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 int ext4_mballoc_ready;
  59 static struct ratelimit_state ext4_mount_msg_ratelimit;
  60
  61 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  62                              unsigned long journal_devnum);
  63 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  64 static int ext4_commit_super(struct super_block *sb, int sync);
  65 static void ext4_mark_recovery_complete(struct super_block *sb,
  66                                         struct ext4_super_block *es);
  67 static void ext4_clear_journal_err(struct super_block *sb,
  68                                    struct ext4_super_block *es);
  69 static int ext4_sync_fs(struct super_block *sb, int wait);
  70 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  71 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  72 static int ext4_unfreeze(struct super_block *sb);
  73 static int ext4_freeze(struct super_block *sb);
  74 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  75                        const char *dev_name, void *data);
  76 static inline int ext2_feature_set_ok(struct super_block *sb);
  77 static inline int ext3_feature_set_ok(struct super_block *sb);
  78 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  79 static void ext4_destroy_lazyinit_thread(void);
  80 static void ext4_unregister_li_request(struct super_block *sb);
  81 static void ext4_clear_request_list(void);
  82
  83 /*
  84  * Lock ordering
  85  *
  86  * Note the difference between i_mmap_sem (EXT4_I(inode)->i_mmap_sem) and
  87  * i_mmap_rwsem (inode->i_mmap_rwsem)!
  88  *
  89  * page fault path:
  90  * mmap_sem -> sb_start_pagefault -> i_mmap_sem (r) -> transaction start ->
  91  *   page lock -> i_data_sem (rw)
  92  *
  93  * buffered write path:
  94  * sb_start_write -> i_mutex -> mmap_sem
  95  * sb_start_write -> i_mutex -> transaction start -> page lock ->
  96  *   i_data_sem (rw)
  97  *
  98  * truncate:
  99  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (w) -> i_mmap_sem (w) ->
 100  *   i_mmap_rwsem (w) -> page lock
 101  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (w) -> i_mmap_sem (w) ->
 102  *   transaction start -> i_data_sem (rw)
 103  *
 104  * direct IO:
 105  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (r) -> mmap_sem
 106  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (r) ->
 107  *   transaction start -> i_data_sem (rw)
 108  *
 109  * writepages:
 110  * transaction start -> page lock(s) -> i_data_sem (rw)
 111  */
 112
 113 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
 114 static struct file_system_type ext2_fs_type = {
 115         .owner          = THIS_MODULE,
 116         .name           = "ext2",
 117         .mount          = ext4_mount,
 118         .kill_sb        = kill_block_super,
 119         .fs_flags       = FS_REQUIRES_DEV,
 120 };
 121 MODULE_ALIAS_FS("ext2");
 122 MODULE_ALIAS("ext2");
 123 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
 124 #else
 125 #define IS_EXT2_SB(sb) (0)
 126 #endif
 127
 128
 129 static struct file_system_type ext3_fs_type = {
 130         .owner          = THIS_MODULE,
 131         .name           = "ext3",
 132         .mount          = ext4_mount,
 133         .kill_sb        = kill_block_super,
 134         .fs_flags       = FS_REQUIRES_DEV,
 135 };
 136 MODULE_ALIAS_FS("ext3");
 137 MODULE_ALIAS("ext3");
 138 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 139
 140 static int ext4_verify_csum_type(struct super_block *sb,
 141                                  struct ext4_super_block *es)
 142 {
 143         if (!ext4_has_feature_metadata_csum(sb))
 144                 return 1;
 145
 146         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 147 }
 148
 149 static __le32 ext4_superblock_csum(struct super_block *sb,
 150                                    struct ext4_super_block *es)
 151 {
 152         struct ext4_sb_info *sbi = EXT4_SB(sb);
 153         int offset = offsetof(struct ext4_super_block, s_checksum);
 154         __u32 csum;
 155
 156         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 157
 158         return cpu_to_le32(csum);
 159 }
 160
 161 static int ext4_superblock_csum_verify(struct super_block *sb,
 162                                        struct ext4_super_block *es)
 163 {
 164         if (!ext4_has_metadata_csum(sb))
 165                 return 1;
 166
 167         return es->s_checksum == ext4_superblock_csum(sb, es);
 168 }
 169
 170 void ext4_superblock_csum_set(struct super_block *sb)
 171 {
 172         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 173
 174         if (!ext4_has_metadata_csum(sb))
 175                 return;
 176
 177         es->s_checksum = ext4_superblock_csum(sb, es);
 178 }
 179
 180 void *ext4_kvmalloc(size_t size, gfp_t flags)
 181 {
 182         void *ret;
 183
 184         ret = kmalloc(size, flags | __GFP_NOWARN);
 185         if (!ret)
 186                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 187         return ret;
 188 }
 189
 190 void *ext4_kvzalloc(size_t size, gfp_t flags)
 191 {
 192         void *ret;
 193
 194         ret = kzalloc(size, flags | __GFP_NOWARN);
 195         if (!ret)
 196                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 197         return ret;
 198 }
 199
 200 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 201                                struct ext4_group_desc *bg)
 202 {
 203         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 204                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 205                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 206 }
 207
 208 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 209                                struct ext4_group_desc *bg)
 210 {
 211         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 212                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 213                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 214 }
 215
 216 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 217                               struct ext4_group_desc *bg)
 218 {
 219         return le32_to_cpu(bg->bg_inode_table_lo) |
 220                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 221                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 222 }
 223
 224 __u32 ext4_free_group_clusters(struct super_block *sb,
 225                                struct ext4_group_desc *bg)
 226 {
 227         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 228                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 229                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 230 }
 231
 232 __u32 ext4_free_inodes_count(struct super_block *sb,
 233                               struct ext4_group_desc *bg)
 234 {
 235         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 236                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 237                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 238 }
 239
 240 __u32 ext4_used_dirs_count(struct super_block *sb,
 241                               struct ext4_group_desc *bg)
 242 {
 243         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 244                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 245                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 246 }
 247
 248 __u32 ext4_itable_unused_count(struct super_block *sb,
 249                               struct ext4_group_desc *bg)
 250 {
 251         return le16_to_cpu(bg->bg_itable_unused_lo) |
 252                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 253                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 254 }
 255
 256 void ext4_block_bitmap_set(struct super_block *sb,
 257                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 258 {
 259         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 260         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 261                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 262 }
 263
 264 void ext4_inode_bitmap_set(struct super_block *sb,
 265                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 266 {
 267         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 268         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 269                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 270 }
 271
 272 void ext4_inode_table_set(struct super_block *sb,
 273                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 274 {
 275         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 276         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 277                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 278 }
 279
 280 void ext4_free_group_clusters_set(struct super_block *sb,
 281                                   struct ext4_group_desc *bg, __u32 count)
 282 {
 283         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 284         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 285                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 286 }
 287
 288 void ext4_free_inodes_set(struct super_block *sb,
 289                           struct ext4_group_desc *bg, __u32 count)
 290 {
 291         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 292         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 293                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 294 }
 295
 296 void ext4_used_dirs_set(struct super_block *sb,
 297                           struct ext4_group_desc *bg, __u32 count)
 298 {
 299         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 300         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 301                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 302 }
 303
 304 void ext4_itable_unused_set(struct super_block *sb,
 305                           struct ext4_group_desc *bg, __u32 count)
 306 {
 307         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 308         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 309                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 310 }
 311
 312
 313 static void __save_error_info(struct super_block *sb, const char *func,
 314                             unsigned int line)
 315 {
 316         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 317
 318         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 319         if (bdev_read_only(sb->s_bdev))
 320                 return;
 321         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 322         es->s_last_error_time = cpu_to_le32(get_seconds());
 323         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 324         es->s_last_error_line = cpu_to_le32(line);
 325         if (!es->s_first_error_time) {
 326                 es->s_first_error_time = es->s_last_error_time;
 327                 strncpy(es->s_first_error_func, func,
 328                         sizeof(es->s_first_error_func));
 329                 es->s_first_error_line = cpu_to_le32(line);
 330                 es->s_first_error_ino = es->s_last_error_ino;
 331                 es->s_first_error_block = es->s_last_error_block;
 332         }
 333         /*
 334          * Start the daily error reporting function if it hasn't been
 335          * started already
 336          */
 337         if (!es->s_error_count)
 338                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 339         le32_add_cpu(&es->s_error_count, 1);
 340 }
 341
 342 static void save_error_info(struct super_block *sb, const char *func,
 343                             unsigned int line)
 344 {
 345         __save_error_info(sb, func, line);
 346         ext4_commit_super(sb, 1);
 347 }
 348
 349 /*
 350  * The del_gendisk() function uninitializes the disk-specific data
 351  * structures, including the bdi structure, without telling anyone
 352  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 353  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 354  * This is a kludge to prevent these oops until we can put in a proper
 355  * hook in del_gendisk() to inform the VFS and file system layers.
 356  */
 357 static int block_device_ejected(struct super_block *sb)
 358 {
 359         struct inode *bd_inode = sb->s_bdev->bd_inode;
 360         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 361
 362         return bdi->dev == NULL;
 363 }
 364
 365 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 366 {
 367         struct super_block              *sb = journal->j_private;
 368         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 369         int                             error = is_journal_aborted(journal);
 370         struct ext4_journal_cb_entry    *jce;
 371
 372         BUG_ON(txn->t_state == T_FINISHED);
 373         spin_lock(&sbi->s_md_lock);
 374         while (!list_empty(&txn->t_private_list)) {
 375                 jce = list_entry(txn->t_private_list.next,
 376                                  struct ext4_journal_cb_entry, jce_list);
 377                 list_del_init(&jce->jce_list);
 378                 spin_unlock(&sbi->s_md_lock);
 379                 jce->jce_func(sb, jce, error);
 380                 spin_lock(&sbi->s_md_lock);
 381         }
 382         spin_unlock(&sbi->s_md_lock);
 383 }
 384
 385 /* Deal with the reporting of failure conditions on a filesystem such as
 386  * inconsistencies detected or read IO failures.
 387  *
 388  * On ext2, we can store the error state of the filesystem in the
 389  * superblock.  That is not possible on ext4, because we may have other
 390  * write ordering constraints on the superblock which prevent us from
 391  * writing it out straight away; and given that the journal is about to
 392  * be aborted, we can't rely on the current, or future, transactions to
 393  * write out the superblock safely.
 394  *
 395  * We'll just use the jbd2_journal_abort() error code to record an error in
 396  * the journal instead.  On recovery, the journal will complain about
 397  * that error until we've noted it down and cleared it.
 398  */
 399
 400 static void ext4_handle_error(struct super_block *sb)
 401 {
 402         if (sb->s_flags & MS_RDONLY)
 403                 return;
 404
 405         if (!test_opt(sb, ERRORS_CONT)) {
 406                 journal_t *journal = EXT4_SB(sb)->s_journal;
 407
 408                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 409                 if (journal)
 410                         jbd2_journal_abort(journal, -EIO);
 411         }
 412         if (test_opt(sb, ERRORS_RO)) {
 413                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 414                 /*
 415                  * Make sure updated value of ->s_mount_flags will be visible
 416                  * before ->s_flags update
 417                  */
 418                 smp_wmb();
 419                 sb->s_flags |= MS_RDONLY;
 420         }
 421         if (test_opt(sb, ERRORS_PANIC)) {
 422                 if (EXT4_SB(sb)->s_journal &&
 423                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 424                         return;
 425                 panic("EXT4-fs (device %s): panic forced after error\n",
 426                         sb->s_id);
 427         }
 428 }
 429
 430 #define ext4_error_ratelimit(sb)                                        \
 431                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 432                              "EXT4-fs error")
 433
 434 void __ext4_error(struct super_block *sb, const char *function,
 435                   unsigned int line, const char *fmt, ...)
 436 {
 437         struct va_format vaf;
 438         va_list args;
 439
 440         if (ext4_error_ratelimit(sb)) {
 441                 va_start(args, fmt);
 442                 vaf.fmt = fmt;
 443                 vaf.va = &args;
 444                 printk(KERN_CRIT
 445                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 446                        sb->s_id, function, line, current->comm, &vaf);
 447                 va_end(args);
 448         }
 449         save_error_info(sb, function, line);
 450         ext4_handle_error(sb);
 451 }
 452
 453 void __ext4_error_inode(struct inode *inode, const char *function,
 454                         unsigned int line, ext4_fsblk_t block,
 455                         const char *fmt, ...)
 456 {
 457         va_list args;
 458         struct va_format vaf;
 459         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 460
 461         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 462         es->s_last_error_block = cpu_to_le64(block);
 463         if (ext4_error_ratelimit(inode->i_sb)) {
 464                 va_start(args, fmt);
 465                 vaf.fmt = fmt;
 466                 vaf.va = &args;
 467                 if (block)
 468                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 469                                "inode #%lu: block %llu: comm %s: %pV\n",
 470                                inode->i_sb->s_id, function, line, inode->i_ino,
 471                                block, current->comm, &vaf);
 472                 else
 473                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 474                                "inode #%lu: comm %s: %pV\n",
 475                                inode->i_sb->s_id, function, line, inode->i_ino,
 476                                current->comm, &vaf);
 477                 va_end(args);
 478         }
 479         save_error_info(inode->i_sb, function, line);
 480         ext4_handle_error(inode->i_sb);
 481 }
 482
 483 void __ext4_error_file(struct file *file, const char *function,
 484                        unsigned int line, ext4_fsblk_t block,
 485                        const char *fmt, ...)
 486 {
 487         va_list args;
 488         struct va_format vaf;
 489         struct ext4_super_block *es;
 490         struct inode *inode = file_inode(file);
 491         char pathname[80], *path;
 492
 493         es = EXT4_SB(inode->i_sb)->s_es;
 494         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 495         if (ext4_error_ratelimit(inode->i_sb)) {
 496                 path = file_path(file, pathname, sizeof(pathname));
 497                 if (IS_ERR(path))
 498                         path = "(unknown)";
 499                 va_start(args, fmt);
 500                 vaf.fmt = fmt;
 501                 vaf.va = &args;
 502                 if (block)
 503                         printk(KERN_CRIT
 504                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 505                                "block %llu: comm %s: path %s: %pV\n",
 506                                inode->i_sb->s_id, function, line, inode->i_ino,
 507                                block, current->comm, path, &vaf);
 508                 else
 509                         printk(KERN_CRIT
 510                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 511                                "comm %s: path %s: %pV\n",
 512                                inode->i_sb->s_id, function, line, inode->i_ino,
 513                                current->comm, path, &vaf);
 514                 va_end(args);
 515         }
 516         save_error_info(inode->i_sb, function, line);
 517         ext4_handle_error(inode->i_sb);
 518 }
 519
 520 const char *ext4_decode_error(struct super_block *sb, int errno,
 521                               char nbuf[16])
 522 {
 523         char *errstr = NULL;
 524
 525         switch (errno) {
 526         case -EFSCORRUPTED:
 527                 errstr = "Corrupt filesystem";
 528                 break;
 529         case -EFSBADCRC:
 530                 errstr = "Filesystem failed CRC";
 531                 break;
 532         case -EIO:
 533                 errstr = "IO failure";
 534                 break;
 535         case -ENOMEM:
 536                 errstr = "Out of memory";
 537                 break;
 538         case -EROFS:
 539                 if (!sb || (EXT4_SB(sb)->s_journal &&
 540                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 541                         errstr = "Journal has aborted";
 542                 else
 543                         errstr = "Readonly filesystem";
 544                 break;
 545         default:
 546                 /* If the caller passed in an extra buffer for unknown
 547                  * errors, textualise them now.  Else we just return
 548                  * NULL. */
 549                 if (nbuf) {
 550                         /* Check for truncated error codes... */
 551                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 552                                 errstr = nbuf;
 553                 }
 554                 break;
 555         }
 556
 557         return errstr;
 558 }
 559
 560 /* __ext4_std_error decodes expected errors from journaling functions
 561  * automatically and invokes the appropriate error response.  */
 562
 563 void __ext4_std_error(struct super_block *sb, const char *function,
 564                       unsigned int line, int errno)
 565 {
 566         char nbuf[16];
 567         const char *errstr;
 568
 569         /* Special case: if the error is EROFS, and we're not already
 570          * inside a transaction, then there's really no point in logging
 571          * an error. */
 572         if (errno == -EROFS && journal_current_handle() == NULL &&
 573             (sb->s_flags & MS_RDONLY))
 574                 return;
 575
 576         if (ext4_error_ratelimit(sb)) {
 577                 errstr = ext4_decode_error(sb, errno, nbuf);
 578                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 579                        sb->s_id, function, line, errstr);
 580         }
 581
 582         save_error_info(sb, function, line);
 583         ext4_handle_error(sb);
 584 }
 585
 586 /*
 587  * ext4_abort is a much stronger failure handler than ext4_error.  The
 588  * abort function may be used to deal with unrecoverable failures such
 589  * as journal IO errors or ENOMEM at a critical moment in log management.
 590  *
 591  * We unconditionally force the filesystem into an ABORT|READONLY state,
 592  * unless the error response on the fs has been set to panic in which
 593  * case we take the easy way out and panic immediately.
 594  */
 595
 596 void __ext4_abort(struct super_block *sb, const char *function,
 597                 unsigned int line, const char *fmt, ...)
 598 {
 599         va_list args;
 600
 601         save_error_info(sb, function, line);
 602         va_start(args, fmt);
 603         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 604                function, line);
 605         vprintk(fmt, args);
 606         printk("\n");
 607         va_end(args);
 608
 609         if ((sb->s_flags & MS_RDONLY) == 0) {
 610                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 611                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 612                 /*
 613                  * Make sure updated value of ->s_mount_flags will be visible
 614                  * before ->s_flags update
 615                  */
 616                 smp_wmb();
 617                 sb->s_flags |= MS_RDONLY;
 618                 if (EXT4_SB(sb)->s_journal)
 619                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 620                 save_error_info(sb, function, line);
 621         }
 622         if (test_opt(sb, ERRORS_PANIC)) {
 623                 if (EXT4_SB(sb)->s_journal &&
 624                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 625                         return;
 626                 panic("EXT4-fs panic from previous error\n");
 627         }
 628 }
 629
 630 void __ext4_msg(struct super_block *sb,
 631                 const char *prefix, const char *fmt, ...)
 632 {
 633         struct va_format vaf;
 634         va_list args;
 635
 636         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 637                 return;
 638
 639         va_start(args, fmt);
 640         vaf.fmt = fmt;
 641         vaf.va = &args;
 642         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 643         va_end(args);
 644 }
 645
 646 #define ext4_warning_ratelimit(sb)                                      \
 647                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
 648                              "EXT4-fs warning")
 649
 650 void __ext4_warning(struct super_block *sb, const char *function,
 651                     unsigned int line, const char *fmt, ...)
 652 {
 653         struct va_format vaf;
 654         va_list args;
 655
 656         if (!ext4_warning_ratelimit(sb))
 657                 return;
 658
 659         va_start(args, fmt);
 660         vaf.fmt = fmt;
 661         vaf.va = &args;
 662         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 663                sb->s_id, function, line, &vaf);
 664         va_end(args);
 665 }
 666
 667 void __ext4_warning_inode(const struct inode *inode, const char *function,
 668                           unsigned int line, const char *fmt, ...)
 669 {
 670         struct va_format vaf;
 671         va_list args;
 672
 673         if (!ext4_warning_ratelimit(inode->i_sb))
 674                 return;
 675
 676         va_start(args, fmt);
 677         vaf.fmt = fmt;
 678         vaf.va = &args;
 679         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 680                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 681                function, line, inode->i_ino, current->comm, &vaf);
 682         va_end(args);
 683 }
 684
 685 void __ext4_grp_locked_error(const char *function, unsigned int line,
 686                              struct super_block *sb, ext4_group_t grp,
 687                              unsigned long ino, ext4_fsblk_t block,
 688                              const char *fmt, ...)
