load(journal);
4296                 if (save)
4297                         memcpy(((char *) es) + EXT4_S_ERR_START,
4298                                save, EXT4_S_ERR_LEN);
4299                 kfree(save);
4300         }
4301
4302         if (err) {
4303                 ext4_msg(sb, KERN_ERR, "error loading journal");
4304                 jbd2_journal_destroy(journal);
4305                 return err;
4306         }
4307
4308         EXT4_SB(sb)->s_journal = journal;
4309         ext4_clear_journal_err(sb, es);
4310
4311         if (!really_read_only && journal_devnum &&
4312             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4313                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4314
4315                 /* Make sure we flush the recovery flag to disk. */
4316                 ext4_commit_super(sb, 1);
4317         }
4318
4319         return 0;
4320 }
4321
4322 static int ext4_commit_super(struct super_block *sb, int sync)
4323 {
4324         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4325         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4326         int error = 0;
4327
4328         if (!sbh || block_device_ejected(sb))
4329                 return error;
4330         if (buffer_write_io_error(sbh)) {
4331                 /*
4332                  * Oh, dear.  A previous attempt to write the
4333                  * superblock failed.  This could happen because the
4334                  * USB device was yanked out.  Or it could happen to
4335                  * be a transient write error and maybe the block will
4336                  * be remapped.  Nothing we can do but to retry the
4337                  * write and hope for the best.
4338                  */
4339                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4340                        "superblock detected");
4341                 clear_buffer_write_io_error(sbh);
4342                 set_buffer_uptodate(sbh);
4343         }
4344         /*
4345          * If the file system is mounted read-only, don't update the
4346          * superblock write time.  This avoids updating the superblock
4347          * write time when we are mounting the root file system
4348          * read/only but we need to replay the journal; at that point,
4349          * for people who are east of GMT and who make their clock
4350          * tick in localtime for Windows bug-for-bug compatibility,
4351          * the clock is set in the future, and this will cause e2fsck
4352          * to complain and force a full file system check.
4353          */
4354         if (!(sb->s_flags & MS_RDONLY))
4355                 es->s_wtime = cpu_to_le32(get_seconds());
4356         if (sb->s_bdev->bd_part)
4357                 es->s_kbytes_written =
4358                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4359                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4360                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4361         else
4362                 es->s_kbytes_written =
4363                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4364         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4365                 ext4_free_blocks_count_set(es,
4366                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4367                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4368         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4369                 es->s_free_inodes_count =
4370                         cpu_to_le32(percpu_counter_sum_positive(
4371                                 &EXT4_SB(sb)->s_freeinodes_counter));
4372         BUFFER_TRACE(sbh, "marking dirty");
4373         ext4_superblock_csum_set(sb);
4374         mark_buffer_dirty(sbh);
4375         if (sync) {
4376                 error = __sync_dirty_buffer(sbh,
4377                         test_opt(sb, BARRIER) ? WRITE_FUA : WRITE_SYNC);
4378                 if (error)
4379                         return error;
4380
4381                 error = buffer_write_io_error(sbh);
4382                 if (error) {
4383                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4384                                "superblock");
4385                         clear_buffer_write_io_error(sbh);
4386                         set_buffer_uptodate(sbh);
4387                 }
4388         }
4389         return error;
4390 }
4391
4392 /*
4393  * Have we just finished recovery?  If so, and if we are mounting (or
4394  * remounting) the filesystem readonly, then we will end up with a
4395  * consistent fs on disk.  Record that fact.
4396  */
4397 static void ext4_mark_recovery_complete(struct super_block *sb,
4398                                         struct ext4_super_block *es)
4399 {
4400         journal_t *journal = EXT4_SB(sb)->s_journal;
4401
4402         if (!ext4_has_feature_journal(sb)) {
4403                 BUG_ON(journal != NULL);
4404                 return;
4405         }
4406         jbd2_journal_lock_updates(journal);
4407         if (jbd2_journal_flush(journal) < 0)
4408                 goto out;
4409
4410         if (ext4_has_feature_journal_needs_recovery(sb) &&
4411             sb->s_flags & MS_RDONLY) {
4412                 ext4_clear_feature_journal_needs_recovery(sb);
4413                 ext4_commit_super(sb, 1);
4414         }
4415
4416 out:
4417         jbd2_journal_unlock_updates(journal);
4418 }
4419
4420 /*
4421  * If we are mounting (or read-write remounting) a filesystem whose journal
4422  * has recorded an error from a previous lifetime, move that error to the
4423  * main filesystem now.