 689 __releases(bitlock)
 690 __acquires(bitlock)
 691 {
 692         struct va_format vaf;
 693         va_list args;
 694         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 695
 696         es->s_last_error_ino = cpu_to_le32(ino);
 697         es->s_last_error_block = cpu_to_le64(block);
 698         __save_error_info(sb, function, line);
 699
 700         if (ext4_error_ratelimit(sb)) {
 701                 va_start(args, fmt);
 702                 vaf.fmt = fmt;
 703                 vaf.va = &args;
 704                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 705                        sb->s_id, function, line, grp);
 706                 if (ino)
 707                         printk(KERN_CONT "inode %lu: ", ino);
 708                 if (block)
 709                         printk(KERN_CONT "block %llu:",
 710                                (unsigned long long) block);
 711                 printk(KERN_CONT "%pV\n", &vaf);
 712                 va_end(args);
 713         }
 714
 715         if (test_opt(sb, ERRORS_CONT)) {
 716                 ext4_commit_super(sb, 0);
 717                 return;
 718         }
 719
 720         ext4_unlock_group(sb, grp);
 721         ext4_handle_error(sb);
 722         /*
 723          * We only get here in the ERRORS_RO case; relocking the group
 724          * may be dangerous, but nothing bad will happen since the
 725          * filesystem will have already been marked read/only and the
 726          * journal has been aborted.  We return 1 as a hint to callers
 727          * who might what to use the return value from
 728          * ext4_grp_locked_error() to distinguish between the
 729          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 730          * aggressively from the ext4 function in question, with a
 731          * more appropriate error code.
 732          */
 733         ext4_lock_group(sb, grp);
 734         return;
 735 }
 736
 737 void ext4_update_dynamic_rev(struct super_block *sb)
 738 {
 739         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 740
 741         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 742                 return;
 743
 744         ext4_warning(sb,
 745                      "updating to rev %d because of new feature flag, "
 746                      "running e2fsck is recommended",
 747                      EXT4_DYNAMIC_REV);
 748
 749         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 750         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 751         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 752         /* leave es->s_feature_*compat flags alone */
 753         /* es->s_uuid will be set by e2fsck if empty */
 754
 755         /*
 756          * The rest of the superblock fields should be zero, and if not it
 757          * means they are likely already in use, so leave them alone.  We
 758          * can leave it up to e2fsck to clean up any inconsistencies there.
 759          */
 760 }
 761
 762 /*
 763  * Open the external journal device
 764  */
 765 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 766 {
 767         struct block_device *bdev;
 768         char b[BDEVNAME_SIZE];
 769
 770         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 771         if (IS_ERR(bdev))
 772                 goto fail;
 773         return bdev;
 774
 775 fail:
 776         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 777                         __bdevname(dev, b), PTR_ERR(bdev));
 778         return NULL;
 779 }
 780
 781 /*
 782  * Release the journal device
 783  */
 784 static void ext4_blkdev_put(struct block_device *bdev)
 785 {
 786         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 787 }
 788
 789 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 790 {
 791         struct block_device *bdev;
 792         bdev = sbi->journal_bdev;
 793         if (bdev) {
 794                 ext4_blkdev_put(bdev);
 795                 sbi->journal_bdev = NULL;
 796         }
 797 }
 798
 799 static inline struct inode *orphan_list_entry(struct list_head *l)
 800 {
 801         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 802 }
 803
 804 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 805 {
 806         struct list_head *l;
 807
 808         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 809                  le32_to_cpu(sbi->s_es->s_last_orphan));
 810
 811         printk(KERN_ERR "sb_info orphan list:\n");
 812         list_for_each(l, &sbi->s_orphan) {
 813                 struct inode *inode = orphan_list_entry(l);
 814                 printk(KERN_ERR "  "
 815                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 816                        inode->i_sb->s_id, inode->i_ino, inode,
 817                        inode->i_mode, inode->i_nlink,
 818                        NEXT_ORPHAN(inode));
 819         }
 820 }
 821
 822 static void ext4_put_super(struct super_block *sb)
 823 {
 824         struct ext4_sb_info *sbi = EXT4_SB(sb);
 825         struct ext4_super_block *es = sbi->s_es;
 826         int i, err;
 827
 828         ext4_unregister_li_request(sb);
 829         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 830
 831         flush_workqueue(sbi->rsv_conversion_wq);
 832         destroy_workqueue(sbi->rsv_conversion_wq);
 833
 834         if (sbi->s_journal) {
 835                 err = jbd2_journal_destroy(sbi->s_journal);
 836                 sbi->s_journal = NULL;
 837                 if (err < 0)
 838                         ext4_abort(sb, "Couldn't clean up the journal");
 839         }
 840
 841         ext4_unregister_sysfs(sb);
 842         ext4_es_unregister_shrinker(sbi);
 843         del_timer_sync(&sbi->s_err_report);
 844         ext4_release_system_zone(sb);
 845         ext4_mb_release(sb);
 846         ext4_ext_release(sb);
 847         ext4_xattr_put_super(sb);
 848
 849         if (!(sb->s_flags & MS_RDONLY)) {
 850                 ext4_clear_feature_journal_needs_recovery(sb);
 851                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 852         }
 853         if (!(sb->s_flags & MS_RDONLY))
 854                 ext4_commit_super(sb, 1);
 855
 856         for (i = 0; i < sbi->s_gdb_count; i++)
 857                 brelse(sbi->s_group_desc[i]);
 858         kvfree(sbi->s_group_desc);
 859         kvfree(sbi->s_flex_groups);
 860         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 861         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 862         percpu_counter_destroy(&sbi->s_dirs_counter);
 863         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 864         brelse(sbi->s_sbh);
 865 #ifdef CONFIG_QUOTA
 866         for (i = 0; i < EXT4_MAXQUOTAS; i++)
 867                 kfree(sbi->s_qf_names[i]);
 868 #endif
 869
 870         /* Debugging code just in case the in-memory inode orphan list
 871          * isn't empty.  The on-disk one can be non-empty if we've
 872          * detected an error and taken the fs readonly, but the
 873          * in-memory list had better be clean by this point. */
 874         if (!list_empty(&sbi->s_orphan))
 875                 dump_orphan_list(sb, sbi);
 876         J_ASSERT(list_empty(&sbi->s_orphan));
 877
 878         sync_blockdev(sb->s_bdev);
 879         invalidate_bdev(sb->s_bdev);
 880         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 881                 /*
 882                  * Invalidate the journal device's buffers.  We don't want them
 883                  * floating about in memory - the physical journal device may
 884                  * hotswapped, and it breaks the `ro-after' testing code.
 885                  */
 886                 sync_blockdev(sbi->journal_bdev);
 887                 invalidate_bdev(sbi->journal_bdev);
 888                 ext4_blkdev_remove(sbi);
 889         }
 890         if (sbi->s_mb_cache) {
 891                 ext4_xattr_destroy_cache(sbi->s_mb_cache);
 892                 sbi->s_mb_cache = NULL;
 893         }
 894         if (sbi->s_mmp_tsk)
 895                 kthread_stop(sbi->s_mmp_tsk);
 896         sb->s_fs_info = NULL;
 897         /*
 898          * Now that we are completely done shutting down the
 899          * superblock, we need to actually destroy the kobject.
 900          */
 901         kobject_put(&sbi->s_kobj);
 902         wait_for_completion(&sbi->s_kobj_unregister);
 903         if (sbi->s_chksum_driver)
 904                 crypto_free_shash(sbi->s_chksum_driver);
 905         kfree(sbi->s_blockgroup_lock);
 906         kfree(sbi);
 907 }
 908
 909 static struct kmem_cache *ext4_inode_cachep;
 910
 911 /*
 912  * Called inside transaction, so use GFP_NOFS
 913  */
 914 static struct inode *ext4_alloc_inode(struct super_block *sb)
 915 {
 916         struct ext4_inode_info *ei;
 917
 918         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 919         if (!ei)
 920                 return NULL;
 921
 922         ei->vfs_inode.i_version = 1;
 923         spin_lock_init(&ei->i_raw_lock);
 924         INIT_LIST_HEAD(&ei->i_prealloc_list);
 925         spin_lock_init(&ei->i_prealloc_lock);
 926         ext4_es_init_tree(&ei->i_es_tree);
 927         rwlock_init(&ei->i_es_lock);
 928         INIT_LIST_HEAD(&ei->i_es_list);
 929         ei->i_es_all_nr = 0;
 930         ei->i_es_shk_nr = 0;
 931         ei->i_es_shrink_lblk = 0;
 932         ei->i_reserved_data_blocks = 0;
 933         ei->i_reserved_meta_blocks = 0;
 934         ei->i_allocated_meta_blocks = 0;
 935         ei->i_da_metadata_calc_len = 0;
 936         ei->i_da_metadata_calc_last_lblock = 0;
 937         spin_lock_init(&(ei->i_block_reservation_lock));
 938 #ifdef CONFIG_QUOTA
 939         ei->i_reserved_quota = 0;
 940         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
 941 #endif
 942         ei->jinode = NULL;
 943         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
 944         spin_lock_init(&ei->i_completed_io_lock);
 945         ei->i_sync_tid = 0;
 946         ei->i_datasync_tid = 0;
 947         atomic_set(&ei->i_ioend_count, 0);
 948         atomic_set(&ei->i_unwritten, 0);
 949         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
 950 #ifdef CONFIG_EXT4_FS_ENCRYPTION
 951         ei->i_crypt_info = NULL;
 952 #endif
 953         return &ei->vfs_inode;
 954 }
 955
 956 static int ext4_drop_inode(struct inode *inode)
 957 {
 958         int drop = generic_drop_inode(inode);
 959
 960         trace_ext4_drop_inode(inode, drop);
 961         return drop;
 962 }
 963
 964 static void ext4_i_callback(struct rcu_head *head)
 965 {
 966         struct inode *inode = container_of(head, struct inode, i_rcu);
 967         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 968 }
 969
 970 static void ext4_destroy_inode(struct inode *inode)
 971 {
 972         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 973                 ext4_msg(inode->i_sb, KERN_ERR,
 974                          "Inode %lu (%p): orphan list check failed!",
 975                          inode->i_ino, EXT4_I(inode));
 976                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 977                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 978                                 true);
 979                 dump_stack();
 980         }
 981         call_rcu(&inode->i_rcu, ext4_i_callback);
 982 }
 983
 984 static void init_once(void *foo)
 985 {
 986         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 987
 988         INIT_LIST_HEAD(&ei->i_orphan);
 989         init_rwsem(&ei->xattr_sem);
 990         init_rwsem(&ei->i_data_sem);
 991         init_rwsem(&ei->i_mmap_sem);
 992         inode_init_once(&ei->vfs_inode);
 993 }
 994
 995 static int __init init_inodecache(void)
 996 {
 997         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 998                                              sizeof(struct ext4_inode_info),
 999                                              0, (SLAB_RECLAIM_ACCOUNT|
1000                                                 SLAB_MEM_SPREAD),
1001                                              init_once);
1002         if (ext4_inode_cachep == NULL)
1003                 return -ENOMEM;
1004         return 0;
1005 }
1006
1007 static void destroy_inodecache(void)
1008 {
1009         /*
1010          * Make sure all delayed rcu free inodes are flushed before we
1011          * destroy cache.
1012          */
1013         rcu_barrier();
1014         kmem_cache_destroy(ext4_inode_cachep);
1015 }
1016
1017 void ext4_clear_inode(struct inode *inode)
1018 {
1019         invalidate_inode_buffers(inode);
1020         clear_inode(inode);
1021         dquot_drop(inode);
1022         ext4_discard_preallocations(inode);
1023         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1024         if (EXT4_I(inode)->jinode) {
1025                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1026                                                EXT4_I(inode)->jinode);
1027                 jbd2_free_inode(EXT4_I(inode)->jinode);
1028                 EXT4_I(inode)->jinode = NULL;
1029         }
1030 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1031         if (EXT4_I(inode)->i_crypt_info)
1032                 ext4_free_encryption_info(inode, EXT4_I(inode)->i_crypt_info);
1033 #endif
1034 }
1035
1036 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1037                                         u64 ino, u32 generation)
1038 {
1039         struct inode *inode;
1040
1041         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1042                 return ERR_PTR(-ESTALE);
1043         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1044                 return ERR_PTR(-ESTALE);
1045
1046         /* iget isn't really right if the inode is currently unallocated!!
1047          *
1048          * ext4_read_inode will return a bad_inode if the inode had been
1049          * deleted, so we should be safe.
1050          *
1051          * Currently we don't know the generation for parent directory, so
1052          * a generation of 0 means "accept any"
1053          */
1054         inode = ext4_iget_normal(sb, ino);
1055         if (IS_ERR(inode))
1056                 return ERR_CAST(inode);
1057         if (generation && inode->i_generation != generation) {
1058                 iput(inode);
1059                 return ERR_PTR(-ESTALE);
1060         }
1061
1062         return inode;
1063 }
1064
1065 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1066                                         int fh_len, int fh_type)
1067 {
1068         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1069                                     ext4_nfs_get_inode);
1070 }
1071
1072 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1073                                         int fh_len, int fh_type)
1074 {
1075         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1076                                     ext4_nfs_get_inode);
1077 }
1078
1079 /*
1080  * Try to release metadata pages (indirect blocks, directories) which are
1081  * mapped via the block device.  Since these pages could have journal heads
1082  * which would prevent try_to_free_buffers() from freeing them, we must use
1083  * jbd2 layer's try_to_free_buffers() function to release them.
1084  */
1085 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1086                                  gfp_t wait)
1087 {
1088         journal_t *journal = EXT4_SB(sb)->s_journal;
1089
1090         WARN_ON(PageChecked(page));
1091         if (!page_has_buffers(page))
1092                 return 0;
1093         if (journal)
1094                 return jbd2_journal_try_to_free_buffers(journal, page,
1095                                                 wait & ~__GFP_DIRECT_RECLAIM);
1096         return try_to_free_buffers(page);
1097 }
1098
1099 #ifdef CONFIG_QUOTA
1100 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1101 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1102
1103 static int ext4_write_dquot(struct dquot *dquot);
1104 static int ext4_acquire_dquot(struct dquot *dquot);
1105 static int ext4_release_dquot(struct dquot *dquot);
1106 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1107 static int ext4_write_info(struct super_block *sb, int type);
1108 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1109                          struct path *path);
1110 static int ext4_quota_off(struct super_block *sb, int type);
1111 static int ext4_quota_on_mount(struct super_block *sb, int type);
1112 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1113                                size_t len, loff_t off);
1114 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1115                                 const char *data, size_t len, loff_t off);
1116 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1117                              unsigned int flags);
1118 static int ext4_enable_quotas(struct super_block *sb);
1119
1120 static struct dquot **ext4_get_dquots(struct inode *inode)
1121 {
1122         return EXT4_I(inode)->i_dquot;
1123 }
1124
1125 static const struct dquot_operations ext4_quota_operations = {
1126         .get_reserved_space = ext4_get_reserved_space,
1127         .write_dquot    = ext4_write_dquot,
1128         .acquire_dquot  = ext4_acquire_dquot,
1129         .release_dquot  = ext4_release_dquot,
1130         .mark_dirty     = ext4_mark_dquot_dirty,
1131         .write_info     = ext4_write_info,
1132         .alloc_dquot    = dquot_alloc,
1133         .destroy_dquot  = dquot_destroy,
1134 };
1135
1136 static const struct quotactl_ops ext4_qctl_operations = {
1137         .quota_on       = ext4_quota_on,
1138         .quota_off      = ext4_quota_off,
1139         .quota_sync     = dquot_quota_sync,
1140         .get_state      = dquot_get_state,
1141         .set_info       = dquot_set_dqinfo,
1142         .get_dqblk      = dquot_get_dqblk,
1143         .set_dqblk      = dquot_set_dqblk
1144 };
1145 #endif
1146
1147 static const struct super_operations ext4_sops = {
1148         .alloc_inode    = ext4_alloc_inode,
1149         .destroy_inode  = ext4_destroy_inode,
1150         .write_inode    = ext4_write_inode,
1151         .dirty_inode    = ext4_dirty_inode,
1152         .drop_inode     = ext4_drop_inode,
1153         .evict_inode    = ext4_evict_inode,
1154         .put_super      = ext4_put_super,
1155         .sync_fs        = ext4_sync_fs,
1156         .freeze_fs      = ext4_freeze,
1157         .unfreeze_fs    = ext4_unfreeze,
1158         .statfs         = ext4_statfs,
1159         .remount_fs     = ext4_remount,
1160         .show_options   = ext4_show_options,
1161 #ifdef CONFIG_QUOTA
1162         .quota_read     = ext4_quota_read,
1163         .quota_write    = ext4_quota_write,
1164         .get_dquots     = ext4_get_dquots,
1165 #endif
1166         .bdev_try_to_free_page = bdev_try_to_free_page,
1167 };
1168
1169 static const struct export_operations ext4_export_ops = {
1170         .fh_to_dentry = ext4_fh_to_dentry,
1171         .fh_to_parent = ext4_fh_to_parent,
1172         .get_parent = ext4_get_parent,
1173 };
1174
1175 enum {
1176         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1177         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1178         Opt_nouid32, Opt_debug, Opt_removed,
1179         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1180         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1181         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1182         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1183         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1184         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1185         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1186         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1187         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1188         Opt_usrquota, Opt_grpquota, Opt_i_version, Opt_dax,
1189         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1190         Opt_lazytime, Opt_nolazytime,
1191         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1192         Opt_inode_readahead_blks, Opt_journal_ioprio,
1193         Opt_dioread_nolock, Opt_dioread_lock,
1194         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1195         Opt_max_dir_size_kb, Opt_nojournal_checksum,
1196 };
1197
1198 static const match_table_t tokens = {
1199         {Opt_bsd_df, "bsddf"},
1200         {Opt_minix_df, "minixdf"},
1201         {Opt_grpid, "grpid"},
1202         {Opt_grpid, "bsdgroups"},
1203         {Opt_nogrpid, "nogrpid"},
1204         {Opt_nogrpid, "sysvgroups"},
1205         {Opt_resgid, "resgid=%u"},
1206         {Opt_resuid, "resuid=%u"},
1207         {Opt_sb, "sb=%u"},
1208         {Opt_err_cont, "errors=continue"},
1209         {Opt_err_panic, "errors=panic"},
1210         {Opt_err_ro, "errors=remount-ro"},
1211         {Opt_nouid32, "nouid32"},
1212         {Opt_debug, "debug"},
1213         {Opt_removed, "oldalloc"},
1214         {Opt_removed, "orlov"},
1215         {Opt_user_xattr, "user_xattr"},
1216         {Opt_nouser_xattr, "nouser_xattr"},
1217         {Opt_acl, "acl"},
1218         {Opt_noacl, "noacl"},
1219         {Opt_noload, "norecovery"},
1220         {Opt_noload, "noload"},
1221         {Opt_removed, "nobh"},
1222         {Opt_removed, "bh"},
1223         {Opt_commit, "commit=%u"},
1224         {Opt_min_batch_time, "min_batch_time=%u"},
1225         {Opt_max_batch_time, "max_batch_time=%u"},
1226         {Opt_journal_dev, "journal_dev=%u"},
1227         {Opt_journal_path, "journal_path=%s"},
1228         {Opt_journal_checksum, "journal_checksum"},
1229         {Opt_nojournal_checksum, "nojournal_checksum"},
1230         {Opt_journal_async_commit, "journal_async_commit"},
1231         {Opt_abort, "abort"},
1232         {Opt_data_journal, "data=journal"},
1233         {Opt_data_ordered, "data=ordered"},
1234         {Opt_data_writeback, "data=writeback"},
1235         {Opt_data_err_abort, "data_err=abort"},
1236         {Opt_data_err_ignore, "data_err=ignore"},
1237         {Opt_offusrjquota, "usrjquota="},
1238         {Opt_usrjquota, "usrjquota=%s"},
1239         {Opt_offgrpjquota, "grpjquota="},
1240         {Opt_grpjquota, "grpjquota=%s"},
1241         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1242         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1243         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1244         {Opt_grpquota, "grpquota"},
1245         {Opt_noquota, "noquota"},
1246         {Opt_quota, "quota"},
1247         {Opt_usrquota, "usrquota"},
1248         {Opt_barrier, "barrier=%u"},
1249         {Opt_barrier, "barrier"},
1250         {Opt_nobarrier, "nobarrier"},
1251         {Opt_i_version, "i_version"},
1252         {Opt_dax, "dax"},
1253         {Opt_stripe, "stripe=%u"},
1254         {Opt_delalloc, "delalloc"},
1255         {Opt_lazytime, "lazytime"},
1256         {Opt_nolazytime, "nolazytime"},
1257         {Opt_nodelalloc, "nodelalloc"},
1258         {Opt_removed, "mblk_io_submit"},
1259         {Opt_removed, "nomblk_io_submit"},
1260         {Opt_block_validity, "block_validity"},
1261         {Opt_noblock_validity, "noblock_validity"},
1262         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1263         {Opt_journal_ioprio, "journal_ioprio=%u"},
1264         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1265         {Opt_auto_da_alloc, "auto_da_alloc"},
1266         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1267         {Opt_dioread_nolock, "dioread_nolock"},
1268         {Opt_dioread_lock, "dioread_lock"},
1269         {Opt_discard, "discard"},
1270         {Opt_nodiscard, "nodiscard"},
1271         {Opt_init_itable, "init_itable=%u"},
1272         {Opt_init_itable, "init_itable"},
1273         {Opt_noinit_itable, "noinit_itable"},
1274         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1275         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1276         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1277         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1278         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1279         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1280         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1281         {Opt_err, NULL},
1282 };
1283
1284 static ext4_fsblk_t get_sb_block(void **data)
1285 {
1286         ext4_fsblk_t    sb_block;
1287         char            *options = (char *) *data;
1288
1289         if (!options || strncmp(options, "sb=", 3) != 0)
1290                 return 1;       /* Default location */
1291
1292         options += 3;
1293         /* TODO: use simple_strtoll with >32bit ext4 */
1294         sb_block = simple_strtoul(options, &options, 0);
1295         if (*options && *options != ',') {
1296                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1297                        (char *) *data);
1298                 return 1;
1299         }
1300         if (*options == ',')
1301                 options++;
1302         *data = (void *) options;
1303
1304         return sb_block;
1305 }
1306
1307 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1308 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1309         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1310
1311 #ifdef CONFIG_QUOTA
1312 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1313 {
1314         struct ext4_sb_info *sbi = EXT4_SB(sb);
1315         char *qname;
1316         int ret = -1;
1317
1318         if (sb_any_quota_loaded(sb) &&
1319                 !sbi->s_qf_names[qtype]) {
1320                 ext4_msg(sb, KERN_ERR,
1321                         "Cannot change journaled "
1322                         "quota options when quota turned on");
1323                 return -1;
1324         }
1325         if (ext4_has_feature_quota(sb)) {
1326                 ext4_msg(sb, KERN_ERR, "Cannot set journaled quota options "
1327                          "when QUOTA feature is enabled");
1328                 return -1;
1329         }
1330         qname = match_strdup(args);
1331         if (!qname) {
1332                 ext4_msg(sb, KERN_ERR,
1333                         "Not enough memory for storing quotafile name");
1334                 return -1;
1335         }
1336         if (sbi->s_qf_names[qtype]) {
1337                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1338                         ret = 1;
1339                 else
1340                         ext4_msg(sb, KERN_ERR,
1341                                  "%s quota file already specified",
1342                                  QTYPE2NAME(qtype));
1343                 goto errout;
1344         }
1345         if (strchr(qname, '/')) {
1346                 ext4_msg(sb, KERN_ERR,
1347                         "quotafile must be on filesystem root");
1348                 goto errout;
1349         }
1350         sbi->s_qf_names[qtype] = qname;
1351         set_opt(sb, QUOTA);
1352         return 1;
1353 errout:
1354         kfree(qname);
1355         return ret;
1356 }
1357
1358 static int clear_qf_name(struct super_block *sb, int qtype)
1359 {
1360
1361         struct ext4_sb_info *sbi = EXT4_SB(sb);
1362
1363         if (sb_any_quota_loaded(sb) &&
1364                 sbi->s_qf_names[qtype]) {
1365                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1366                         " when quota turned on");
1367                 return -1;
1368         }
1369         kfree(sbi->s_qf_names[qtype]);
1370         sbi->s_qf_names[qtype] = NULL;
1371         return 1;
1372 }
1373 #endif
1374
1375 #define MOPT_SET        0x0001
1376 #define MOPT_CLEAR      0x0002
1377 #define MOPT_NOSUPPORT  0x0004
1378 #define MOPT_EXPLICIT   0x0008
1379 #define MOPT_CLEAR_ERR  0x0010
1380 #define MOPT_GTE0       0x0020
1381 #ifdef CONFIG_QUOTA
1382 #define MOPT_Q          0
1383 #define MOPT_QFMT       0x0040
1384 #else
1385 #define MOPT_Q          MOPT_NOSUPPORT
1386 #define MOPT_QFMT       MOPT_NOSUPPORT
1387 #endif
1388 #define MOPT_DATAJ      0x0080
1389 #define MOPT_NO_EXT2    0x0100
1390 #define MOPT_NO_EXT3    0x0200
1391 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1392 #define MOPT_STRING     0x0400
1393
1394 static const struct mount_opts {
1395         int     token;
1396         int     mount_opt;
1397         int     flags;
1398 } ext4_mount_opts[] = {
1399         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1400         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1401         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1402         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1403         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1404         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1405         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1406          MOPT_EXT4_ONLY | MOPT_SET},
1407         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1408          MOPT_EXT4_ONLY | MOPT_CLEAR},
1409         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1410         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1411         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1412          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1413         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1414          MOPT_EXT4_ONLY | MOPT_CLEAR},
1415         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1416          MOPT_EXT4_ONLY | MOPT_CLEAR},
1417         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1418          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1419         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1420                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1421          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1422         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1423         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1424         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1425         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1426         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1427          MOPT_NO_EXT2 | MOPT_SET},
1428         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1429          MOPT_NO_EXT2 | MOPT_CLEAR},
1430         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1431         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1432         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1433         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1434         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1435         {Opt_commit, 0, MOPT_GTE0},
1436         {Opt_max_batch_time, 0, MOPT_GTE0},
1437         {Opt_min_batch_time, 0, MOPT_GTE0},
1438         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1439         {Opt_init_itable, 0, MOPT_GTE0},
1440         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1441         {Opt_stripe, 0, MOPT_GTE0},
1442         {Opt_resuid, 0, MOPT_GTE0},
1443         {Opt_resgid, 0, MOPT_GTE0},
1444         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1445         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1446         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1447         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1448         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1449         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1450          MOPT_NO_EXT2 | MOPT_DATAJ},
1451         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1452         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1453 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1454         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1455         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1456 #else
1457         {Opt_acl, 0, MOPT_NOSUPPORT},
1458         {Opt_noacl, 0, MOPT_NOSUPPORT},
1459 #endif
1460         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1461         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1462         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1463         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1464                                                         MOPT_SET | MOPT_Q},
1465         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1466                                                         MOPT_SET | MOPT_Q},
1467         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1468                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1469         {Opt_usrjquota, 0, MOPT_Q},
1470         {Opt_grpjquota, 0, MOPT_Q},
1471         {Opt_offusrjquota, 0, MOPT_Q},
1472         {Opt_offgrpjquota, 0, MOPT_Q},
1473         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1474         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1475         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1476         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1477         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1478         {Opt_err, 0, 0}
1479 };
1480
1481 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1482                             substring_t *args, unsigned long *journal_devnum,
1483                             unsigned int *journal_ioprio, int is_remount)
1484 {
1485         struct ext4_sb_info *sbi = EXT4_SB(sb);
1486         const struct mount_opts *m;
1487         kuid_t uid;
1488         kgid_t gid;
1489         int arg = 0;
1490
1491 #ifdef CONFIG_QUOTA
1492         if (token == Opt_usrjquota)
1493                 return set_qf_name(sb, USRQUOTA, &args[0]);
1494         else if (token == Opt_grpjquota)
1495                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1496         else if (token == Opt_offusrjquota)
1497                 return clear_qf_name(sb, USRQUOTA);
1498         else if (token == Opt_offgrpjquota)
1499                 return clear_qf_name(sb, GRPQUOTA);
1500 #endif
1501         switch (token) {
1502         case Opt_noacl:
1503         case Opt_nouser_xattr:
1504                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1505                 break;
1506         case Opt_sb:
1507                 return 1;       /* handled by get_sb_block() */
1508         case Opt_removed:
1509                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1510                 return 1;
1511         case Opt_abort:
1512                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1513                 return 1;
1514         case Opt_i_version:
1515                 sb->s_flags |= MS_I_VERSION;
1516                 return 1;
1517         case Opt_lazytime:
1518                 sb->s_flags |= MS_LAZYTIME;
1519                 return 1;
1520         case Opt_nolazytime:
1521                 sb->s_flags &= ~MS_LAZYTIME;
1522                 return 1;
1523         }
1524
1525         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1526                 if (token == m->token)
1527                         break;
1528
1529         if (m->token == Opt_err) {
1530                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1531                          "or missing value", opt);
1532                 return -1;
1533         }
1534
1535         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1536                 ext4_msg(sb, KERN_ERR,
1537                          "Mount option \"%s\" incompatible with ext2", opt);
1538                 return -1;
1539         }
1540         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1541                 ext4_msg(sb, KERN_ERR,
1542                          "Mount option \"%s\" incompatible with ext3", opt);
1543                 return -1;
1544         }
1545
1546         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1547                 return -1;
1548         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1549                 return -1;
1550         if (m->flags & MOPT_EXPLICIT) {
1551                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1552                         set_opt2(sb, EXPLICIT_DELALLOC);
1553                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1554                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1555                 } else
1556                         return -1;
1557         }
1558         if (m->flags & MOPT_CLEAR_ERR)
1559                 clear_opt(sb, ERRORS_MASK);
1560         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1561                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1562                          "options when quota turned on");
1563                 return -1;
1564         }
1565
1566         if (m->flags & MOPT_NOSUPPORT) {
1567                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1568         } else if (token == Opt_commit) {
1569                 if (arg == 0)
1570                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1571                 sbi->s_commit_interval = HZ * arg;
1572         } else if (token == Opt_max_batch_time) {
1573                 sbi->s_max_batch_time = arg;
1574         } else if (token == Opt_min_batch_time) {
1575                 sbi->s_min_batch_time = arg;
1576         } else if (token == Opt_inode_readahead_blks) {
1577                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1578                         ext4_msg(sb, KERN_ERR,
1579                                  "EXT4-fs: inode_readahead_blks must be "
1580                                  "0 or a power of 2 smaller than 2^31");
1581                         return -1;
1582                 }
1583                 sbi->s_inode_readahead_blks = arg;
1584         } else if (token == Opt_init_itable) {
1585                 set_opt(sb, INIT_INODE_TABLE);
1586                 if (!args->from)
1587                         arg = EXT4_DEF_LI_WAIT_MULT;
1588                 sbi->s_li_wait_mult = arg;
1589         } else if (token == Opt_max_dir_size_kb) {
1590                 sbi->s_max_dir_size_kb = arg;
1591         } else if (token == Opt_stripe) {
1592                 sbi->s_stripe = arg;
1593         } else if (token == Opt_resuid) {
1594                 uid = make_kuid(current_user_ns(), arg);
1595                 if (!uid_valid(uid)) {
1596                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1597                         return -1;
1598                 }
1599                 sbi->s_resuid = uid;
1600         } else if (token == Opt_resgid) {
1601                 gid = make_kgid(current_user_ns(), arg);
1602                 if (!gid_valid(gid)) {
1603                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1604                         return -1;
1605                 }
1606                 sbi->s_resgid = gid;
1607         } else if (token == Opt_journal_dev) {
1608                 if (is_remount) {
1609                         ext4_msg(sb, KERN_ERR,
1610                                  "Cannot specify journal on remount");
1611                         return -1;
1612                 }
1613                 *journal_devnum = arg;
1614         } else if (token == Opt_journal_path) {
1615                 char *journal_path;
1616                 struct inode *journal_inode;
1617                 struct path path;
1618                 int error;
1619
1620                 if (is_remount) {
1621                         ext4_msg(sb, KERN_ERR,
1622                                  "Cannot specify journal on remount");
1623                         return -1;
1624                 }
1625                 journal_path = match_strdup(&args[0]);
1626                 if (!journal_path) {
1627                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1628                                 "journal device string");
1629                         return -1;
1630                 }
1631
1632                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1633                 if (error) {
1634                         ext4_msg(sb, KERN_ERR, "error: could not find "
1635                                 "journal device path: error %d", error);
1636                         kfree(journal_path);
1637                         return -1;
1638                 }
1639
1640                 journal_inode = d_inode(path.dentry);
1641                 if (!S_ISBLK(journal_inode->i_mode)) {
1642                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1643                                 "is not a block device", journal_path);
1644                         path_put(&path);
1645                         kfree(journal_path);
1646                         return -1;
1647                 }
1648
1649                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1650                 path_put(&path);
1651                 kfree(journal_path);
1652         } else if (token == Opt_journal_ioprio) {
1653                 if (arg > 7) {
1654                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1655                                  " (must be 0-7)");
1656                         return -1;
1657                 }
1658                 *journal_ioprio =
1659                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1660         } else if (token == Opt_test_dummy_encryption) {
1661 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1662                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1663                 ext4_msg(sb, KERN_WARNING,
1664                          "Test dummy encryption mode enabled");
1665 #else
1666                 ext4_msg(sb, KERN_WARNING,
1667                          "Test dummy encryption mount option ignored");
1668 #endif
1669         } else if (m->flags & MOPT_DATAJ) {
1670                 if (is_remount) {
1671                         if (!sbi->s_journal)
1672                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1673                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1674                                 ext4_msg(sb, KERN_ERR,
1675                                          "Cannot change data mode on remount");
1676                                 return -1;
1677                         }
1678                 } else {
1679                         clear_opt(sb, DATA_FLAGS);
1680                         sbi->s_mount_opt |= m->mount_opt;
1681                 }
1682 #ifdef CONFIG_QUOTA
1683         } else if (m->flags & MOPT_QFMT) {
1684                 if (sb_any_quota_loaded(sb) &&
1685                     sbi->s_jquota_fmt != m->mount_opt) {
1686                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1687                                  "quota options when quota turned on");
1688                         return -1;
1689                 }
1690                 if (ext4_has_feature_quota(sb)) {
1691                         ext4_msg(sb, KERN_ERR,
1692                                  "Cannot set journaled quota options "
1693                                  "when QUOTA feature is enabled");
1694                         return -1;
1695                 }
1696                 sbi->s_jquota_fmt = m->mount_opt;
1697 #endif
1698         } else if (token == Opt_dax) {
1699 #ifdef CONFIG_FS_DAX
1700                 ext4_msg(sb, KERN_WARNING,
1701                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1702                         sbi->s_mount_opt |= m->mount_opt;
1703 #else
1704                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1705                 return -1;
1706 #endif
1707         } else {
1708                 if (!args->from)
1709                         arg = 1;
1710                 if (m->flags & MOPT_CLEAR)
1711                         arg = !arg;
1712                 else if (unlikely(!(m->flags & MOPT_SET))) {
1713                         ext4_msg(sb, KERN_WARNING,
1714                                  "buggy handling of option %s", opt);
1715                         WARN_ON(1);
1716                         return -1;
1717                 }
1718                 if (arg != 0)
1719                         sbi->s_mount_opt |= m->mount_opt;
1720                 else
1721                         sbi->s_mount_opt &= ~m->mount_opt;
1722         }
1723         return 1;
1724 }
1725
1726 static int parse_options(char *options, struct super_block *sb,
1727                          unsigned long *journal_devnum,
1728                          unsigned int *journal_ioprio,
1729                          int is_remount)
1730 {
1731         struct ext4_sb_info *sbi = EXT4_SB(sb);
1732         char *p;
1733         substring_t args[MAX_OPT_ARGS];
1734         int token;
1735
1736         if (!options)
1737                 return 1;
1738
1739         while ((p = strsep(&options, ",")) != NULL) {
1740                 if (!*p)
1741                         continue;
1742                 /*
1743                  * Initialize args struct so we know whether arg was
1744                  * found; some options take optional arguments.