4424  */
4425 static void ext4_clear_journal_err(struct super_block *sb,
4426                                    struct ext4_super_block *es)
4427 {
4428         journal_t *journal;
4429         int j_errno;
4430         const char *errstr;
4431
4432         BUG_ON(!ext4_has_feature_journal(sb));
4433
4434         journal = EXT4_SB(sb)->s_journal;
4435
4436         /*
4437          * Now check for any error status which may have been recorded in the
4438          * journal by a prior ext4_error() or ext4_abort()
4439          */
4440
4441         j_errno = jbd2_journal_errno(journal);
4442         if (j_errno) {
4443                 char nbuf[16];
4444
4445                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4446                 ext4_warning(sb, "Filesystem error recorded "
4447                              "from previous mount: %s", errstr);
4448                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4449
4450                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4451                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4452                 ext4_commit_super(sb, 1);
4453
4454                 jbd2_journal_clear_err(journal);
4455                 jbd2_journal_update_sb_errno(journal);
4456         }
4457 }
4458
4459 /*
4460  * Force the running and committing transactions to commit,
4461  * and wait on the commit.
4462  */
4463 int ext4_force_commit(struct super_block *sb)
4464 {
4465         journal_t *journal;
4466
4467         if (sb->s_flags & MS_RDONLY)
4468                 return 0;
4469
4470         journal = EXT4_SB(sb)->s_journal;
4471         return ext4_journal_force_commit(journal);
4472 }
4473
4474 static int ext4_sync_fs(struct super_block *sb, int wait)
4475 {
4476         int ret = 0;
4477         tid_t target;
4478         bool needs_barrier = false;
4479         struct ext4_sb_info *sbi = EXT4_SB(sb);
4480
4481         trace_ext4_sync_fs(sb, wait);
4482         flush_workqueue(sbi->rsv_conversion_wq);
4483         /*
4484          * Writeback quota in non-journalled quota case - journalled quota has
4485          * no dirty dquots
4486          */
4487         dquot_writeback_dquots(sb, -1);
4488         /*
4489          * Data writeback is possible w/o journal transaction, so barrier must
4490          * being sent at the end of the function. But we can skip it if
4491          * transaction_commit will do it for us.
4492          */
4493         if (sbi->s_journal) {
4494                 target = jbd2_get_latest_transaction(sbi->s_journal);
4495                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4496                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4497                         needs_barrier = true;
4498
4499                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4500                         if (wait)
4501                                 ret = jbd2_log_wait_commit(sbi->s_journal,
4502                                                            target);
4503                 }
4504         } else if (wait && test_opt(sb, BARRIER))
4505                 needs_barrier = true;
4506         if (needs_barrier) {
4507                 int err;
4508                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4509                 if (!ret)
4510                         ret = err;
4511         }
4512
4513         return ret;
4514 }
4515
4516 /*
4517  * LVM calls this function before a (read-only) snapshot is created.  This
4518  * gives us a chance to flush the journal completely and mark the fs clean.
4519  *
4520  * Note that only this function cannot bring a filesystem to be in a clean
4521  * state independently. It relies on upper layer to stop all data & metadata
4522  * modifications.
4523  */
4524 static int ext4_freeze(struct super_block *sb)
4525 {
4526         int error = 0;
4527         journal_t *journal;
4528
4529         if (sb->s_flags & MS_RDONLY)
4530                 return 0;
4531
4532         journal = EXT4_SB(sb)->s_journal;
4533
4534         if (journal) {
4535                 /* Now we set up the journal barrier. */
4536                 jbd2_journal_lock_updates(journal);
4537
4538                 /*
4539                  * Don't clear the needs_recovery flag if we failed to
4540                  * flush the journal.
4541                  */
4542                 error = jbd2_journal_flush(journal);
4543                 if (error < 0)
4544                         goto out;
4545
4546                 /* Journal blocked and flushed, clear needs_recovery flag. */
4547                 ext4_clear_feature_journal_needs_recovery(sb);
4548         }
4549
4550         error = ext4_commit_super(sb, 1);
4551 out:
4552         if (journal)
4553                 /* we rely on upper layer to stop further updates */
4554                 jbd2_journal_unlock_updates(journal);
4555         return error;
4556 }
4557
4558 /*
4559  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4560  * flag here, even though the filesystem is not technically dirty yet.
4561  */
4562 static int ext4_unfreeze(struct super_block *sb)
4563 {
4564         if (sb->s_flags & MS_RDONLY)
4565                 return 0;
4566
4567         if (EXT4_SB(sb)->s_journal) {
4568                 /* Reset the needs_recovery flag before the fs is unlocked. */
4569                 ext4_set_feature_journal_needs_recovery(sb);
4570         }
4571
4572         ext4_commit_super(sb, 1);
4573         return 0;
4574 }
4575
4576 /*
4577  * Structure to save mount options for ext4_remount's benefit
4578  */
4579 struct ext4_mount_options {
4580         unsigned long s_mount_opt;
4581         unsigned long s_mount_opt2;
4582         kuid_t s_resuid;
4583         kgid_t s_resgid;
4584         unsigned long s_commit_interval;
4585         u32 s_min_batch_time, s_max_batch_time;
4586 #ifdef CONFIG_QUOTA
4587         int s_jquota_fmt;
4588         char *s_qf_names[EXT4_MAXQUOTAS];
4589 #endif
4590 };
4591
4592 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4593 {
4594         struct ext4_super_block *es;
4595         struct ext4_sb_info *sbi = EXT4_SB(sb);
4596         unsigned long old_sb_flags;
4597         struct ext4_mount_options old_opts;
4598         int enable_quota = 0;
4599         ext4_group_t g;
4600         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4601         int err = 0;
4602 #ifdef CONFIG_QUOTA
4603         int i, j;
4604 #endif
4605         char *orig_data = kstrdup(data, GFP_KERNEL);
4606
4607         /* Store the original options */
4608         old_sb_flags = sb->s_flags;
4609         old_opts.s_mount_opt = sbi->s_mount_opt;
4610         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4611         old_opts.s_resuid = sbi->s_resuid;
4612         old_opts.s_resgid = sbi->s_resgid;
4613         old_opts.s_commit_interval = sbi->s_commit_interval;
4614         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4615         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4616 #ifdef CONFIG_QUOTA
4617         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4618         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4619                 if (sbi->s_qf_names[i]) {
4620                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4621                                                          GFP_KERNEL);
4622                         if (!old_opts.s_qf_names[i]) {
4623                                 for (j = 0; j < i; j++)
4624                                         kfree(old_opts.s_qf_names[j]);
4625                                 kfree(orig_data);
4626                                 return -ENOMEM;
4627                         }
4628                 } else
4629                         old_opts.s_qf_names[i] = NULL;
4630 #endif
4631         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4632                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4633
4634         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4635                 err = -EINVAL;
4636                 goto restore_opts;
4637         }
4638
4639         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
4640             test_opt(sb, JOURNAL_CHECKSUM)) {
4641                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
4642                          "during remount not supported; ignoring");
4643                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
4644         }
4645
4646         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4647                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4648                         ext4_msg(sb, KERN_ERR, "can't mount with "
4649                                  "both data=journal and delalloc");
4650                         err = -EINVAL;
4651                         goto restore_opts;
4652                 }
4653                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4654                         ext4_msg(sb, KERN_ERR, "can't mount with "
4655                                  "both data=journal and dioread_nolock");
4656                         err = -EINVAL;
4657                         goto restore_opts;
4658                 }
4659                 if (test_opt(sb, DAX)) {
4660                         ext4_msg(sb, KERN_ERR, "can't mount with "
4661                                  "both data=journal and dax");
4662                         err = -EINVAL;
4663                         goto restore_opts;
4664                 }
4665         }
4666
4667         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
4668                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
4669                         "dax flag with busy inodes while remounting");
4670                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
4671         }
4672
4673         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4674                 ext4_abort(sb, "Abort forced by user");
4675
4676         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4677                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4678
4679         es = sbi->s_es;
4680
4681         if (sbi->s_journal) {
4682                 ext4_init_journal_params(sb, sbi->s_journal);
4683                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4684         }
4685
4686         if (*flags & MS_LAZYTIME)
4687                 sb->s_flags |= MS_LAZYTIME;
4688
4689         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4690                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4691                         err = -EROFS;
4692                         goto restore_opts;
4693                 }
4694
4695                 if (*flags & MS_RDONLY) {
4696                         err = sync_filesystem(sb);
4697                         if (err < 0)
4698                                 goto restore_opts;
4699                         err = dquot_suspend(sb, -1);
4700                         if (err < 0)
4701                                 goto restore_opts;
4702
4703                         /*
4704                          * First of all, the unconditional stuff we have to do
4705                          * to disable replay of the journal when we next remount
4706                          */
4707                         sb->s_flags |= MS_RDONLY;
4708
4709                         /*
4710                          * OK, test if we are remounting a valid rw partition
4711                          * readonly, and if so set the rdonly flag and then
4712                          * mark the partition as valid again.