1745                  */
1746                 args[0].to = args[0].from = NULL;
1747                 token = match_token(p, tokens, args);
1748                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1749                                      journal_ioprio, is_remount) < 0)
1750                         return 0;
1751         }
1752 #ifdef CONFIG_QUOTA
1753         if (ext4_has_feature_quota(sb) &&
1754             (test_opt(sb, USRQUOTA) || test_opt(sb, GRPQUOTA))) {
1755                 ext4_msg(sb, KERN_ERR, "Cannot set quota options when QUOTA "
1756                          "feature is enabled");
1757                 return 0;
1758         }
1759         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1760                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1761                         clear_opt(sb, USRQUOTA);
1762
1763                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1764                         clear_opt(sb, GRPQUOTA);
1765
1766                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1767                         ext4_msg(sb, KERN_ERR, "old and new quota "
1768                                         "format mixing");
1769                         return 0;
1770                 }
1771
1772                 if (!sbi->s_jquota_fmt) {
1773                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1774                                         "not specified");
1775                         return 0;
1776                 }
1777         }
1778 #endif
1779         if (test_opt(sb, DIOREAD_NOLOCK)) {
1780                 int blocksize =
1781                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1782
1783                 if (blocksize < PAGE_CACHE_SIZE) {
1784                         ext4_msg(sb, KERN_ERR, "can't mount with "
1785                                  "dioread_nolock if block size != PAGE_SIZE");
1786                         return 0;
1787                 }
1788         }
1789         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
1790             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
1791                 ext4_msg(sb, KERN_ERR, "can't mount with journal_async_commit "
1792                          "in data=ordered mode");
1793                 return 0;
1794         }
1795         return 1;
1796 }
1797
1798 static inline void ext4_show_quota_options(struct seq_file *seq,
1799                                            struct super_block *sb)
1800 {
1801 #if defined(CONFIG_QUOTA)
1802         struct ext4_sb_info *sbi = EXT4_SB(sb);
1803
1804         if (sbi->s_jquota_fmt) {
1805                 char *fmtname = "";
1806
1807                 switch (sbi->s_jquota_fmt) {
1808                 case QFMT_VFS_OLD:
1809                         fmtname = "vfsold";
1810                         break;
1811                 case QFMT_VFS_V0:
1812                         fmtname = "vfsv0";
1813                         break;
1814                 case QFMT_VFS_V1:
1815                         fmtname = "vfsv1";
1816                         break;
1817                 }
1818                 seq_printf(seq, ",jqfmt=%s", fmtname);
1819         }
1820
1821         if (sbi->s_qf_names[USRQUOTA])
1822                 seq_show_option(seq, "usrjquota", sbi->s_qf_names[USRQUOTA]);
1823
1824         if (sbi->s_qf_names[GRPQUOTA])
1825                 seq_show_option(seq, "grpjquota", sbi->s_qf_names[GRPQUOTA]);
1826 #endif
1827 }
1828
1829 static const char *token2str(int token)
1830 {
1831         const struct match_token *t;
1832
1833         for (t = tokens; t->token != Opt_err; t++)
1834                 if (t->token == token && !strchr(t->pattern, '='))
1835                         break;
1836         return t->pattern;
1837 }
1838
1839 /*
1840  * Show an option if
1841  *  - it's set to a non-default value OR
1842  *  - if the per-sb default is different from the global default
1843  */
1844 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1845                               int nodefs)
1846 {
1847         struct ext4_sb_info *sbi = EXT4_SB(sb);
1848         struct ext4_super_block *es = sbi->s_es;
1849         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1850         const struct mount_opts *m;
1851         char sep = nodefs ? '\n' : ',';
1852
1853 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1854 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1855
1856         if (sbi->s_sb_block != 1)
1857                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1858
1859         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1860                 int want_set = m->flags & MOPT_SET;
1861                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1862                     (m->flags & MOPT_CLEAR_ERR))
1863                         continue;
1864                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1865                         continue; /* skip if same as the default */
1866                 if ((want_set &&
1867                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1868                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1869                         continue; /* select Opt_noFoo vs Opt_Foo */
1870                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1871         }
1872
1873         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1874             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1875                 SEQ_OPTS_PRINT("resuid=%u",
1876                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1877         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1878             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1879                 SEQ_OPTS_PRINT("resgid=%u",
1880                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1881         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1882         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1883                 SEQ_OPTS_PUTS("errors=remount-ro");
1884         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1885                 SEQ_OPTS_PUTS("errors=continue");
1886         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1887                 SEQ_OPTS_PUTS("errors=panic");
1888         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1889                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1890         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1891                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1892         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1893                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1894         if (sb->s_flags & MS_I_VERSION)
1895                 SEQ_OPTS_PUTS("i_version");
1896         if (nodefs || sbi->s_stripe)
1897                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1898         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1899                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1900                         SEQ_OPTS_PUTS("data=journal");
1901                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1902                         SEQ_OPTS_PUTS("data=ordered");
1903                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1904                         SEQ_OPTS_PUTS("data=writeback");
1905         }
1906         if (nodefs ||
1907             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1908                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1909                                sbi->s_inode_readahead_blks);
1910
1911         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1912                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1913                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1914         if (nodefs || sbi->s_max_dir_size_kb)
1915                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1916
1917         ext4_show_quota_options(seq, sb);
1918         return 0;
1919 }
1920
1921 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1922 {
1923         return _ext4_show_options(seq, root->d_sb, 0);
1924 }
1925
1926 int ext4_seq_options_show(struct seq_file *seq, void *offset)
1927 {
1928         struct super_block *sb = seq->private;
1929         int rc;
1930
1931         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1932         rc = _ext4_show_options(seq, sb, 1);
1933         seq_puts(seq, "\n");
1934         return rc;
1935 }
1936
1937 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1938                             int read_only)
1939 {
1940         struct ext4_sb_info *sbi = EXT4_SB(sb);
1941         int res = 0;
1942
1943         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1944                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1945                          "forcing read-only mode");
1946                 res = MS_RDONLY;
1947         }
1948         if (read_only)
1949                 goto done;
1950         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1951                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1952                          "running e2fsck is recommended");
1953         else if (sbi->s_mount_state & EXT4_ERROR_FS)
1954                 ext4_msg(sb, KERN_WARNING,
1955                          "warning: mounting fs with errors, "
1956                          "running e2fsck is recommended");
1957         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1958                  le16_to_cpu(es->s_mnt_count) >=
1959                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1960                 ext4_msg(sb, KERN_WARNING,
1961                          "warning: maximal mount count reached, "
1962                          "running e2fsck is recommended");
1963         else if (le32_to_cpu(es->s_checkinterval) &&
1964                 (le32_to_cpu(es->s_lastcheck) +
1965                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1966                 ext4_msg(sb, KERN_WARNING,
1967                          "warning: checktime reached, "
1968                          "running e2fsck is recommended");
1969         if (!sbi->s_journal)
1970                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1971         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1972                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1973         le16_add_cpu(&es->s_mnt_count, 1);
1974         es->s_mtime = cpu_to_le32(get_seconds());
1975         ext4_update_dynamic_rev(sb);
1976         if (sbi->s_journal)
1977                 ext4_set_feature_journal_needs_recovery(sb);
1978
1979         ext4_commit_super(sb, 1);
1980 done:
1981         if (test_opt(sb, DEBUG))
1982                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1983                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1984                         sb->s_blocksize,
1985                         sbi->s_groups_count,
1986                         EXT4_BLOCKS_PER_GROUP(sb),
1987                         EXT4_INODES_PER_GROUP(sb),
1988                         sbi->s_mount_opt, sbi->s_mount_opt2);
1989
1990         cleancache_init_fs(sb);
1991         return res;
1992 }
1993
1994 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1995 {
1996         struct ext4_sb_info *sbi = EXT4_SB(sb);
1997         struct flex_groups *new_groups;
1998         int size;
1999
2000         if (!sbi->s_log_groups_per_flex)
2001                 return 0;
2002
2003         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2004         if (size <= sbi->s_flex_groups_allocated)
2005                 return 0;
2006
2007         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
2008         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
2009         if (!new_groups) {
2010                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
2011                          size / (int) sizeof(struct flex_groups));
2012                 return -ENOMEM;
2013         }
2014
2015         if (sbi->s_flex_groups) {
2016                 memcpy(new_groups, sbi->s_flex_groups,
2017                        (sbi->s_flex_groups_allocated *
2018                         sizeof(struct flex_groups)));
2019                 kvfree(sbi->s_flex_groups);
2020         }
2021         sbi->s_flex_groups = new_groups;
2022         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
2023         return 0;
2024 }
2025
2026 static int ext4_fill_flex_info(struct super_block *sb)
2027 {
2028         struct ext4_sb_info *sbi = EXT4_SB(sb);
2029         struct ext4_group_desc *gdp = NULL;
2030         ext4_group_t flex_group;
2031         int i, err;
2032
2033         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2034         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2035                 sbi->s_log_groups_per_flex = 0;
2036                 return 1;
2037         }
2038
2039         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2040         if (err)
2041                 goto failed;
2042
2043         for (i = 0; i < sbi->s_groups_count; i++) {
2044                 gdp = ext4_get_group_desc(sb, i, NULL);
2045
2046                 flex_group = ext4_flex_group(sbi, i);
2047                 atomic_add(ext4_free_inodes_count(sb, gdp),
2048                            &sbi->s_flex_groups[flex_group].free_inodes);
2049                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2050                              &sbi->s_flex_groups[flex_group].free_clusters);
2051                 atomic_add(ext4_used_dirs_count(sb, gdp),
2052                            &sbi->s_flex_groups[flex_group].used_dirs);
2053         }
2054
2055         return 1;
2056 failed:
2057         return 0;
2058 }
2059
2060 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2061                                    struct ext4_group_desc *gdp)
2062 {
2063         int offset;
2064         __u16 crc = 0;
2065         __le32 le_group = cpu_to_le32(block_group);
2066         struct ext4_sb_info *sbi = EXT4_SB(sb);
2067
2068         if (ext4_has_metadata_csum(sbi->s_sb)) {
2069                 /* Use new metadata_csum algorithm */
2070                 __le16 save_csum;
2071                 __u32 csum32;
2072
2073                 save_csum = gdp->bg_checksum;
2074                 gdp->bg_checksum = 0;
2075                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2076                                      sizeof(le_group));
2077                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
2078                                      sbi->s_desc_size);
2079                 gdp->bg_checksum = save_csum;
2080
2081                 crc = csum32 & 0xFFFF;
2082                 goto out;
2083         }
2084
2085         /* old crc16 code */
2086         if (!ext4_has_feature_gdt_csum(sb))
2087                 return 0;
2088
2089         offset = offsetof(struct ext4_group_desc, bg_checksum);
2090
2091         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2092         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2093         crc = crc16(crc, (__u8 *)gdp, offset);
2094         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2095         /* for checksum of struct ext4_group_desc do the rest...*/
2096         if (ext4_has_feature_64bit(sb) &&
2097             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2098                 crc = crc16(crc, (__u8 *)gdp + offset,
2099                             le16_to_cpu(sbi->s_es->s_desc_size) -
2100                                 offset);
2101
2102 out:
2103         return cpu_to_le16(crc);
2104 }
2105
2106 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2107                                 struct ext4_group_desc *gdp)
2108 {
2109         if (ext4_has_group_desc_csum(sb) &&
2110             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2111                 return 0;
2112
2113         return 1;
2114 }
2115
2116 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2117                               struct ext4_group_desc *gdp)
2118 {
2119         if (!ext4_has_group_desc_csum(sb))
2120                 return;
2121         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2122 }
2123
2124 /* Called at mount-time, super-block is locked */
2125 static int ext4_check_descriptors(struct super_block *sb,
2126                                   ext4_group_t *first_not_zeroed)
2127 {
2128         struct ext4_sb_info *sbi = EXT4_SB(sb);
2129         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2130         ext4_fsblk_t last_block;
2131         ext4_fsblk_t block_bitmap;
2132         ext4_fsblk_t inode_bitmap;
2133         ext4_fsblk_t inode_table;
2134         int flexbg_flag = 0;
2135         ext4_group_t i, grp = sbi->s_groups_count;
2136
2137         if (ext4_has_feature_flex_bg(sb))
2138                 flexbg_flag = 1;
2139
2140         ext4_debug("Checking group descriptors");
2141
2142         for (i = 0; i < sbi->s_groups_count; i++) {
2143                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2144
2145                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2146                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2147                 else
2148                         last_block = first_block +
2149                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2150
2151                 if ((grp == sbi->s_groups_count) &&
2152                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2153                         grp = i;
2154
2155                 block_bitmap = ext4_block_bitmap(sb, gdp);
2156                 if (block_bitmap < first_block || block_bitmap > last_block) {
2157                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2158                                "Block bitmap for group %u not in group "
2159                                "(block %llu)!", i, block_bitmap);
2160                         return 0;
2161                 }
2162                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2163                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2164                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2165                                "Inode bitmap for group %u not in group "
2166                                "(block %llu)!", i, inode_bitmap);
2167                         return 0;
2168                 }
2169                 inode_table = ext4_inode_table(sb, gdp);
2170                 if (inode_table < first_block ||
2171                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2172                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2173                                "Inode table for group %u not in group "
2174                                "(block %llu)!", i, inode_table);
2175                         return 0;
2176                 }
2177                 ext4_lock_group(sb, i);
2178                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2179                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2180                                  "Checksum for group %u failed (%u!=%u)",
2181                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2182                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2183                         if (!(sb->s_flags & MS_RDONLY)) {
2184                                 ext4_unlock_group(sb, i);
2185                                 return 0;
2186                         }
2187                 }
2188                 ext4_unlock_group(sb, i);
2189                 if (!flexbg_flag)
2190                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2191         }
2192         if (NULL != first_not_zeroed)
2193                 *first_not_zeroed = grp;
2194         return 1;
2195 }
2196
2197 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2198  * the superblock) which were deleted from all directories, but held open by
2199  * a process at the time of a crash.  We walk the list and try to delete these
2200  * inodes at recovery time (only with a read-write filesystem).
2201  *
2202  * In order to keep the orphan inode chain consistent during traversal (in
2203  * case of crash during recovery), we link each inode into the superblock
2204  * orphan list_head and handle it the same way as an inode deletion during
2205  * normal operation (which journals the operations for us).
2206  *
2207  * We only do an iget() and an iput() on each inode, which is very safe if we
2208  * accidentally point at an in-use or already deleted inode.  The worst that
2209  * can happen in this case is that we get a "bit already cleared" message from
2210  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2211  * e2fsck was run on this filesystem, and it must have already done the orphan
2212  * inode cleanup for us, so we can safely abort without any further action.
2213  */
2214 static void ext4_orphan_cleanup(struct super_block *sb,
2215                                 struct ext4_super_block *es)
2216 {
2217         unsigned int s_flags = sb->s_flags;
2218         int nr_orphans = 0, nr_truncates = 0;
2219 #ifdef CONFIG_QUOTA
2220         int i;
2221 #endif
2222         if (!es->s_last_orphan) {
2223                 jbd_debug(4, "no orphan inodes to clean up\n");
2224                 return;
2225         }
2226
2227         if (bdev_read_only(sb->s_bdev)) {
2228                 ext4_msg(sb, KERN_ERR, "write access "
2229                         "unavailable, skipping orphan cleanup");
2230                 return;
2231         }
2232
2233         /* Check if feature set would not allow a r/w mount */
2234         if (!ext4_feature_set_ok(sb, 0)) {
2235                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2236                          "unknown ROCOMPAT features");
2237                 return;
2238         }
2239
2240         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2241                 /* don't clear list on RO mount w/ errors */
2242                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2243                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2244                                   "clearing orphan list.\n");
2245                         es->s_last_orphan = 0;
2246                 }
2247                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2248                 return;
2249         }
2250
2251         if (s_flags & MS_RDONLY) {
2252                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2253                 sb->s_flags &= ~MS_RDONLY;
2254         }
2255 #ifdef CONFIG_QUOTA
2256         /* Needed for iput() to work correctly and not trash data */
2257         sb->s_flags |= MS_ACTIVE;
2258         /* Turn on quotas so that they are updated correctly */
2259         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2260                 if (EXT4_SB(sb)->s_qf_names[i]) {
2261                         int ret = ext4_quota_on_mount(sb, i);
2262                         if (ret < 0)
2263                                 ext4_msg(sb, KERN_ERR,
2264                                         "Cannot turn on journaled "
2265                                         "quota: error %d", ret);
2266                 }
2267         }
2268 #endif
2269
2270         while (es->s_last_orphan) {
2271                 struct inode *inode;
2272
2273                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2274                 if (IS_ERR(inode)) {
2275                         es->s_last_orphan = 0;
2276                         break;
2277                 }
2278
2279                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2280                 dquot_initialize(inode);
2281                 if (inode->i_nlink) {
2282                         if (test_opt(sb, DEBUG))
2283                                 ext4_msg(sb, KERN_DEBUG,
2284                                         "%s: truncating inode %lu to %lld bytes",
2285                                         __func__, inode->i_ino, inode->i_size);
2286                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2287                                   inode->i_ino, inode->i_size);
2288                         mutex_lock(&inode->i_mutex);
2289                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2290                         ext4_truncate(inode);
2291                         mutex_unlock(&inode->i_mutex);
2292                         nr_truncates++;
2293                 } else {
2294                         if (test_opt(sb, DEBUG))
2295                                 ext4_msg(sb, KERN_DEBUG,
2296                                         "%s: deleting unreferenced inode %lu",
2297                                         __func__, inode->i_ino);
2298                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2299                                   inode->i_ino);
2300                         nr_orphans++;
2301                 }
2302                 iput(inode);  /* The delete magic happens here! */
2303         }
2304
2305 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2306
2307         if (nr_orphans)
2308                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2309                        PLURAL(nr_orphans));
2310         if (nr_truncates)
2311                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2312                        PLURAL(nr_truncates));
2313 #ifdef CONFIG_QUOTA
2314         /* Turn quotas off */
2315         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2316                 if (sb_dqopt(sb)->files[i])
2317                         dquot_quota_off(sb, i);
2318         }
2319 #endif
2320         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2321 }
2322
2323 /*
2324  * Maximal extent format file size.
2325  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2326  * extent format containers, within a sector_t, and within i_blocks
2327  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2328  * so that won't be a limiting factor.
2329  *
2330  * However there is other limiting factor. We do store extents in the form
2331  * of starting block and length, hence the resulting length of the extent
2332  * covering maximum file size must fit into on-disk format containers as
2333  * well. Given that length is always by 1 unit bigger than max unit (because
2334  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2335  *
2336  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2337  */
2338 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2339 {
2340         loff_t res;
2341         loff_t upper_limit = MAX_LFS_FILESIZE;
2342
2343         /* small i_blocks in vfs inode? */
2344         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2345                 /*
2346                  * CONFIG_LBDAF is not enabled implies the inode
2347                  * i_block represent total blocks in 512 bytes
2348                  * 32 == size of vfs inode i_blocks * 8
2349                  */
2350                 upper_limit = (1LL << 32) - 1;
2351
2352                 /* total blocks in file system block size */
2353                 upper_limit >>= (blkbits - 9);
2354                 upper_limit <<= blkbits;
2355         }
2356
2357         /*
2358          * 32-bit extent-start container, ee_block. We lower the maxbytes
2359          * by one fs block, so ee_len can cover the extent of maximum file
2360          * size
2361          */
2362         res = (1LL << 32) - 1;
2363         res <<= blkbits;
2364
2365         /* Sanity check against vm- & vfs- imposed limits */
2366         if (res > upper_limit)
2367                 res = upper_limit;
2368
2369         return res;
2370 }
2371
2372 /*
2373  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2374  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2375  * We need to be 1 filesystem block less than the 2^48 sector limit.
2376  */
2377 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2378 {
2379         loff_t res = EXT4_NDIR_BLOCKS;
2380         int meta_blocks;
2381         loff_t upper_limit;
2382         /* This is calculated to be the largest file size for a dense, block
2383          * mapped file such that the file's total number of 512-byte sectors,
2384          * including data and all indirect blocks, does not exceed (2^48 - 1).
2385          *
2386          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2387          * number of 512-byte sectors of the file.
2388          */
2389
2390         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2391                 /*
2392                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2393                  * the inode i_block field represents total file blocks in
2394                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2395                  */
2396                 upper_limit = (1LL << 32) - 1;
2397
2398                 /* total blocks in file system block size */
2399                 upper_limit >>= (bits - 9);
2400
2401         } else {
2402                 /*
2403                  * We use 48 bit ext4_inode i_blocks
2404                  * With EXT4_HUGE_FILE_FL set the i_blocks
2405                  * represent total number of blocks in
2406                  * file system block size
2407                  */
2408                 upper_limit = (1LL << 48) - 1;
2409
2410         }
2411
2412         /* indirect blocks */
2413         meta_blocks = 1;
2414         /* double indirect blocks */
2415         meta_blocks += 1 + (1LL << (bits-2));
2416         /* tripple indirect blocks */
2417         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2418
2419         upper_limit -= meta_blocks;
2420         upper_limit <<= bits;
2421
2422         res += 1LL << (bits-2);
2423         res += 1LL << (2*(bits-2));
2424         res += 1LL << (3*(bits-2));
2425         res <<= bits;
2426         if (res > upper_limit)
2427                 res = upper_limit;
2428
2429         if (res > MAX_LFS_FILESIZE)
2430                 res = MAX_LFS_FILESIZE;
2431
2432         return res;
2433 }
2434
2435 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2436                                    ext4_fsblk_t logical_sb_block, int nr)
2437 {
2438         struct ext4_sb_info *sbi = EXT4_SB(sb);
2439         ext4_group_t bg, first_meta_bg;
2440         int has_super = 0;
2441
2442         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2443
2444         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2445                 return logical_sb_block + nr + 1;
2446         bg = sbi->s_desc_per_block * nr;
2447         if (ext4_bg_has_super(sb, bg))
2448                 has_super = 1;
2449
2450         /*
2451          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2452          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2453          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2454          * compensate.
2455          */
2456         if (sb->s_blocksize == 1024 && nr == 0 &&
2457             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2458                 has_super++;
2459
2460         return (has_super + ext4_group_first_block_no(sb, bg));
2461 }
2462
2463 /**
2464  * ext4_get_stripe_size: Get the stripe size.