4713                          */
4714                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4715                             (sbi->s_mount_state & EXT4_VALID_FS))
4716                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4717
4718                         if (sbi->s_journal)
4719                                 ext4_mark_recovery_complete(sb, es);
4720                 } else {
4721                         /* Make sure we can mount this feature set readwrite */
4722                         if (ext4_has_feature_readonly(sb) ||
4723                             !ext4_feature_set_ok(sb, 0)) {
4724                                 err = -EROFS;
4725                                 goto restore_opts;
4726                         }
4727                         /*
4728                          * Make sure the group descriptor checksums
4729                          * are sane.  If they aren't, refuse to remount r/w.
4730                          */
4731                         for (g = 0; g < sbi->s_groups_count; g++) {
4732                                 struct ext4_group_desc *gdp =
4733                                         ext4_get_group_desc(sb, g, NULL);
4734
4735                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4736                                         ext4_msg(sb, KERN_ERR,
4737                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4738                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
4739                                                le16_to_cpu(gdp->bg_checksum));
4740                                         err = -EFSBADCRC;
4741                                         goto restore_opts;
4742                                 }
4743                         }
4744
4745                         /*
4746                          * If we have an unprocessed orphan list hanging
4747                          * around from a previously readonly bdev mount,
4748                          * require a full umount/remount for now.
4749                          */
4750                         if (es->s_last_orphan) {
4751                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4752                                        "remount RDWR because of unprocessed "
4753                                        "orphan inode list.  Please "
4754                                        "umount/remount instead");
4755                                 err = -EINVAL;
4756                                 goto restore_opts;
4757                         }
4758
4759                         /*
4760                          * Mounting a RDONLY partition read-write, so reread
4761                          * and store the current valid flag.  (It may have
4762                          * been changed by e2fsck since we originally mounted
4763                          * the partition.)
4764                          */
4765                         if (sbi->s_journal)
4766                                 ext4_clear_journal_err(sb, es);
4767                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4768                         if (!ext4_setup_super(sb, es, 0))
4769                                 sb->s_flags &= ~MS_RDONLY;
4770                         if (ext4_has_feature_mmp(sb))
4771                                 if (ext4_multi_mount_protect(sb,
4772                                                 le64_to_cpu(es->s_mmp_block))) {
4773                                         err = -EROFS;
4774                                         goto restore_opts;
4775                                 }
4776                         enable_quota = 1;
4777                 }
4778         }
4779
4780         /*
4781          * Reinitialize lazy itable initialization thread based on
4782          * current settings
4783          */
4784         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4785                 ext4_unregister_li_request(sb);
4786         else {
4787                 ext4_group_t first_not_zeroed;
4788                 first_not_zeroed = ext4_has_uninit_itable(sb);
4789                 ext4_register_li_request(sb, first_not_zeroed);
4790         }
4791
4792         ext4_setup_system_zone(sb);
4793         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4794                 ext4_commit_super(sb, 1);
4795
4796 #ifdef CONFIG_QUOTA
4797         /* Release old quota file names */
4798         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4799                 kfree(old_opts.s_qf_names[i]);
4800         if (enable_quota) {
4801                 if (sb_any_quota_suspended(sb))
4802                         dquot_resume(sb, -1);
4803                 else if (ext4_has_feature_quota(sb)) {
4804                         err = ext4_enable_quotas(sb);
4805                         if (err)
4806                                 goto restore_opts;
4807                 }
4808         }
4809 #endif
4810
4811         *flags = (*flags & ~MS_LAZYTIME) | (sb->s_flags & MS_LAZYTIME);
4812         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4813         kfree(orig_data);
4814         return 0;
4815
4816 restore_opts:
4817         sb->s_flags = old_sb_flags;
4818         sbi->s_mount_opt = old_opts.s_mount_opt;
4819         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4820         sbi->s_resuid = old_opts.s_resuid;
4821         sbi->s_resgid = old_opts.s_resgid;
4822         sbi->s_commit_interval = old_opts.s_commit_interval;
4823         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4824         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4825 #ifdef CONFIG_QUOTA
4826         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4827         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
4828                 kfree(sbi->s_qf_names[i]);
4829                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4830         }
4831 #endif
4832         kfree(orig_data);
4833         return err;
4834 }
4835
4836 #ifdef CONFIG_QUOTA
4837 static int ext4_statfs_project(struct super_block *sb,
4838                                kprojid_t projid, struct kstatfs *buf)
4839 {
4840         struct kqid qid;
4841         struct dquot *dquot;
4842         u64 limit;
4843         u64 curblock;
4844
4845         qid = make_kqid_projid(projid);
4846         dquot = dqget(sb, qid);
4847         if (IS_ERR(dquot))
4848                 return PTR_ERR(dquot);
4849         spin_lock(&dq_data_lock);
4850
4851         limit = (dquot->dq_dqb.dqb_bsoftlimit ?
4852                  dquot->dq_dqb.dqb_bsoftlimit :
4853                  dquot->dq_dqb.dqb_bhardlimit) >> sb->s_blocksize_bits;
4854         if (limit && buf->f_blocks > limit) {
4855                 curblock = dquot->dq_dqb.dqb_curspace >> sb->s_blocksize_bits;
4856                 buf->f_blocks = limit;
4857                 buf->f_bfree = buf->f_bavail =
4858                         (buf->f_blocks > curblock) ?
4859                          (buf->f_blocks - curblock) : 0;
4860         }
4861
4862         limit = dquot->dq_dqb.dqb_isoftlimit ?
4863                 dquot->dq_dqb.dqb_isoftlimit :
4864                 dquot->dq_dqb.dqb_ihardlimit;
4865         if (limit && buf->f_files > limit) {
4866                 buf->f_files = limit;
4867                 buf->f_ffree =
4868                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
4869                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
4870         }
4871
4872         spin_unlock(&dq_data_lock);
4873         dqput(dquot);
4874         return 0;
4875 }
4876 #endif
4877
4878 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4879 {
4880         struct super_block *sb = dentry->d_sb;
4881         struct ext4_sb_info *sbi = EXT4_SB(sb);
4882         struct ext4_super_block *es = sbi->s_es;
4883         ext4_fsblk_t overhead = 0, resv_blocks;
4884         u64 fsid;
4885         s64 bfree;
4886         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
4887
4888         if (!test_opt(sb, MINIX_DF))
4889                 overhead = sbi->s_overhead;
4890
4891         buf->f_type = EXT4_SUPER_MAGIC;
4892         buf->f_bsize = sb->s_blocksize;
4893         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
4894         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4895                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4896         /* prevent underflow in case that few free space is available */
4897         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4898         buf->f_bavail = buf->f_bfree -
4899                         (ext4_r_blocks_count(es) + resv_blocks);
4900         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
4901                 buf->f_bavail = 0;
4902         buf->f_files = le32_to_cpu(es->s_inodes_count);
4903         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4904         buf->f_namelen = EXT4_NAME_LEN;
4905         fsid = le64_to_cpup((void *)es->s_uuid) ^
4906                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4907         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4908         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4909
4910 #ifdef CONFIG_QUOTA
4911         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
4912             sb_has_quota_limits_enabled(sb, PRJQUOTA))
4913                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
4914 #endif
4915         return 0;
4916 }
4917
4918 /* Helper function for writing quotas on sync - we need to start transaction
4919  * before quota file is locked for write. Otherwise the are possible deadlocks:
4920  * Process 1                         Process 2
4921  * ext4_create()                     quota_sync()
4922  *   jbd2_journal_start()                  write_dquot()
4923  *   dquot_initialize()                         down(dqio_mutex)
4924  *     down(dqio_mutex)                    jbd2_journal_start()
4925  *
4926  */
4927
4928 #ifdef CONFIG_QUOTA
4929
4930 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4931 {
4932         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
4933 }
4934
4935 static int ext4_write_dquot(struct dquot *dquot)
4936 {
4937         int ret, err;
4938         handle_t *handle;
4939         struct inode *inode;
4940
4941         inode = dquot_to_inode(dquot);
4942         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4943                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4944         if (IS_ERR(handle))
4945                 return PTR_ERR(handle);
4946         ret = dquot_commit(dquot);
4947         err = ext4_journal_stop(handle);
4948         if (!ret)
4949                 ret = err;
4950         return ret;
4951 }
4952
4953 static int ext4_acquire_dquot(struct dquot *dquot)
4954 {
4955         int ret, err;
4956         handle_t *handle;
4957
4958         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4959                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4960         if (IS_ERR(handle))
4961                 return PTR_ERR(handle);
4962         ret = dquot_acquire(dquot);
4963         err = ext4_journal_stop(handle);
4964         if (!ret)
4965                 ret = err;
4966         return ret;
4967 }
4968
4969 static int ext4_release_dquot(struct dquot *dquot)
4970 {
4971         int ret, err;
4972         handle_t *handle;
4973
4974         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4975                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4976         if (IS_ERR(handle)) {
4977                 /* Release dquot anyway to avoid endless cycle in dqput() */
4978                 dquot_release(dquot);
4979                 return PTR_ERR(handle);
4980         }
4981         ret = dquot_release(dquot);
4982         err = ext4_journal_stop(handle);
4983         if (!ret)
4984                 ret = err;
4985         return ret;
4986 }
4987
4988 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4989 {
4990         struct super_block *sb = dquot->dq_sb;
4991         struct ext4_sb_info *sbi = EXT4_SB(sb);