2465  * @sbi: In memory super block info
2466  *
2467  * If we have specified it via mount option, then
2468  * use the mount option value. If the value specified at mount time is
2469  * greater than the blocks per group use the super block value.
2470  * If the super block value is greater than blocks per group return 0.
2471  * Allocator needs it be less than blocks per group.
2472  *
2473  */
2474 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2475 {
2476         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2477         unsigned long stripe_width =
2478                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2479         int ret;
2480
2481         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2482                 ret = sbi->s_stripe;
2483         else if (stripe_width <= sbi->s_blocks_per_group)
2484                 ret = stripe_width;
2485         else if (stride <= sbi->s_blocks_per_group)
2486                 ret = stride;
2487         else
2488                 ret = 0;
2489
2490         /*
2491          * If the stripe width is 1, this makes no sense and
2492          * we set it to 0 to turn off stripe handling code.
2493          */
2494         if (ret <= 1)
2495                 ret = 0;
2496
2497         return ret;
2498 }
2499
2500 /*
2501  * Check whether this filesystem can be mounted based on
2502  * the features present and the RDONLY/RDWR mount requested.
2503  * Returns 1 if this filesystem can be mounted as requested,
2504  * 0 if it cannot be.
2505  */
2506 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2507 {
2508         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2509                 ext4_msg(sb, KERN_ERR,
2510                         "Couldn't mount because of "
2511                         "unsupported optional features (%x)",
2512                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2513                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2514                 return 0;
2515         }
2516
2517         if (readonly)
2518                 return 1;
2519
2520         if (ext4_has_feature_readonly(sb)) {
2521                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2522                 sb->s_flags |= MS_RDONLY;
2523                 return 1;
2524         }
2525
2526         /* Check that feature set is OK for a read-write mount */
2527         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2528                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2529                          "unsupported optional features (%x)",
2530                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2531                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2532                 return 0;
2533         }
2534         /*
2535          * Large file size enabled file system can only be mounted
2536          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2537          */
2538         if (ext4_has_feature_huge_file(sb)) {
2539                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2540                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2541                                  "cannot be mounted RDWR without "
2542                                  "CONFIG_LBDAF");
2543                         return 0;
2544                 }
2545         }
2546         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2547                 ext4_msg(sb, KERN_ERR,
2548                          "Can't support bigalloc feature without "
2549                          "extents feature\n");
2550                 return 0;
2551         }
2552
2553 #ifndef CONFIG_QUOTA
2554         if (ext4_has_feature_quota(sb) && !readonly) {
2555                 ext4_msg(sb, KERN_ERR,
2556                          "Filesystem with quota feature cannot be mounted RDWR "
2557                          "without CONFIG_QUOTA");
2558                 return 0;
2559         }
2560 #endif  /* CONFIG_QUOTA */
2561         return 1;
2562 }
2563
2564 /*
2565  * This function is called once a day if we have errors logged
2566  * on the file system
2567  */
2568 static void print_daily_error_info(unsigned long arg)
2569 {
2570         struct super_block *sb = (struct super_block *) arg;
2571         struct ext4_sb_info *sbi;
2572         struct ext4_super_block *es;
2573
2574         sbi = EXT4_SB(sb);
2575         es = sbi->s_es;
2576
2577         if (es->s_error_count)
2578                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2579                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2580                          le32_to_cpu(es->s_error_count));
2581         if (es->s_first_error_time) {
2582                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2583                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2584                        (int) sizeof(es->s_first_error_func),
2585                        es->s_first_error_func,
2586                        le32_to_cpu(es->s_first_error_line));
2587                 if (es->s_first_error_ino)
2588                         printk(": inode %u",
2589                                le32_to_cpu(es->s_first_error_ino));
2590                 if (es->s_first_error_block)
2591                         printk(": block %llu", (unsigned long long)
2592                                le64_to_cpu(es->s_first_error_block));
2593                 printk("\n");
2594         }
2595         if (es->s_last_error_time) {
2596                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2597                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2598                        (int) sizeof(es->s_last_error_func),
2599                        es->s_last_error_func,
2600                        le32_to_cpu(es->s_last_error_line));
2601                 if (es->s_last_error_ino)
2602                         printk(": inode %u",
2603                                le32_to_cpu(es->s_last_error_ino));
2604                 if (es->s_last_error_block)
2605                         printk(": block %llu", (unsigned long long)
2606                                le64_to_cpu(es->s_last_error_block));
2607                 printk("\n");
2608         }
2609         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2610 }
2611
2612 /* Find next suitable group and run ext4_init_inode_table */
2613 static int ext4_run_li_request(struct ext4_li_request *elr)
2614 {
2615         struct ext4_group_desc *gdp = NULL;
2616         ext4_group_t group, ngroups;
2617         struct super_block *sb;
2618         unsigned long timeout = 0;
2619         int ret = 0;
2620
2621         sb = elr->lr_super;
2622         ngroups = EXT4_SB(sb)->s_groups_count;
2623
2624         sb_start_write(sb);
2625         for (group = elr->lr_next_group; group < ngroups; group++) {
2626                 gdp = ext4_get_group_desc(sb, group, NULL);
2627                 if (!gdp) {
2628                         ret = 1;
2629                         break;
2630                 }
2631
2632                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2633                         break;
2634         }
2635
2636         if (group >= ngroups)
2637                 ret = 1;
2638
2639         if (!ret) {
2640                 timeout = jiffies;
2641                 ret = ext4_init_inode_table(sb, group,
2642                                             elr->lr_timeout ? 0 : 1);
2643                 if (elr->lr_timeout == 0) {
2644                         timeout = (jiffies - timeout) *
2645                                   elr->lr_sbi->s_li_wait_mult;
2646                         elr->lr_timeout = timeout;
2647                 }
2648                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2649                 elr->lr_next_group = group + 1;
2650         }
2651         sb_end_write(sb);
2652
2653         return ret;
2654 }
2655
2656 /*
2657  * Remove lr_request from the list_request and free the
2658  * request structure. Should be called with li_list_mtx held
2659  */
2660 static void ext4_remove_li_request(struct ext4_li_request *elr)
2661 {
2662         struct ext4_sb_info *sbi;
2663
2664         if (!elr)
2665                 return;
2666
2667         sbi = elr->lr_sbi;
2668
2669         list_del(&elr->lr_request);
2670         sbi->s_li_request = NULL;
2671         kfree(elr);
2672 }
2673
2674 static void ext4_unregister_li_request(struct super_block *sb)
2675 {
2676         mutex_lock(&ext4_li_mtx);
2677         if (!ext4_li_info) {
2678                 mutex_unlock(&ext4_li_mtx);
2679                 return;
2680         }
2681
2682         mutex_lock(&ext4_li_info->li_list_mtx);
2683         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2684         mutex_unlock(&ext4_li_info->li_list_mtx);
2685         mutex_unlock(&ext4_li_mtx);
2686 }
2687
2688 static struct task_struct *ext4_lazyinit_task;
2689
2690 /*
2691  * This is the function where ext4lazyinit thread lives. It walks
2692  * through the request list searching for next scheduled filesystem.
2693  * When such a fs is found, run the lazy initialization request
2694  * (ext4_rn_li_request) and keep track of the time spend in this
2695  * function. Based on that time we compute next schedule time of
2696  * the request. When walking through the list is complete, compute
2697  * next waking time and put itself into sleep.
2698  */
2699 static int ext4_lazyinit_thread(void *arg)
2700 {
2701         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2702         struct list_head *pos, *n;
2703         struct ext4_li_request *elr;
2704         unsigned long next_wakeup, cur;
2705
2706         BUG_ON(NULL == eli);
2707
2708 cont_thread:
2709         while (true) {
2710                 next_wakeup = MAX_JIFFY_OFFSET;
2711
2712                 mutex_lock(&eli->li_list_mtx);
2713                 if (list_empty(&eli->li_request_list)) {
2714                         mutex_unlock(&eli->li_list_mtx);
2715                         goto exit_thread;
2716                 }
2717
2718                 list_for_each_safe(pos, n, &eli->li_request_list) {
2719                         elr = list_entry(pos, struct ext4_li_request,
2720                                          lr_request);
2721
2722                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2723                                 if (ext4_run_li_request(elr) != 0) {
2724                                         /* error, remove the lazy_init job */
2725                                         ext4_remove_li_request(elr);
2726                                         continue;
2727                                 }
2728                         }
2729
2730                         if (time_before(elr->lr_next_sched, next_wakeup))
2731                                 next_wakeup = elr->lr_next_sched;
2732                 }
2733                 mutex_unlock(&eli->li_list_mtx);
2734
2735                 try_to_freeze();
2736
2737                 cur = jiffies;
2738                 if ((time_after_eq(cur, next_wakeup)) ||
2739                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2740                         cond_resched();
2741                         continue;
2742                 }
2743
2744                 schedule_timeout_interruptible(next_wakeup - cur);
2745
2746                 if (kthread_should_stop()) {
2747                         ext4_clear_request_list();
2748                         goto exit_thread;
2749                 }
2750         }
2751
2752 exit_thread:
2753         /*
2754          * It looks like the request list is empty, but we need
2755          * to check it under the li_list_mtx lock, to prevent any
2756          * additions into it, and of course we should lock ext4_li_mtx
2757          * to atomically free the list and ext4_li_info, because at
2758          * this point another ext4 filesystem could be registering
2759          * new one.
2760          */
2761         mutex_lock(&ext4_li_mtx);
2762         mutex_lock(&eli->li_list_mtx);
2763         if (!list_empty(&eli->li_request_list)) {
2764                 mutex_unlock(&eli->li_list_mtx);
2765                 mutex_unlock(&ext4_li_mtx);
2766                 goto cont_thread;
2767         }
2768         mutex_unlock(&eli->li_list_mtx);
2769         kfree(ext4_li_info);
2770         ext4_li_info = NULL;
2771         mutex_unlock(&ext4_li_mtx);
2772
2773         return 0;
2774 }
2775
2776 static void ext4_clear_request_list(void)
2777 {
2778         struct list_head *pos, *n;
2779         struct ext4_li_request *elr;
2780
2781         mutex_lock(&ext4_li_info->li_list_mtx);
2782         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2783                 elr = list_entry(pos, struct ext4_li_request,
2784                                  lr_request);
2785                 ext4_remove_li_request(elr);
2786         }
2787         mutex_unlock(&ext4_li_info->li_list_mtx);
2788 }
2789
2790 static int ext4_run_lazyinit_thread(void)
2791 {
2792         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2793                                          ext4_li_info, "ext4lazyinit");
2794         if (IS_ERR(ext4_lazyinit_task)) {
2795                 int err = PTR_ERR(ext4_lazyinit_task);
2796                 ext4_clear_request_list();
2797                 kfree(ext4_li_info);
2798                 ext4_li_info = NULL;
2799                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2800                                  "initialization thread\n",
2801                                  err);
2802                 return err;
2803         }
2804         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2805         return 0;
2806 }
2807
2808 /*
2809  * Check whether it make sense to run itable init. thread or not.
2810  * If there is at least one uninitialized inode table, return
2811  * corresponding group number, else the loop goes through all
2812  * groups and return total number of groups.
2813  */
2814 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2815 {
2816         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2817         struct ext4_group_desc *gdp = NULL;
2818
2819         for (group = 0; group < ngroups; group++) {
2820                 gdp = ext4_get_group_desc(sb, group, NULL);
2821                 if (!gdp)
2822                         continue;
2823
2824                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2825                         break;
2826         }
2827
2828         return group;
2829 }
2830
2831 static int ext4_li_info_new(void)
2832 {
2833         struct ext4_lazy_init *eli = NULL;
2834
2835         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2836         if (!eli)
2837                 return -ENOMEM;
2838
2839         INIT_LIST_HEAD(&eli->li_request_list);
2840         mutex_init(&eli->li_list_mtx);
2841
2842         eli->li_state |= EXT4_LAZYINIT_QUIT;
2843
2844         ext4_li_info = eli;
2845
2846         return 0;
2847 }
2848
2849 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2850                                             ext4_group_t start)
2851 {
2852         struct ext4_sb_info *sbi = EXT4_SB(sb);
2853         struct ext4_li_request *elr;
2854
2855         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2856         if (!elr)
2857                 return NULL;
2858
2859         elr->lr_super = sb;
2860         elr->lr_sbi = sbi;
2861         elr->lr_next_group = start;
2862
2863         /*
2864          * Randomize first schedule time of the request to
2865          * spread the inode table initialization requests
2866          * better.
2867          */
2868         elr->lr_next_sched = jiffies + (prandom_u32() %
2869                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
2870         return elr;
2871 }
2872
2873 int ext4_register_li_request(struct super_block *sb,
2874                              ext4_group_t first_not_zeroed)
2875 {
2876         struct ext4_sb_info *sbi = EXT4_SB(sb);
2877         struct ext4_li_request *elr = NULL;
2878         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2879         int ret = 0;
2880
2881         mutex_lock(&ext4_li_mtx);
2882         if (sbi->s_li_request != NULL) {
2883                 /*
2884                  * Reset timeout so it can be computed again, because
2885                  * s_li_wait_mult might have changed.
2886                  */
2887                 sbi->s_li_request->lr_timeout = 0;
2888                 goto out;
2889         }
2890
2891         if (first_not_zeroed == ngroups ||
2892             (sb->s_flags & MS_RDONLY) ||
2893             !test_opt(sb, INIT_INODE_TABLE))
2894                 goto out;
2895
2896         elr = ext4_li_request_new(sb, first_not_zeroed);
2897         if (!elr) {
2898                 ret = -ENOMEM;
2899                 goto out;
2900         }
2901
2902         if (NULL == ext4_li_info) {
2903                 ret = ext4_li_info_new();
2904                 if (ret)
2905                         goto out;
2906         }
2907
2908         mutex_lock(&ext4_li_info->li_list_mtx);
2909         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2910         mutex_unlock(&ext4_li_info->li_list_mtx);
2911
2912         sbi->s_li_request = elr;
2913         /*
2914          * set elr to NULL here since it has been inserted to
2915          * the request_list and the removal and free of it is
2916          * handled by ext4_clear_request_list from now on.
2917          */
2918         elr = NULL;
2919
2920         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
2921                 ret = ext4_run_lazyinit_thread();
2922                 if (ret)
2923                         goto out;
2924         }
2925 out:
2926         mutex_unlock(&ext4_li_mtx);
2927         if (ret)
2928                 kfree(elr);
2929         return ret;
2930 }
2931
2932 /*
2933  * We do not need to lock anything since this is called on
2934  * module unload.
2935  */
2936 static void ext4_destroy_lazyinit_thread(void)
2937 {
2938         /*
2939          * If thread exited earlier
2940          * there's nothing to be done.
2941          */
2942         if (!ext4_li_info || !ext4_lazyinit_task)
2943                 return;
2944
2945         kthread_stop(ext4_lazyinit_task);
2946 }
2947
2948 static int set_journal_csum_feature_set(struct super_block *sb)
2949 {
2950         int ret = 1;
2951         int compat, incompat;
2952         struct ext4_sb_info *sbi = EXT4_SB(sb);
2953
2954         if (ext4_has_metadata_csum(sb)) {
2955                 /* journal checksum v3 */
2956                 compat = 0;
2957                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
2958         } else {
2959                 /* journal checksum v1 */
2960                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
2961                 incompat = 0;
2962         }
2963
2964         jbd2_journal_clear_features(sbi->s_journal,
2965                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2966                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
2967                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
2968         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2969                 ret = jbd2_journal_set_features(sbi->s_journal,
2970                                 compat, 0,
2971                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
2972                                 incompat);
2973         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2974                 ret = jbd2_journal_set_features(sbi->s_journal,
2975                                 compat, 0,
2976                                 incompat);
2977                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2978                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2979         } else {
2980                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2981                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2982         }
2983
2984         return ret;
2985 }
2986
2987 /*
2988  * Note: calculating the overhead so we can be compatible with
2989  * historical BSD practice is quite difficult in the face of
2990  * clusters/bigalloc.  This is because multiple metadata blocks from
2991  * different block group can end up in the same allocation cluster.
2992  * Calculating the exact overhead in the face of clustered allocation
2993  * requires either O(all block bitmaps) in memory or O(number of block
2994  * groups**2) in time.  We will still calculate the superblock for
2995  * older file systems --- and if we come across with a bigalloc file
2996  * system with zero in s_overhead_clusters the estimate will be close to
2997  * correct especially for very large cluster sizes --- but for newer
2998  * file systems, it's better to calculate this figure once at mkfs
2999  * time, and store it in the superblock.  If the superblock value is
3000  * present (even for non-bigalloc file systems), we will use it.
3001  */
3002 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3003                           char *buf)
3004 {
3005         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3006         struct ext4_group_desc  *gdp;
3007         ext4_fsblk_t            first_block, last_block, b;
3008         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3009         int                     s, j, count = 0;
3010
3011         if (!ext4_has_feature_bigalloc(sb))
3012                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3013                         sbi->s_itb_per_group + 2);
3014
3015         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3016                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3017         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3018         for (i = 0; i < ngroups; i++) {
3019                 gdp = ext4_get_group_desc(sb, i, NULL);
3020                 b = ext4_block_bitmap(sb, gdp);
3021                 if (b >= first_block && b <= last_block) {
3022                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3023                         count++;
3024                 }
3025                 b = ext4_inode_bitmap(sb, gdp);
3026                 if (b >= first_block && b <= last_block) {
3027                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3028                         count++;
3029                 }
3030                 b = ext4_inode_table(sb, gdp);
3031                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3032                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3033                                 int c = EXT4_B2C(sbi, b - first_block);
3034                                 ext4_set_bit(c, buf);
3035                                 count++;
3036                         }
3037                 if (i != grp)
3038                         continue;
3039                 s = 0;
3040                 if (ext4_bg_has_super(sb, grp)) {
3041                         ext4_set_bit(s++, buf);
3042                         count++;
3043                 }
3044                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3045                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3046                         count++;
3047                 }
3048         }
3049         if (!count)
3050                 return 0;
3051         return EXT4_CLUSTERS_PER_GROUP(sb) -
3052                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3053 }
3054
3055 /*
3056  * Compute the overhead and stash it in sbi->s_overhead
3057  */
3058 int ext4_calculate_overhead(struct super_block *sb)
3059 {
3060         struct ext4_sb_info *sbi = EXT4_SB(sb);
3061         struct ext4_super_block *es = sbi->s_es;
3062         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3063         ext4_fsblk_t overhead = 0;
3064         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3065
3066         if (!buf)
3067                 return -ENOMEM;
3068
3069         /*
3070          * Compute the overhead (FS structures).  This is constant
3071          * for a given filesystem unless the number of block groups
3072          * changes so we cache the previous value until it does.
3073          */
3074
3075         /*
3076          * All of the blocks before first_data_block are overhead
3077          */
3078         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3079
3080         /*
3081          * Add the overhead found in each block group
3082          */
3083         for (i = 0; i < ngroups; i++) {
3084                 int blks;
3085
3086                 blks = count_overhead(sb, i, buf);
3087                 overhead += blks;
3088                 if (blks)
3089                         memset(buf, 0, PAGE_SIZE);
3090                 cond_resched();
3091         }
3092         /* Add the internal journal blocks as well */
3093         if (sbi->s_journal && !sbi->journal_bdev)
3094                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3095
3096         sbi->s_overhead = overhead;
3097         smp_wmb();
3098         free_page((unsigned long) buf);
3099         return 0;
3100 }
3101
3102 static void ext4_set_resv_clusters(struct super_block *sb)
3103 {
3104         ext4_fsblk_t resv_clusters;
3105         struct ext4_sb_info *sbi = EXT4_SB(sb);
3106
3107         /*
3108          * There's no need to reserve anything when we aren't using extents.