4992
4993         /* Are we journaling quotas? */
4994         if (ext4_has_feature_quota(sb) ||
4995             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
4996                 dquot_mark_dquot_dirty(dquot);
4997                 return ext4_write_dquot(dquot);
4998         } else {
4999                 return dquot_mark_dquot_dirty(dquot);
5000         }
5001 }
5002
5003 static int ext4_write_info(struct super_block *sb, int type)
5004 {
5005         int ret, err;
5006         handle_t *handle;
5007
5008         /* Data block + inode block */
5009         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5010         if (IS_ERR(handle))
5011                 return PTR_ERR(handle);
5012         ret = dquot_commit_info(sb, type);
5013         err = ext4_journal_stop(handle);
5014         if (!ret)
5015                 ret = err;
5016         return ret;
5017 }
5018
5019 /*
5020  * Turn on quotas during mount time - we need to find
5021  * the quota file and such...
5022  */
5023 static int ext4_quota_on_mount(struct super_block *sb, int type)
5024 {
5025         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5026                                         EXT4_SB(sb)->s_jquota_fmt, type);
5027 }
5028
5029 /*
5030  * Standard function to be called on quota_on
5031  */
5032 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5033                          struct path *path)
5034 {
5035         int err;
5036
5037         if (!test_opt(sb, QUOTA))
5038                 return -EINVAL;
5039
5040         /* Quotafile not on the same filesystem? */
5041         if (path->dentry->d_sb != sb)
5042                 return -EXDEV;
5043         /* Journaling quota? */
5044         if (EXT4_SB(sb)->s_qf_names[type]) {
5045                 /* Quotafile not in fs root? */
5046                 if (path->dentry->d_parent != sb->s_root)
5047                         ext4_msg(sb, KERN_WARNING,
5048                                 "Quota file not on filesystem root. "
5049                                 "Journaled quota will not work");
5050         }
5051
5052         /*
5053          * When we journal data on quota file, we have to flush journal to see
5054          * all updates to the file when we bypass pagecache...
5055          */
5056         if (EXT4_SB(sb)->s_journal &&
5057             ext4_should_journal_data(d_inode(path->dentry))) {
5058                 /*
5059                  * We don't need to lock updates but journal_flush() could
5060                  * otherwise be livelocked...
5061                  */
5062                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5063                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5064                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5065                 if (err)
5066                         return err;
5067         }
5068
5069         return dquot_quota_on(sb, type, format_id, path);
5070 }
5071
5072 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5073                              unsigned int flags)
5074 {
5075         int err;
5076         struct inode *qf_inode;
5077         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5078                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5079                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5080                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5081         };
5082
5083         BUG_ON(!ext4_has_feature_quota(sb));
5084
5085         if (!qf_inums[type])
5086                 return -EPERM;
5087
5088         qf_inode = ext4_iget(sb, qf_inums[type]);
5089         if (IS_ERR(qf_inode)) {
5090                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5091                 return PTR_ERR(qf_inode);
5092         }
5093
5094         /* Don't account quota for quota files to avoid recursion */
5095         qf_inode->i_flags |= S_NOQUOTA;
5096         err = dquot_enable(qf_inode, type, format_id, flags);
5097         iput(qf_inode);
5098
5099         return err;
5100 }
5101
5102 /* Enable usage tracking for all quota types. */
5103 static int ext4_enable_quotas(struct super_block *sb)
5104 {
5105         int type, err = 0;
5106         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5107                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5108                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5109                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5110         };
5111
5112         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5113         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5114                 if (qf_inums[type]) {
5115                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5116                                                 DQUOT_USAGE_ENABLED);
5117                         if (err) {
5118                                 ext4_warning(sb,
5119                                         "Failed to enable quota tracking "
5120                                         "(type=%d, err=%d). Please run "
5121                                         "e2fsck to fix.", type, err);
5122                                 return err;
5123                         }
5124                 }
5125         }
5126         return 0;
5127 }
5128
5129 static int ext4_quota_off(struct super_block *sb, int type)
5130 {
5131         struct inode *inode = sb_dqopt(sb)->files[type];
5132         handle_t *handle;
5133
5134         /* Force all delayed allocation blocks to be allocated.
5135          * Caller already holds s_umount sem */
5136         if (test_opt(sb, DELALLOC))
5137                 sync_filesystem(sb);
5138
5139         if (!inode)
5140                 goto out;
5141
5142         /* Update modification times of quota files when userspace can
5143          * start looking at them */
5144         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5145         if (IS_ERR(handle))
5146                 goto out;
5147         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5148         ext4_mark_inode_dirty(handle, inode);
5149         ext4_journal_stop(handle);
5150
5151 out:
5152         return dquot_quota_off(sb, type);
5153 }
5154
5155 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5156  * acquiring the locks... As quota files are never truncated and quota code
5157  * itself serializes the operations (and no one else should touch the files)
5158  * we don't have to be afraid of races */
5159 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5160                                size_t len, loff_t off)
5161 {
5162         struct inode *inode = sb_dqopt(sb)->files[type];
5163         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5164         int offset = off & (sb->s_blocksize - 1);
5165         int tocopy;
5166         size_t toread;
5167         struct buffer_head *bh;
5168         loff_t i_size = i_size_read(inode);
5169
5170         if (off > i_size)
5171                 return 0;
5172         if (off+len > i_size)
5173                 len = i_size-off;
5174         toread = len;
5175         while (toread > 0) {
5176                 tocopy = sb->s_blocksize - offset < toread ?
5177                                 sb->s_blocksize - offset : toread;
5178                 bh = ext4_bread(NULL, inode, blk, 0);
5179                 if (IS_ERR(bh))
5180                         return PTR_ERR(bh);
5181                 if (!bh)        /* A hole? */
5182                         memset(data, 0, tocopy);
5183                 else
5184                         memcpy(data, bh->b_data+offset, tocopy);
5185                 brelse(bh);
5186                 offset = 0;
5187                 toread -= tocopy;
5188                 data += tocopy;
5189                 blk++;
5190         }
5191         return len;
5192 }
5193
5194 /* Write to quotafile (we know the transaction is already started and has
5195  * enough credits) */
5196 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5197                                 const char *data, size_t len, loff_t off)
5198 {
5199         struct inode *inode = sb_dqopt(sb)->files[type];
5200         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5201         int err, offset = off & (sb->s_blocksize - 1);
5202         int retries = 0;
5203         struct buffer_head *bh;
5204         handle_t *handle = journal_current_handle();
5205
5206         if (EXT4_SB(sb)->s_journal && !handle) {
5207                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5208                         " cancelled because transaction is not started",
5209                         (unsigned long long)off, (unsigned long long)len);
5210                 return -EIO;
5211         }
5212         /*
5213          * Since we account only one data block in transaction credits,
5214          * then it is impossible to cross a block boundary.
5215          */
5216         if (sb->s_blocksize - offset < len) {
5217                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5218                         " cancelled because not block aligned",
5219                         (unsigned long long)off, (unsigned long long)len);
5220                 return -EIO;
5221         }
5222
5223         do {
5224                 bh = ext4_bread(handle, inode, blk,
5225                                 EXT4_GET_BLOCKS_CREATE |
5226                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5227         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5228                  ext4_should_retry_alloc(inode->i_sb, &retries));
5229         if (IS_ERR(bh))
5230                 return PTR_ERR(bh);
5231         if (!bh)
5232                 goto out;
5233         BUFFER_TRACE(bh, "get write access");
5234         err = ext4_journal_get_write_access(handle, bh);
5235         if (err) {
5236                 brelse(bh);
5237                 return err;
5238         }
5239         lock_buffer(bh);
5240         memcpy(bh->b_data+offset, data, len);
5241         flush_dcache_page(bh->b_page);
5242         unlock_buffer(bh);
5243         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5244         brelse(bh);
5245 out:
5246         if (inode->i_size < off + len) {
5247                 i_size_write(inode, off + len);
5248                 EXT4_I(inode)->i_disksize = inode->i_size;