3109          * The space estimates are exact, there are no unwritten extents,
3110          * hole punching doesn't need new metadata... This is needed especially
3111          * to keep ext2/3 backward compatibility.
3112          */
3113         if (!ext4_has_feature_extents(sb))
3114                 return;
3115         /*
3116          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3117          * This should cover the situations where we can not afford to run
3118          * out of space like for example punch hole, or converting
3119          * unwritten extents in delalloc path. In most cases such
3120          * allocation would require 1, or 2 blocks, higher numbers are
3121          * very rare.
3122          */
3123         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3124                          sbi->s_cluster_bits);
3125
3126         do_div(resv_clusters, 50);
3127         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3128
3129         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3130 }
3131
3132 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3133 {
3134         char *orig_data = kstrdup(data, GFP_KERNEL);
3135         struct buffer_head *bh;
3136         struct ext4_super_block *es = NULL;
3137         struct ext4_sb_info *sbi;
3138         ext4_fsblk_t block;
3139         ext4_fsblk_t sb_block = get_sb_block(&data);
3140         ext4_fsblk_t logical_sb_block;
3141         unsigned long offset = 0;
3142         unsigned long journal_devnum = 0;
3143         unsigned long def_mount_opts;
3144         struct inode *root;
3145         const char *descr;
3146         int ret = -ENOMEM;
3147         int blocksize, clustersize;
3148         unsigned int db_count;
3149         unsigned int i;
3150         int needs_recovery, has_huge_files, has_bigalloc;
3151         __u64 blocks_count;
3152         int err = 0;
3153         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3154         ext4_group_t first_not_zeroed;
3155
3156         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3157         if (!sbi)
3158                 goto out_free_orig;
3159
3160         sbi->s_blockgroup_lock =
3161                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3162         if (!sbi->s_blockgroup_lock) {
3163                 kfree(sbi);
3164                 goto out_free_orig;
3165         }
3166         sb->s_fs_info = sbi;
3167         sbi->s_sb = sb;
3168         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3169         sbi->s_sb_block = sb_block;
3170         if (sb->s_bdev->bd_part)
3171                 sbi->s_sectors_written_start =
3172                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3173
3174         /* Cleanup superblock name */
3175         strreplace(sb->s_id, '/', '!');
3176
3177         /* -EINVAL is default */
3178         ret = -EINVAL;
3179         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3180         if (!blocksize) {
3181                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3182                 goto out_fail;
3183         }
3184
3185         /*
3186          * The ext4 superblock will not be buffer aligned for other than 1kB
3187          * block sizes.  We need to calculate the offset from buffer start.
3188          */
3189         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3190                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3191                 offset = do_div(logical_sb_block, blocksize);
3192         } else {
3193                 logical_sb_block = sb_block;
3194         }
3195
3196         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3197                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3198                 goto out_fail;
3199         }
3200         /*
3201          * Note: s_es must be initialized as soon as possible because
3202          *       some ext4 macro-instructions depend on its value
3203          */
3204         es = (struct ext4_super_block *) (bh->b_data + offset);
3205         sbi->s_es = es;
3206         sb->s_magic = le16_to_cpu(es->s_magic);
3207         if (sb->s_magic != EXT4_SUPER_MAGIC)
3208                 goto cantfind_ext4;
3209         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3210
3211         /* Warn if metadata_csum and gdt_csum are both set. */
3212         if (ext4_has_feature_metadata_csum(sb) &&
3213             ext4_has_feature_gdt_csum(sb))
3214                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3215                              "redundant flags; please run fsck.");
3216
3217         /* Check for a known checksum algorithm */
3218         if (!ext4_verify_csum_type(sb, es)) {
3219                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3220                          "unknown checksum algorithm.");
3221                 silent = 1;
3222                 goto cantfind_ext4;
3223         }
3224
3225         /* Load the checksum driver */
3226         if (ext4_has_feature_metadata_csum(sb)) {
3227                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3228                 if (IS_ERR(sbi->s_chksum_driver)) {
3229                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3230                         ret = PTR_ERR(sbi->s_chksum_driver);
3231                         sbi->s_chksum_driver = NULL;
3232                         goto failed_mount;
3233                 }
3234         }
3235
3236         /* Check superblock checksum */
3237         if (!ext4_superblock_csum_verify(sb, es)) {
3238                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3239                          "invalid superblock checksum.  Run e2fsck?");
3240                 silent = 1;
3241                 ret = -EFSBADCRC;
3242                 goto cantfind_ext4;
3243         }
3244
3245         /* Precompute checksum seed for all metadata */
3246         if (ext4_has_feature_csum_seed(sb))
3247                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3248         else if (ext4_has_metadata_csum(sb))
3249                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3250                                                sizeof(es->s_uuid));
3251
3252         /* Set defaults before we parse the mount options */
3253         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3254         set_opt(sb, INIT_INODE_TABLE);
3255         if (def_mount_opts & EXT4_DEFM_DEBUG)
3256                 set_opt(sb, DEBUG);
3257         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3258                 set_opt(sb, GRPID);
3259         if (def_mount_opts & EXT4_DEFM_UID16)
3260                 set_opt(sb, NO_UID32);
3261         /* xattr user namespace & acls are now defaulted on */
3262         set_opt(sb, XATTR_USER);
3263 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3264         set_opt(sb, POSIX_ACL);
3265 #endif
3266         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3267         if (ext4_has_metadata_csum(sb))
3268                 set_opt(sb, JOURNAL_CHECKSUM);
3269
3270         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3271                 set_opt(sb, JOURNAL_DATA);
3272         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3273                 set_opt(sb, ORDERED_DATA);
3274         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3275                 set_opt(sb, WRITEBACK_DATA);
3276
3277         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3278                 set_opt(sb, ERRORS_PANIC);
3279         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3280                 set_opt(sb, ERRORS_CONT);
3281         else
3282                 set_opt(sb, ERRORS_RO);
3283         /* block_validity enabled by default; disable with noblock_validity */
3284         set_opt(sb, BLOCK_VALIDITY);
3285         if (def_mount_opts & EXT4_DEFM_DISCARD)
3286                 set_opt(sb, DISCARD);
3287
3288         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3289         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3290         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3291         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3292         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3293
3294         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3295                 set_opt(sb, BARRIER);
3296
3297         /*
3298          * enable delayed allocation by default
3299          * Use -o nodelalloc to turn it off
3300          */
3301         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3302             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3303                 set_opt(sb, DELALLOC);
3304
3305         /*
3306          * set default s_li_wait_mult for lazyinit, for the case there is
3307          * no mount option specified.
3308          */
3309         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3310
3311         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3312                            &journal_devnum, &journal_ioprio, 0)) {
3313                 ext4_msg(sb, KERN_WARNING,
3314                          "failed to parse options in superblock: %s",
3315                          sbi->s_es->s_mount_opts);
3316         }
3317         sbi->s_def_mount_opt = sbi->s_mount_opt;
3318         if (!parse_options((char *) data, sb, &journal_devnum,
3319                            &journal_ioprio, 0))
3320                 goto failed_mount;
3321
3322         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3323                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3324                             "with data=journal disables delayed "
3325                             "allocation and O_DIRECT support!\n");
3326                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3327                         ext4_msg(sb, KERN_ERR, "can't mount with "
3328                                  "both data=journal and delalloc");
3329                         goto failed_mount;
3330                 }
3331                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3332                         ext4_msg(sb, KERN_ERR, "can't mount with "
3333                                  "both data=journal and dioread_nolock");
3334                         goto failed_mount;
3335                 }
3336                 if (test_opt(sb, DAX)) {
3337                         ext4_msg(sb, KERN_ERR, "can't mount with "
3338                                  "both data=journal and dax");
3339                         goto failed_mount;
3340                 }
3341                 if (test_opt(sb, DELALLOC))
3342                         clear_opt(sb, DELALLOC);
3343         } else {
3344                 sb->s_iflags |= SB_I_CGROUPWB;
3345         }
3346
3347         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3348                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3349
3350         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3351             (ext4_has_compat_features(sb) ||
3352              ext4_has_ro_compat_features(sb) ||
3353              ext4_has_incompat_features(sb)))
3354                 ext4_msg(sb, KERN_WARNING,
3355                        "feature flags set on rev 0 fs, "
3356                        "running e2fsck is recommended");
3357
3358         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3359                 set_opt2(sb, HURD_COMPAT);
3360                 if (ext4_has_feature_64bit(sb)) {
3361                         ext4_msg(sb, KERN_ERR,
3362                                  "The Hurd can't support 64-bit file systems");
3363                         goto failed_mount;
3364                 }
3365         }
3366
3367         if (IS_EXT2_SB(sb)) {
3368                 if (ext2_feature_set_ok(sb))
3369                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3370                                  "using the ext4 subsystem");
3371                 else {
3372                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3373                                  "to feature incompatibilities");
3374                         goto failed_mount;
3375                 }
3376         }
3377
3378         if (IS_EXT3_SB(sb)) {
3379                 if (ext3_feature_set_ok(sb))
3380                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3381                                  "using the ext4 subsystem");
3382                 else {
3383                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3384                                  "to feature incompatibilities");
3385                         goto failed_mount;
3386                 }
3387         }
3388
3389         /*
3390          * Check feature flags regardless of the revision level, since we
3391          * previously didn't change the revision level when setting the flags,
3392          * so there is a chance incompat flags are set on a rev 0 filesystem.
3393          */
3394         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3395                 goto failed_mount;
3396
3397         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3398         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3399             blocksize > EXT4_MAX_BLOCK_SIZE) {
3400                 ext4_msg(sb, KERN_ERR,
3401                        "Unsupported filesystem blocksize %d", blocksize);
3402                 goto failed_mount;
3403         }
3404
3405         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3406                 if (blocksize != PAGE_SIZE) {
3407                         ext4_msg(sb, KERN_ERR,
3408                                         "error: unsupported blocksize for dax");
3409                         goto failed_mount;
3410                 }
3411                 if (!sb->s_bdev->bd_disk->fops->direct_access) {
3412                         ext4_msg(sb, KERN_ERR,
3413                                         "error: device does not support dax");
3414                         goto failed_mount;
3415                 }
3416         }
3417
3418         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3419                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3420                          es->s_encryption_level);
3421                 goto failed_mount;
3422         }
3423
3424         if (sb->s_blocksize != blocksize) {
3425                 /* Validate the filesystem blocksize */
3426                 if (!sb_set_blocksize(sb, blocksize)) {
3427                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3428                                         blocksize);
3429                         goto failed_mount;
3430                 }
3431
3432                 brelse(bh);
3433                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3434                 offset = do_div(logical_sb_block, blocksize);
3435                 bh = sb_bread_unmovable(sb, logical_sb_block);
3436                 if (!bh) {
3437                         ext4_msg(sb, KERN_ERR,
3438                                "Can't read superblock on 2nd try");
3439                         goto failed_mount;
3440                 }
3441                 es = (struct ext4_super_block *)(bh->b_data + offset);
3442                 sbi->s_es = es;
3443                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3444                         ext4_msg(sb, KERN_ERR,
3445                                "Magic mismatch, very weird!");
3446                         goto failed_mount;
3447                 }
3448         }
3449
3450         has_huge_files = ext4_has_feature_huge_file(sb);
3451         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3452                                                       has_huge_files);
3453         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3454
3455         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3456                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3457                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3458         } else {
3459                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3460                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3461                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3462                     (!is_power_of_2(sbi->s_inode_size)) ||
3463                     (sbi->s_inode_size > blocksize)) {
3464                         ext4_msg(sb, KERN_ERR,
3465                                "unsupported inode size: %d",
3466                                sbi->s_inode_size);
3467                         goto failed_mount;
3468                 }
3469                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3470                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3471         }
3472
3473         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3474         if (ext4_has_feature_64bit(sb)) {
3475                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3476                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3477                     !is_power_of_2(sbi->s_desc_size)) {
3478                         ext4_msg(sb, KERN_ERR,
3479                                "unsupported descriptor size %lu",
3480                                sbi->s_desc_size);
3481                         goto failed_mount;
3482                 }
3483         } else
3484                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3485
3486         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3487         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3488         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3489                 goto cantfind_ext4;
3490
3491         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3492         if (sbi->s_inodes_per_block == 0)
3493                 goto cantfind_ext4;
3494         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3495                                         sbi->s_inodes_per_block;
3496         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3497         sbi->s_sbh = bh;
3498         sbi->s_mount_state = le16_to_cpu(es->s_state);
3499         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3500         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3501
3502         for (i = 0; i < 4; i++)
3503                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3504         sbi->s_def_hash_version = es->s_def_hash_version;
3505         if (ext4_has_feature_dir_index(sb)) {
3506                 i = le32_to_cpu(es->s_flags);
3507                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3508                         sbi->s_hash_unsigned = 3;
3509                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3510 #ifdef __CHAR_UNSIGNED__
3511                         if (!(sb->s_flags & MS_RDONLY))
3512                                 es->s_flags |=
3513                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3514                         sbi->s_hash_unsigned = 3;
3515 #else
3516                         if (!(sb->s_flags & MS_RDONLY))
3517                                 es->s_flags |=
3518                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3519 #endif
3520                 }
3521         }
3522
3523         /* Handle clustersize */
3524         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3525         has_bigalloc = ext4_has_feature_bigalloc(sb);
3526         if (has_bigalloc) {
3527                 if (clustersize < blocksize) {
3528                         ext4_msg(sb, KERN_ERR,
3529                                  "cluster size (%d) smaller than "
3530                                  "block size (%d)", clustersize, blocksize);
3531                         goto failed_mount;
3532                 }
3533                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3534                         le32_to_cpu(es->s_log_block_size);
3535                 sbi->s_clusters_per_group =
3536                         le32_to_cpu(es->s_clusters_per_group);
3537                 if (sbi->s_clusters_per_group > blocksize * 8) {
3538                         ext4_msg(sb, KERN_ERR,
3539                                  "#clusters per group too big: %lu",
3540                                  sbi->s_clusters_per_group);
3541                         goto failed_mount;
3542                 }
3543                 if (sbi->s_blocks_per_group !=
3544                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3545                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3546                                  "clusters per group (%lu) inconsistent",
3547                                  sbi->s_blocks_per_group,
3548                                  sbi->s_clusters_per_group);
3549                         goto failed_mount;
3550                 }
3551         } else {
3552                 if (clustersize != blocksize) {
3553                         ext4_warning(sb, "fragment/cluster size (%d) != "
3554                                      "block size (%d)", clustersize,
3555                                      blocksize);
3556                         clustersize = blocksize;
3557                 }
3558                 if (sbi->s_blocks_per_group > blocksize * 8) {
3559                         ext4_msg(sb, KERN_ERR,
3560                                  "#blocks per group too big: %lu",
3561                                  sbi->s_blocks_per_group);
3562                         goto failed_mount;
3563                 }
3564                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3565                 sbi->s_cluster_bits = 0;
3566         }
3567         sbi->s_cluster_ratio = clustersize / blocksize;
3568
3569         if (sbi->s_inodes_per_group > blocksize * 8) {
3570                 ext4_msg(sb, KERN_ERR,
3571                        "#inodes per group too big: %lu",
3572                        sbi->s_inodes_per_group);
3573                 goto failed_mount;
3574         }
3575
3576         /* Do we have standard group size of clustersize * 8 blocks ? */
3577         if (sbi->s_blocks_per_group == clustersize << 3)
3578                 set_opt2(sb, STD_GROUP_SIZE);
3579
3580         /*
3581          * Test whether we have more sectors than will fit in sector_t,
3582          * and whether the max offset is addressable by the page cache.
3583          */
3584         err = generic_check_addressable(sb->s_blocksize_bits,
3585                                         ext4_blocks_count(es));
3586         if (err) {
3587                 ext4_msg(sb, KERN_ERR, "filesystem"
3588                          " too large to mount safely on this system");
3589                 if (sizeof(sector_t) < 8)
3590                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3591                 goto failed_mount;
3592         }
3593
3594         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3595                 goto cantfind_ext4;
3596
3597         /* check blocks count against device size */
3598         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3599         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3600                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3601                        "exceeds size of device (%llu blocks)",
3602                        ext4_blocks_count(es), blocks_count);
3603                 goto failed_mount;
3604         }
3605
3606         /*
3607          * It makes no sense for the first data block to be beyond the end
3608          * of the filesystem.
3609          */
3610         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3611                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3612                          "block %u is beyond end of filesystem (%llu)",
3613                          le32_to_cpu(es->s_first_data_block),
3614                          ext4_blocks_count(es));
3615                 goto failed_mount;
3616         }
3617         blocks_count = (ext4_blocks_count(es) -
3618                         le32_to_cpu(es->s_first_data_block) +
3619                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3620         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3621         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3622                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3623                        "(block count %llu, first data block %u, "
3624                        "blocks per group %lu)", sbi->s_groups_count,
3625                        ext4_blocks_count(es),
3626                        le32_to_cpu(es->s_first_data_block),
3627                        EXT4_BLOCKS_PER_GROUP(sb));
3628                 goto failed_mount;
3629         }
3630         sbi->s_groups_count = blocks_count;
3631         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3632                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3633         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3634                    EXT4_DESC_PER_BLOCK(sb);
3635         sbi->s_group_desc = ext4_kvmalloc(db_count *
3636                                           sizeof(struct buffer_head *),
3637                                           GFP_KERNEL);
3638         if (sbi->s_group_desc == NULL) {
3639                 ext4_msg(sb, KERN_ERR, "not enough memory");
3640                 ret = -ENOMEM;
3641                 goto failed_mount;
3642         }
3643
3644         bgl_lock_init(sbi->s_blockgroup_lock);
3645
3646         for (i = 0; i < db_count; i++) {
3647                 block = descriptor_loc(sb, logical_sb_block, i);
3648                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
3649                 if (!sbi->s_group_desc[i]) {
3650                         ext4_msg(sb, KERN_ERR,
3651                                "can't read group descriptor %d", i);
3652                         db_count = i;
3653                         goto failed_mount2;
3654                 }
3655         }
3656         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3657                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3658                 ret = -EFSCORRUPTED;
3659                 goto failed_mount2;
3660         }
3661
3662         sbi->s_gdb_count = db_count;
3663         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3664         spin_lock_init(&sbi->s_next_gen_lock);
3665
3666         setup_timer(&sbi->s_err_report, print_daily_error_info,
3667                 (unsigned long) sb);
3668
3669         /* Register extent status tree shrinker */
3670         if (ext4_es_register_shrinker(sbi))
3671                 goto failed_mount3;
3672
3673         sbi->s_stripe = ext4_get_stripe_size(sbi);
3674         sbi->s_extent_max_zeroout_kb = 32;
3675
3676         /*
3677          * set up enough so that it can read an inode
3678          */
3679         sb->s_op = &ext4_sops;
3680         sb->s_export_op = &ext4_export_ops;
3681         sb->s_xattr = ext4_xattr_handlers;
3682 #ifdef CONFIG_QUOTA
3683         sb->dq_op = &ext4_quota_operations;
3684         if (ext4_has_feature_quota(sb))
3685                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
3686         else
3687                 sb->s_qcop = &ext4_qctl_operations;
3688         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
3689 #endif
3690         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3691
3692         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3693         mutex_init(&sbi->s_orphan_lock);
3694
3695         sb->s_root = NULL;
3696
3697         needs_recovery = (es->s_last_orphan != 0 ||
3698                           ext4_has_feature_journal_needs_recovery(sb));
3699
3700         if (ext4_has_feature_mmp(sb) && !(sb->s_flags & MS_RDONLY))
3701                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3702                         goto failed_mount3a;
3703
3704         /*
3705          * The first inode we look at is the journal inode.  Don't try
3706          * root first: it may be modified in the journal!
3707          */
3708         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
3709                 if (ext4_load_journal(sb, es, journal_devnum))
3710                         goto failed_mount3a;
3711         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3712                    ext4_has_feature_journal_needs_recovery(sb)) {
3713                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3714                        "suppressed and not mounted read-only");
3715                 goto failed_mount_wq;
3716         } else {
3717                 /* Nojournal mode, all journal mount options are illegal */
3718                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
3719                         ext4_msg(sb, KERN_ERR, "can't mount with "
3720                                  "journal_checksum, fs mounted w/o journal");
3721                         goto failed_mount_wq;
3722                 }
3723                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3724                         ext4_msg(sb, KERN_ERR, "can't mount with "
3725                                  "journal_async_commit, fs mounted w/o journal");
3726                         goto failed_mount_wq;
3727                 }
3728                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
3729                         ext4_msg(sb, KERN_ERR, "can't mount with "
3730                                  "commit=%lu, fs mounted w/o journal",
3731                                  sbi->s_commit_interval / HZ);
3732                         goto failed_mount_wq;
3733                 }
3734                 if (EXT4_MOUNT_DATA_FLAGS &
3735                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
3736                         ext4_msg(sb, KERN_ERR, "can't mount with "
3737                                  "data=, fs mounted w/o journal");
3738                         goto failed_mount_wq;
3739                 }
3740                 sbi->s_def_mount_opt &= EXT4_MOUNT_JOURNAL_CHECKSUM;
3741                 clear_opt(sb, JOURNAL_CHECKSUM);
3742                 clear_opt(sb, DATA_FLAGS);
3743                 sbi->s_journal = NULL;
3744                 needs_recovery = 0;
3745                 goto no_journal;
3746         }
3747
3748         if (ext4_has_feature_64bit(sb) &&
3749             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3750                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3751                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3752                 goto failed_mount_wq;
3753         }
3754
3755         if (!set_journal_csum_feature_set(sb)) {
3756                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3757                          "feature set");
3758                 goto failed_mount_wq;
3759         }
3760
3761         /* We have now updated the journal if required, so we can
3762          * validate the data journaling mode. */
3763         switch (test_opt(sb, DATA_FLAGS)) {
3764         case 0:
3765                 /* No mode set, assume a default based on the journal
3766                  * capabilities: ORDERED_DATA if the journal can
3767                  * cope, else JOURNAL_DATA
3768                  */
3769                 if (jbd2_journal_check_available_features
3770                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3771                         set_opt(sb, ORDERED_DATA);
3772                 else
3773                         set_opt(sb, JOURNAL_DATA);
3774                 break;
3775
3776         case EXT4_MOUNT_ORDERED_DATA:
3777         case EXT4_MOUNT_WRITEBACK_DATA:
3778                 if (!jbd2_journal_check_available_features
3779                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3780                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3781                                "requested data journaling mode");
3782                         goto failed_mount_wq;
3783                 }
3784         default:
3785                 break;
3786         }
3787         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3788
3789         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3790
3791 no_journal:
3792         if (ext4_mballoc_ready) {
3793                 sbi->s_mb_cache = ext4_xattr_create_cache(sb->s_id);
3794                 if (!sbi->s_mb_cache) {
3795                         ext4_msg(sb, KERN_ERR, "Failed to create an mb_cache");
3796                         goto failed_mount_wq;
3797                 }
3798         }
3799
3800         if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
3801             (blocksize != PAGE_CACHE_SIZE)) {
3802                 ext4_msg(sb, KERN_ERR,
3803                          "Unsupported blocksize for fs encryption");
3804                 goto failed_mount_wq;
3805         }
3806
3807         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !(sb->s_flags & MS_RDONLY) &&
3808             !ext4_has_feature_encrypt(sb)) {
3809                 ext4_set_feature_encrypt(sb);
3810                 ext4_commit_super(sb, 1);
3811         }
3812
3813         /*
3814          * Get the # of file system overhead blocks from the
3815          * superblock if present.
3816          */
3817         if (es->s_overhead_clusters)
3818                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3819         else {
3820                 err = ext4_calculate_overhead(sb);
3821                 if (err)
3822                         goto failed_mount_wq;
3823         }
3824
3825         /*
3826          * The maximum number of concurrent works can be high and
3827          * concurrency isn't really necessary.  Limit it to 1.
3828          */
3829         EXT4_SB(sb)->rsv_conversion_wq =
3830                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3831         if (!EXT4_SB(sb)->rsv_conversion_wq) {
3832                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
3833                 ret = -ENOMEM;
3834                 goto failed_mount4;
3835         }
3836
3837         /*
3838          * The jbd2_journal_load will have done any necessary log recovery,
3839          * so we can safely mount the rest of the filesystem now.
3840          */
3841
3842         root = ext4_iget(sb, EXT4_ROOT_INO);
3843         if (IS_ERR(root)) {
3844                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3845                 ret = PTR_ERR(root);
3846                 root = NULL;
3847                 goto failed_mount4;
3848         }
3849         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3850                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3851                 iput(root);
3852                 goto failed_mount4;
3853         }
3854         sb->s_root = d_make_root(root);
3855         if (!sb->s_root) {
3856                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3857                 ret = -ENOMEM;
3858                 goto failed_mount4;
3859         }
3860
3861         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3862                 sb->s_flags |= MS_RDONLY;
3863
3864         /* determine the minimum size of new large inodes, if present */
3865         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3866                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3867                                                      EXT4_GOOD_OLD_INODE_SIZE;
3868                 if (ext4_has_feature_extra_isize(sb)) {
3869                         if (sbi->s_want_extra_isize <
3870                             le16_to_cpu(es->s_want_extra_isize))
3871                                 sbi->s_want_extra_isize =
3872                                         le16_to_cpu(es->s_want_extra_isize);
3873                         if (sbi->s_want_extra_isize <
3874                             le16_to_cpu(es->s_min_extra_isize))
3875                                 sbi->s_want_extra_isize =
3876                                         le16_to_cpu(es->s_min_extra_isize);
3877                 }
3878         }
3879         /* Check if enough inode space is available */
3880         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3881                                                         sbi->s_inode_size) {
3882                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3883                                                        EXT4_GOOD_OLD_INODE_SIZE;
3884                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3885                          "available");
3886         }
3887
3888         ext4_set_resv_clusters(sb);
3889
3890         err = ext4_setup_system_zone(sb);
3891         if (err) {
3892                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3893                          "zone (%d)", err);
3894                 goto failed_mount4a;
3895         }
3896
3897         ext4_ext_init(sb);
3898         err = ext4_mb_init(sb);
3899         if (err) {
3900                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3901                          err);
3902                 goto failed_mount5;
3903         }
3904
3905         block = ext4_count_free_clusters(sb);
3906         ext4_free_blocks_count_set(sbi->s_es,
3907                                    EXT4_C2B(sbi, block));
3908         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
3909                                   GFP_KERNEL);
3910         if (!err) {
3911                 unsigned long freei = ext4_count_free_inodes(sb);
3912                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
3913                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
3914                                           GFP_KERNEL);
3915         }
3916         if (!err)
3917                 err = percpu_counter_init(&sbi->s_dirs_counter,
3918                                           ext4_count_dirs(sb), GFP_KERNEL);
3919         if (!err)
3920                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
3921                                           GFP_KERNEL);
3922         if (err) {
3923                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3924                 goto failed_mount6;
3925         }
3926
3927         if (ext4_has_feature_flex_bg(sb))
3928                 if (!ext4_fill_flex_info(sb)) {
3929                         ext4_msg(sb, KERN_ERR,
3930                                "unable to initialize "
3931                                "flex_bg meta info!");
3932                         goto failed_mount6;
3933                 }
3934
3935         err = ext4_register_li_request(sb, first_not_zeroed);
3936         if (err)
3937                 goto failed_mount6;
3938
3939         err = ext4_register_sysfs(sb);
3940         if (err)
3941                 goto failed_mount7;
3942
3943 #ifdef CONFIG_QUOTA
3944         /* Enable quota usage during mount. */
3945         if (ext4_has_feature_quota(sb) && !(sb->s_flags & MS_RDONLY)) {
3946                 err = ext4_enable_quotas(sb);
3947                 if (err)
3948                         goto failed_mount8;
3949         }
3950 #endif  /* CONFIG_QUOTA */
3951
3952         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3953         ext4_orphan_cleanup(sb, es);
3954         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3955         if (needs_recovery) {
3956                 ext4_msg(sb, KERN_INFO, "recovery complete");
3957                 ext4_mark_recovery_complete(sb, es);
3958         }
3959         if (EXT4_SB(sb)->s_journal) {
3960                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3961                         descr = " journalled data mode";
3962                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3963                         descr = " ordered data mode";
3964                 else
3965                         descr = " writeback data mode";
3966         } else
3967                 descr = "out journal";
3968
3969         if (test_opt(sb, DISCARD)) {
3970                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
3971                 if (!blk_queue_discard(q))
3972                         ext4_msg(sb, KERN_WARNING,
3973                                  "mounting with \"discard\" option, but "
3974                                  "the device does not support discard");
3975         }
3976
3977         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
3978                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3979                          "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3980                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3981
3982         if (es->s_error_count)
3983                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3984
3985         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
3986         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
3987         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
3988         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
3989
3990         kfree(orig_data);
3991         return 0;
3992
3993 cantfind_ext4:
3994         if (!silent)
3995                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3996         goto failed_mount;
3997
3998 #ifdef CONFIG_QUOTA
3999 failed_mount8:
4000         ext4_unregister_sysfs(sb);
4001 #endif
4002 failed_mount7:
4003         ext4_unregister_li_request(sb);
4004 failed_mount6:
4005         ext4_mb_release(sb);
4006         if (sbi->s_flex_groups)
4007                 kvfree(sbi->s_flex_groups);
4008         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4009         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4010         percpu_counter_destroy(&sbi->s_dirs_counter);
4011         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4012 failed_mount5:
4013         ext4_ext_release(sb);
4014         ext4_release_system_zone(sb);
4015 failed_mount4a:
4016         dput(sb->s_root);
4017         sb->s_root = NULL;
4018 failed_mount4:
4019         ext4_msg(sb, KERN_ERR, "mount failed");
4020         if (EXT4_SB(sb)->rsv_conversion_wq)
4021                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4022 failed_mount_wq:
4023         if (sbi->s_journal) {
4024                 jbd2_journal_destroy(sbi->s_journal);
4025                 sbi->s_journal = NULL;
4026         }
4027 failed_mount3a:
4028         ext4_es_unregister_shrinker(sbi);
4029 failed_mount3:
4030         del_timer_sync(&sbi->s_err_report);
4031         if (sbi->s_mmp_tsk)
4032                 kthread_stop(sbi->s_mmp_tsk);
4033 failed_mount2:
4034         for (i = 0; i < db_count; i++)
4035                 brelse(sbi->s_group_desc[i]);
4036         kvfree(sbi->s_group_desc);
4037 failed_mount:
4038         if (sbi->s_chksum_driver)
4039                 crypto_free_shash(sbi->s_chksum_driver);
4040 #ifdef CONFIG_QUOTA
4041         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4042                 kfree(sbi->s_qf_names[i]);
4043 #endif
4044         ext4_blkdev_remove(sbi);
4045         brelse(bh);
4046 out_fail:
4047         sb->s_fs_info = NULL;
4048         kfree(sbi->s_blockgroup_lock);
4049         kfree(sbi);
4050 out_free_orig:
4051         kfree(orig_data);
4052         return err ? err : ret;
4053 }
4054
4055 /*
4056  * Setup any per-fs journal parameters now.  We'll do this both on
4057  * initial mount, once the journal has been initialised but before we've
4058  * done any recovery; and again on any subsequent remount.
4059  */
4060 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4061 {
4062         struct ext4_sb_info *sbi = EXT4_SB(sb);
4063
4064         journal->j_commit_interval = sbi->s_commit_interval;
4065         journal->j_min_batch_time = sbi->s_min_batch_time;
4066         journal->j_max_batch_time = sbi->s_max_batch_time;
4067
4068         write_lock(&journal->j_state_lock);
4069         if (test_opt(sb, BARRIER))
4070                 journal->j_flags |= JBD2_BARRIER;
4071         else
4072                 journal->j_flags &= ~JBD2_BARRIER;
4073         if (test_opt(sb, DATA_ERR_ABORT))
4074                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4075         else
4076                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4077         write_unlock(&journal->j_state_lock);
4078 }
4079
4080 static journal_t *ext4_get_journal(struct super_block *sb,
4081                                    unsigned int journal_inum)
4082 {
4083         struct inode *journal_inode;
4084         journal_t *journal;
4085
4086         BUG_ON(!ext4_has_feature_journal(sb));
4087
4088         /* First, test for the existence of a valid inode on disk.  Bad
4089          * things happen if we iget() an unused inode, as the subsequent
4090          * iput() will try to delete it. */
4091
4092         journal_inode = ext4_iget(sb, journal_inum);
4093         if (IS_ERR(journal_inode)) {
4094                 ext4_msg(sb, KERN_ERR, "no journal found");
4095                 return NULL;
4096         }
4097         if (!journal_inode->i_nlink) {
4098                 make_bad_inode(journal_inode);
4099                 iput(journal_inode);
4100                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4101                 return NULL;
4102         }
4103
4104         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4105                   journal_inode, journal_inode->i_size);
4106         if (!S_ISREG(journal_inode->i_mode)) {
4107                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4108                 iput(journal_inode);
4109                 return NULL;
4110         }
4111
4112         journal = jbd2_journal_init_inode(journal_inode);
4113         if (!journal) {
4114                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4115                 iput(journal_inode);
4116                 return NULL;
4117         }
4118         journal->j_private = sb;
4119         ext4_init_journal_params(sb, journal);
4120         return journal;
4121 }
4122
4123 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4124                                        dev_t j_dev)
4125 {
4126         struct buffer_head *bh;
4127         journal_t *journal;
4128         ext4_fsblk_t start;
4129         ext4_fsblk_t len;
4130         int hblock, blocksize;
4131         ext4_fsblk_t sb_block;
4132         unsigned long offset;
4133         struct ext4_super_block *es;
4134         struct block_device *bdev;
4135
4136         BUG_ON(!ext4_has_feature_journal(sb));
4137
4138         bdev = ext4_blkdev_get(j_dev, sb);
4139         if (bdev == NULL)
4140                 return NULL;
4141
4142         blocksize = sb->s_blocksize;
4143         hblock = bdev_logical_block_size(bdev);
4144         if (blocksize < hblock) {
4145                 ext4_msg(sb, KERN_ERR,
4146                         "blocksize too small for journal device");
4147                 goto out_bdev;
4148         }
4149
4150         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4151         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4152         set_blocksize(bdev, blocksize);
4153         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4154                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4155                        "external journal");
4156                 goto out_bdev;
4157         }
4158
4159         es = (struct ext4_super_block *) (bh->b_data + offset);
4160         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4161             !(le32_to_cpu(es->s_feature_incompat) &
4162               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4163                 ext4_msg(sb, KERN_ERR, "external journal has "
4164                                         "bad superblock");
4165                 brelse(bh);
4166                 goto out_bdev;
4167         }
4168
4169         if ((le32_to_cpu(es->s_feature_ro_compat) &
4170              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4171             es->s_checksum != ext4_superblock_csum(sb, es)) {
4172                 ext4_msg(sb, KERN_ERR, "external journal has "
4173                                        "corrupt superblock");
4174                 brelse(bh);
4175                 goto out_bdev;
4176         }
4177
4178         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4179                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4180                 brelse(bh);
4181                 goto out_bdev;
4182         }
4183
4184         len = ext4_blocks_count(es);
4185         start = sb_block + 1;
4186         brelse(bh);     /* we're done with the superblock */
4187
4188         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4189                                         start, len, blocksize);
4190         if (!journal) {
4191                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4192                 goto out_bdev;
4193         }
4194         journal->j_private = sb;
4195         ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4196         wait_on_buffer(journal->j_sb_buffer);
4197         if (!buffer_uptodate(journal->j_sb_buffer)) {
4198                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4199                 goto out_journal;
4200         }
4201         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4202                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4203                                         "user (unsupported) - %d",
4204                         be32_to_cpu(journal->j_superblock->s_nr_users));
4205                 goto out_journal;
4206         }
4207         EXT4_SB(sb)->journal_bdev = bdev;
4208         ext4_init_journal_params(sb, journal);
4209         return journal;
4210
4211 out_journal:
4212         jbd2_journal_destroy(journal);
4213 out_bdev:
4214         ext4_blkdev_put(bdev);
4215         return NULL;
4216 }
4217
4218 static int ext4_load_journal(struct super_block *sb,
4219                              struct ext4_super_block *es,
4220                              unsigned long journal_devnum)
4221 {
4222         journal_t *journal;
4223         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4224         dev_t journal_dev;
4225         int err = 0;
4226         int really_read_only;
4227
4228         BUG_ON(!ext4_has_feature_journal(sb));
4229
4230         if (journal_devnum &&
4231             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4232                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4233                         "numbers have changed");
4234                 journal_dev = new_decode_dev(journal_devnum);
4235         } else
4236                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4237
4238         really_read_only = bdev_read_only(sb->s_bdev);
4239
4240         /*
4241          * Are we loading a blank journal or performing recovery after a
4242          * crash?  For recovery, we need to check in advance whether we
4243          * can get read-write access to the device.
4244          */
4245         if (ext4_has_feature_journal_needs_recovery(sb)) {
4246                 if (sb->s_flags & MS_RDONLY) {
4247                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4248                                         "required on readonly filesystem");
4249                         if (really_read_only) {
4250                                 ext4_msg(sb, KERN_ERR, "write access "
4251                                         "unavailable, cannot proceed");
4252                                 return -EROFS;
4253                         }
4254                         ext4_msg(sb, KERN_INFO, "write access will "
4255                                "be enabled during recovery");
4256                 }
4257         }
4258
4259         if (journal_inum && journal_dev) {
4260                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4261                        "and inode journals!");
4262                 return -EINVAL;
4263         }
4264
4265         if (journal_inum) {
4266                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4267                         return -EINVAL;
4268         } else {
4269                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4270                         return -EINVAL;
4271         }
4272
4273         if (!(journal->j_flags & JBD2_BARRIER))
4274                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4275
4276         if (!ext4_has_feature_journal_needs_recovery(sb))
4277                 err = jbd2_journal_wipe(journal, !really_read_only);
4278         if (!err) {
4279                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4280                 if (save)
4281                         memcpy(save, ((char *) es) +
4282                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4283                 err = jbd2_journal_load(journal);
4284                 if (save)
4285                         memcpy(((char *) es) + EXT4_S_ERR_START,
4286                                save, EXT4_S_ERR_LEN);
4287                 kfree(save);
4288         }
4289
4290         if (err) {
4291                 ext4_msg(sb, KERN_ERR, "error loading journal");
4292                 jbd2_journal_destroy(journal);
4293                 return err;
4294         }
4295
4296         EXT4_SB(sb)->s_journal = journal;
4297         ext4_clear_journal_err(sb, es);
4298
4299         if (!really_read_only && journal_devnum &&
4300             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4301                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4302
4303                 /* Make sure we flush the recovery flag to disk. */
4304                 ext4_commit_super(sb, 1);
4305         }
4306
4307         return 0;
4308 }
4309
4310 static int ext4_commit_super(struct super_block *sb, int sync)
4311 {
4312         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4313         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4314         int error = 0;
4315
4316         if (!sbh || block_device_ejected(sb))
4317                 return error;
4318         if (buffer_write_io_error(sbh)) {
4319                 /*
4320                  * Oh, dear.  A previous attempt to write the
4321                  * superblock failed.  This could happen because the
4322                  * USB device was yanked out.  Or it could happen to
4323                  * be a transient write error and maybe the block will
4324                  * be remapped.  Nothing we can do but to retry the
4325                  * write and hope for the best.