5249                 ext4_mark_inode_dirty(handle, inode);
5250         }
5251         return len;
5252 }
5253
5254 #endif
5255
5256 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5257                        const char *dev_name, void *data)
5258 {
5259         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5260 }
5261
5262 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5263 static inline void register_as_ext2(void)
5264 {
5265         int err = register_filesystem(&ext2_fs_type);
5266         if (err)
5267                 printk(KERN_WARNING
5268                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5269 }
5270
5271 static inline void unregister_as_ext2(void)
5272 {
5273         unregister_filesystem(&ext2_fs_type);
5274 }
5275
5276 static inline int ext2_feature_set_ok(struct super_block *sb)
5277 {
5278         if (ext4_has_unknown_ext2_incompat_features(sb))
5279                 return 0;
5280         if (sb->s_flags & MS_RDONLY)
5281                 return 1;
5282         if (ext4_has_unknown_ext2_ro_compat_features(sb))
5283                 return 0;
5284         return 1;
5285 }
5286 #else
5287 static inline void register_as_ext2(void) { }
5288 static inline void unregister_as_ext2(void) { }
5289 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5290 #endif
5291
5292 static inline void register_as_ext3(void)
5293 {
5294         int err = register_filesystem(&ext3_fs_type);
5295         if (err)
5296                 printk(KERN_WARNING
5297                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5298 }
5299
5300 static inline void unregister_as_ext3(void)
5301 {
5302         unregister_filesystem(&ext3_fs_type);
5303 }
5304
5305 static inline int ext3_feature_set_ok(struct super_block *sb)
5306 {
5307         if (ext4_has_unknown_ext3_incompat_features(sb))
5308                 return 0;
5309         if (!ext4_has_feature_journal(sb))
5310                 return 0;
5311         if (sb->s_flags & MS_RDONLY)
5312                 return 1;
5313         if (ext4_has_unknown_ext3_ro_compat_features(sb))
5314                 return 0;
5315         return 1;
5316 }
5317
5318 static struct file_system_type ext4_fs_type = {
5319         .owner          = THIS_MODULE,
5320         .name           = "ext4",
5321         .mount          = ext4_mount,
5322         .kill_sb        = kill_block_super,
5323         .fs_flags       = FS_REQUIRES_DEV,
5324 };
5325 MODULE_ALIAS_FS("ext4");
5326
5327 /* Shared across all ext4 file systems */
5328 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5329
5330 static int __init ext4_init_fs(void)
5331 {
5332         int i, err;
5333
5334         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
5335         ext4_li_info = NULL;
5336         mutex_init(&ext4_li_mtx);
5337
5338         /* Build-time check for flags consistency */
5339         ext4_check_flag_values();
5340
5341         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
5342                 init_waitqueue_head(&ext4__ioend_wq[i]);
5343
5344         err = ext4_init_es();
5345         if (err)
5346                 return err;
5347
5348         err = ext4_init_pageio();
5349         if (err)
5350                 goto out5;
5351
5352         err = ext4_init_system_zone();
5353         if (err)
5354                 goto out4;
5355
5356         err = ext4_init_sysfs();
5357         if (err)
5358                 goto out3;
5359
5360         err = ext4_init_mballoc();
5361         if (err)
5362                 goto out2;
5363         err = init_inodecache();
5364         if (err)
5365                 goto out1;
5366         register_as_ext3();
5367         register_as_ext2();
5368         err = register_filesystem(&ext4_fs_type);
5369         if (err)
5370                 goto out;
5371
5372         return 0;
5373 out:
5374         unregister_as_ext2();
5375         unregister_as_ext3();
5376         destroy_inodecache();
5377 out1:
5378         ext4_exit_mballoc();
5379 out2:
5380         ext4_exit_sysfs();
5381 out3:
5382         ext4_exit_system_zone();
5383 out4:
5384         ext4_exit_pageio();
5385 out5:
5386         ext4_exit_es();
5387
5388         return err;
5389 }
5390
5391 static void __exit ext4_exit_fs(void)
5392 {
5393         ext4_exit_crypto();
5394         ext4_destroy_lazyinit_thread();
5395         unregister_as_ext2();
5396         unregister_as_ext3();
5397         unregister_filesystem(&ext4_fs_type);
5398         destroy_inodecache();
5399         ext4_exit_mballoc();
5400         ext4_exit_sysfs();
5401         ext4_exit_system_zone();
5402         ext4_exit_pageio();
5403         ext4_exit_es();
5404 }
5405
5406 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5407 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5408 MODULE_LICENSE("GPL");
5409 module_init(ext4_init_fs)
5410 module_exit(ext4_exit_fs)