4326                  */
4327                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4328                        "superblock detected");
4329                 clear_buffer_write_io_error(sbh);
4330                 set_buffer_uptodate(sbh);
4331         }
4332         /*
4333          * If the file system is mounted read-only, don't update the
4334          * superblock write time.  This avoids updating the superblock
4335          * write time when we are mounting the root file system
4336          * read/only but we need to replay the journal; at that point,
4337          * for people who are east of GMT and who make their clock
4338          * tick in localtime for Windows bug-for-bug compatibility,
4339          * the clock is set in the future, and this will cause e2fsck
4340          * to complain and force a full file system check.
4341          */
4342         if (!(sb->s_flags & MS_RDONLY))
4343                 es->s_wtime = cpu_to_le32(get_seconds());
4344         if (sb->s_bdev->bd_part)
4345                 es->s_kbytes_written =
4346                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4347                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4348                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4349         else
4350                 es->s_kbytes_written =
4351                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4352         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4353                 ext4_free_blocks_count_set(es,
4354                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4355                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4356         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4357                 es->s_free_inodes_count =
4358                         cpu_to_le32(percpu_counter_sum_positive(
4359                                 &EXT4_SB(sb)->s_freeinodes_counter));
4360         BUFFER_TRACE(sbh, "marking dirty");
4361         ext4_superblock_csum_set(sb);
4362         mark_buffer_dirty(sbh);
4363         if (sync) {
4364                 error = __sync_dirty_buffer(sbh,
4365                         test_opt(sb, BARRIER) ? WRITE_FUA : WRITE_SYNC);
4366                 if (error)
4367                         return error;
4368
4369                 error = buffer_write_io_error(sbh);
4370                 if (error) {
4371                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4372                                "superblock");
4373                         clear_buffer_write_io_error(sbh);
4374                         set_buffer_uptodate(sbh);
4375                 }
4376         }
4377         return error;
4378 }
4379
4380 /*
4381  * Have we just finished recovery?  If so, and if we are mounting (or
4382  * remounting) the filesystem readonly, then we will end up with a
4383  * consistent fs on disk.  Record that fact.
4384  */
4385 static void ext4_mark_recovery_complete(struct super_block *sb,
4386                                         struct ext4_super_block *es)
4387 {
4388         journal_t *journal = EXT4_SB(sb)->s_journal;
4389
4390         if (!ext4_has_feature_journal(sb)) {
4391                 BUG_ON(journal != NULL);
4392                 return;
4393         }
4394         jbd2_journal_lock_updates(journal);
4395         if (jbd2_journal_flush(journal) < 0)
4396                 goto out;
4397
4398         if (ext4_has_feature_journal_needs_recovery(sb) &&
4399             sb->s_flags & MS_RDONLY) {
4400                 ext4_clear_feature_journal_needs_recovery(sb);
4401                 ext4_commit_super(sb, 1);
4402         }
4403
4404 out:
4405         jbd2_journal_unlock_updates(journal);
4406 }
4407
4408 /*
4409  * If we are mounting (or read-write remounting) a filesystem whose journal
4410  * has recorded an error from a previous lifetime, move that error to the
4411  * main filesystem now.
4412  */
4413 static void ext4_clear_journal_err(struct super_block *sb,
4414                                    struct ext4_super_block *es)
4415 {
4416         journal_t *journal;
4417         int j_errno;
4418         const char *errstr;
4419
4420         BUG_ON(!ext4_has_feature_journal(sb));
4421
4422         journal = EXT4_SB(sb)->s_journal;
4423
4424         /*
4425          * Now check for any error status which may have been recorded in the
4426          * journal by a prior ext4_error() or ext4_abort()
4427          */
4428
4429         j_errno = jbd2_journal_errno(journal);
4430         if (j_errno) {
4431                 char nbuf[16];
4432
4433                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4434                 ext4_warning(sb, "Filesystem error recorded "
4435                              "from previous mount: %s", errstr);
4436                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4437
4438                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4439                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4440                 ext4_commit_super(sb, 1);
4441
4442                 jbd2_journal_clear_err(journal);
4443                 jbd2_journal_update_sb_errno(journal);
4444         }
4445 }
4446
4447 /*
4448  * Force the running and committing transactions to commit,
4449  * and wait on the commit.
4450  */
4451 int ext4_force_commit(struct super_block *sb)
4452 {
4453         journal_t *journal;
4454
4455         if (sb->s_flags & MS_RDONLY)
4456                 return 0;
4457
4458         journal = EXT4_SB(sb)->s_journal;
4459         return ext4_journal_force_commit(journal);
4460 }
4461
4462 static int ext4_sync_fs(struct super_block *sb, int wait)
4463 {
4464         int ret = 0;
4465         tid_t target;
4466         bool needs_barrier = false;
4467         struct ext4_sb_info *sbi = EXT4_SB(sb);
4468
4469         trace_ext4_sync_fs(sb, wait);
4470         flush_workqueue(sbi->rsv_conversion_wq);
4471         /*
4472          * Writeback quota in non-journalled quota case - journalled quota has
4473          * no dirty dquots
4474          */
4475         dquot_writeback_dquots(sb, -1);
4476         /*
4477          * Data writeback is possible w/o journal transaction, so barrier must
4478          * being sent at the end of the function. But we can skip it if
4479          * transaction_commit will do it for us.
4480          */
4481         if (sbi->s_journal) {
4482                 target = jbd2_get_latest_transaction(sbi->s_journal);
4483                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4484                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4485                         needs_barrier = true;
4486
4487                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4488                         if (wait)
4489                                 ret = jbd2_log_wait_commit(sbi->s_journal,
4490                                                            target);
4491                 }
4492         } else if (wait && test_opt(sb, BARRIER))
4493                 needs_barrier = true;
4494         if (needs_barrier) {
4495                 int err;
4496                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4497                 if (!ret)
4498                         ret = err;
4499         }
4500
4501         return ret;
4502 }
4503
4504 /*
4505  * LVM calls this function before a (read-only) snapshot is created.  This
4506  * gives us a chance to flush the journal completely and mark the fs clean.
4507  *
4508  * Note that only this function cannot bring a filesystem to be in a clean
4509  * state independently. It relies on upper layer to stop all data & metadata
4510  * modifications.
4511  */
4512 static int ext4_freeze(struct super_block *sb)
4513 {
4514         int error = 0;
4515         journal_t *journal;
4516
4517         if (sb->s_flags & MS_RDONLY)
4518                 return 0;
4519
4520         journal = EXT4_SB(sb)->s_journal;
4521
4522         if (journal) {
4523                 /* Now we set up the journal barrier. */
4524                 jbd2_journal_lock_updates(journal);
4525
4526                 /*
4527                  * Don't clear the needs_recovery flag if we failed to
4528                  * flush the journal.
4529                  */
4530                 error = jbd2_journal_flush(journal);
4531                 if (error < 0)
4532                         goto out;
4533
4534                 /* Journal blocked and flushed, clear needs_recovery flag. */
4535                 ext4_clear_feature_journal_needs_recovery(sb);
4536         }
4537
4538         error = ext4_commit_super(sb, 1);
4539 out:
4540         if (journal)
4541                 /* we rely on upper layer to stop further updates */
4542                 jbd2_journal_unlock_updates(journal);
4543         return error;
4544 }
4545
4546 /*
4547  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4548  * flag here, even though the filesystem is not technically dirty yet.
4549  */
4550 static int ext4_unfreeze(struct super_block *sb)
4551 {
4552         if (sb->s_flags & MS_RDONLY)
4553                 return 0;
4554
4555         if (EXT4_SB(sb)->s_journal) {
4556                 /* Reset the needs_recovery flag before the fs is unlocked. */
4557                 ext4_set_feature_journal_needs_recovery(sb);
4558         }
4559
4560         ext4_commit_super(sb, 1);
4561         return 0;
4562 }
4563
4564 /*
4565  * Structure to save mount options for ext4_remount's benefit
4566  */
4567 struct ext4_mount_options {
4568         unsigned long s_mount_opt;
4569         unsigned long s_mount_opt2;
4570         kuid_t s_resuid;
4571         kgid_t s_resgid;
4572         unsigned long s_commit_interval;
4573         u32 s_min_batch_time, s_max_batch_time;
4574 #ifdef CONFIG_QUOTA
4575         int s_jquota_fmt;
4576         char *s_qf_names[EXT4_MAXQUOTAS];
4577 #endif
4578 };
4579
4580 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4581 {
4582         struct ext4_super_block *es;
4583         struct ext4_sb_info *sbi = EXT4_SB(sb);
4584         unsigned long old_sb_flags;
4585         struct ext4_mount_options old_opts;
4586         int enable_quota = 0;
4587         ext4_group_t g;
4588         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4589         int err = 0;
4590 #ifdef CONFIG_QUOTA
4591         int i, j;
4592 #endif
4593         char *orig_data = kstrdup(data, GFP_KERNEL);
4594
4595         /* Store the original options */
4596         old_sb_flags = sb->s_flags;
4597         old_opts.s_mount_opt = sbi->s_mount_opt;
4598         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4599         old_opts.s_resuid = sbi->s_resuid;
4600         old_opts.s_resgid = sbi->s_resgid;
4601         old_opts.s_commit_interval = sbi->s_commit_interval;
4602         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4603         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4604 #ifdef CONFIG_QUOTA
4605         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4606         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4607                 if (sbi->s_qf_names[i]) {
4608                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4609                                                          GFP_KERNEL);
4610                         if (!old_opts.s_qf_names[i]) {
4611                                 for (j = 0; j < i; j++)
4612                                         kfree(old_opts.s_qf_names[j]);
4613                                 kfree(orig_data);
4614                                 return -ENOMEM;
4615                         }
4616                 } else
4617                         old_opts.s_qf_names[i] = NULL;
4618 #endif
4619         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4620                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4621
4622         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4623                 err = -EINVAL;
4624                 goto restore_opts;
4625         }
4626
4627         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
4628             test_opt(sb, JOURNAL_CHECKSUM)) {
4629                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
4630                          "during remount not supported; ignoring");
4631                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
4632         }
4633
4634         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4635                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4636                         ext4_msg(sb, KERN_ERR, "can't mount with "
4637                                  "both data=journal and delalloc");
4638                         err = -EINVAL;
4639                         goto restore_opts;
4640                 }
4641                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4642                         ext4_msg(sb, KERN_ERR, "can't mount with "
4643                                  "both data=journal and dioread_nolock");
4644                         err = -EINVAL;
4645                         goto restore_opts;
4646                 }
4647                 if (test_opt(sb, DAX)) {
4648                         ext4_msg(sb, KERN_ERR, "can't mount with "
4649                                  "both data=journal and dax");
4650                         err = -EINVAL;
4651                         goto restore_opts;
4652                 }
4653         }
4654
4655         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
4656                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
4657                         "dax flag with busy inodes while remounting");
4658                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
4659         }
4660
4661         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4662                 ext4_abort(sb, "Abort forced by user");
4663
4664         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4665                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4666
4667         es = sbi->s_es;
4668
4669         if (sbi->s_journal) {
4670                 ext4_init_journal_params(sb, sbi->s_journal);
4671                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4672         }
4673
4674         if (*flags & MS_LAZYTIME)
4675                 sb->s_flags |= MS_LAZYTIME;
4676
4677         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4678                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4679                         err = -EROFS;
4680                         goto restore_opts;
4681                 }
4682
4683                 if (*flags & MS_RDONLY) {
4684                         err = sync_filesystem(sb);
4685                         if (err < 0)
4686                                 goto restore_opts;
4687                         err = dquot_suspend(sb, -1);
4688                         if (err < 0)
4689                                 goto restore_opts;
4690
4691                         /*
4692                          * First of all, the unconditional stuff we have to do
4693                          * to disable replay of the journal when we next remount
4694                          */
4695                         sb->s_flags |= MS_RDONLY;
4696
4697                         /*
4698                          * OK, test if we are remounting a valid rw partition
4699                          * readonly, and if so set the rdonly flag and then
4700                          * mark the partition as valid again.
4701                          */
4702                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4703                             (sbi->s_mount_state & EXT4_VALID_FS))
4704                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4705
4706                         if (sbi->s_journal)
4707                                 ext4_mark_recovery_complete(sb, es);
4708                 } else {
4709                         /* Make sure we can mount this feature set readwrite */
4710                         if (ext4_has_feature_readonly(sb) ||
4711                             !ext4_feature_set_ok(sb, 0)) {
4712                                 err = -EROFS;
4713                                 goto restore_opts;
4714                         }
4715                         /*
4716                          * Make sure the group descriptor checksums
4717                          * are sane.  If they aren't, refuse to remount r/w.
4718                          */
4719                         for (g = 0; g < sbi->s_groups_count; g++) {
4720                                 struct ext4_group_desc *gdp =
4721                                         ext4_get_group_desc(sb, g, NULL);
4722
4723                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4724                                         ext4_msg(sb, KERN_ERR,
4725                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4726                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
4727                                                le16_to_cpu(gdp->bg_checksum));
4728                                         err = -EFSBADCRC;
4729                                         goto restore_opts;
4730                                 }
4731                         }
4732
4733                         /*
4734                          * If we have an unprocessed orphan list hanging
4735                          * around from a previously readonly bdev mount,
4736                          * require a full umount/remount for now.
4737                          */
4738                         if (es->s_last_orphan) {
4739                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4740                                        "remount RDWR because of unprocessed "
4741                                        "orphan inode list.  Please "
4742                                        "umount/remount instead");
4743                                 err = -EINVAL;
4744                                 goto restore_opts;
4745                         }
4746
4747                         /*
4748                          * Mounting a RDONLY partition read-write, so reread
4749                          * and store the current valid flag.  (It may have
4750                          * been changed by e2fsck since we originally mounted
4751                          * the partition.)
4752                          */
4753                         if (sbi->s_journal)
4754                                 ext4_clear_journal_err(sb, es);
4755                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4756                         if (!ext4_setup_super(sb, es, 0))
4757                                 sb->s_flags &= ~MS_RDONLY;
4758                         if (ext4_has_feature_mmp(sb))
4759                                 if (ext4_multi_mount_protect(sb,
4760                                                 le64_to_cpu(es->s_mmp_block))) {
4761                                         err = -EROFS;
4762                                         goto restore_opts;
4763                                 }
4764                         enable_quota = 1;
4765                 }
4766         }
4767
4768         /*
4769          * Reinitialize lazy itable initialization thread based on
4770          * current settings
4771          */
4772         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4773                 ext4_unregister_li_request(sb);
4774         else {
4775                 ext4_group_t first_not_zeroed;
4776                 first_not_zeroed = ext4_has_uninit_itable(sb);
4777                 ext4_register_li_request(sb, first_not_zeroed);
4778         }
4779
4780         ext4_setup_system_zone(sb);
4781         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4782                 ext4_commit_super(sb, 1);
4783
4784 #ifdef CONFIG_QUOTA
4785         /* Release old quota file names */
4786         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4787                 kfree(old_opts.s_qf_names[i]);
4788         if (enable_quota) {
4789                 if (sb_any_quota_suspended(sb))
4790                         dquot_resume(sb, -1);
4791                 else if (ext4_has_feature_quota(sb)) {
4792                         err = ext4_enable_quotas(sb);
4793                         if (err)
4794                                 goto restore_opts;
4795                 }
4796         }
4797 #endif
4798
4799         *flags = (*flags & ~MS_LAZYTIME) | (sb->s_flags & MS_LAZYTIME);
4800         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4801         kfree(orig_data);
4802         return 0;
4803
4804 restore_opts:
4805         sb->s_flags = old_sb_flags;
4806         sbi->s_mount_opt = old_opts.s_mount_opt;
4807         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4808         sbi->s_resuid = old_opts.s_resuid;
4809         sbi->s_resgid = old_opts.s_resgid;
4810         sbi->s_commit_interval = old_opts.s_commit_interval;
4811         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4812         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4813 #ifdef CONFIG_QUOTA
4814         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4815         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
4816                 kfree(sbi->s_qf_names[i]);
4817                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4818         }
4819 #endif
4820         kfree(orig_data);
4821         return err;
4822 }
4823
4824 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4825 {
4826         struct super_block *sb = dentry->d_sb;
4827         struct ext4_sb_info *sbi = EXT4_SB(sb);
4828         struct ext4_super_block *es = sbi->s_es;
4829         ext4_fsblk_t overhead = 0, resv_blocks;
4830         u64 fsid;
4831         s64 bfree;
4832         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
4833
4834         if (!test_opt(sb, MINIX_DF))
4835                 overhead = sbi->s_overhead;
4836
4837         buf->f_type = EXT4_SUPER_MAGIC;
4838         buf->f_bsize = sb->s_blocksize;
4839         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
4840         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4841                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4842         /* prevent underflow in case that few free space is available */
4843         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4844         buf->f_bavail = buf->f_bfree -
4845                         (ext4_r_blocks_count(es) + resv_blocks);
4846         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
4847                 buf->f_bavail = 0;
4848         buf->f_files = le32_to_cpu(es->s_inodes_count);
4849         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4850         buf->f_namelen = EXT4_NAME_LEN;
4851         fsid = le64_to_cpup((void *)es->s_uuid) ^
4852                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4853         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4854         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4855
4856         return 0;
4857 }
4858
4859 /* Helper function for writing quotas on sync - we need to start transaction
4860  * before quota file is locked for write. Otherwise the are possible deadlocks:
4861  * Process 1                         Process 2
4862  * ext4_create()                     quota_sync()
4863  *   jbd2_journal_start()                  write_dquot()
4864  *   dquot_initialize()                         down(dqio_mutex)
4865  *     down(dqio_mutex)                    jbd2_journal_start()
4866  *
4867  */
4868
4869 #ifdef CONFIG_QUOTA
4870
4871 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4872 {
4873         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
4874 }
4875
4876 static int ext4_write_dquot(struct dquot *dquot)
4877 {
4878         int ret, err;
4879         handle_t *handle;
4880         struct inode *inode;
4881
4882         inode = dquot_to_inode(dquot);
4883         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4884                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4885         if (IS_ERR(handle))
4886                 return PTR_ERR(handle);
4887         ret = dquot_commit(dquot);
4888         err = ext4_journal_stop(handle);
4889         if (!ret)
4890                 ret = err;
4891         return ret;
4892 }
4893
4894 static int ext4_acquire_dquot(struct dquot *dquot)
4895 {
4896         int ret, err;
4897         handle_t *handle;
4898
4899         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4900                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4901         if (IS_ERR(handle))
4902                 return PTR_ERR(handle);
4903         ret = dquot_acquire(dquot);
4904         err = ext4_journal_stop(handle);
4905         if (!ret)
4906                 ret = err;
4907         return ret;
4908 }
4909
4910 static int ext4_release_dquot(struct dquot *dquot)
4911 {
4912         int ret, err;
4913         handle_t *handle;
4914
4915         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4916                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4917         if (IS_ERR(handle)) {
4918                 /* Release dquot anyway to avoid endless cycle in dqput() */
4919                 dquot_release(dquot);
4920                 return PTR_ERR(handle);
4921         }
4922         ret = dquot_release(dquot);
4923         err = ext4_journal_stop(handle);
4924         if (!ret)
4925                 ret = err;
4926         return ret;
4927 }
4928
4929 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4930 {
4931         struct super_block *sb = dquot->dq_sb;
4932         struct ext4_sb_info *sbi = EXT4_SB(sb);
4933
4934         /* Are we journaling quotas? */
4935         if (ext4_has_feature_quota(sb) ||
4936             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
4937                 dquot_mark_dquot_dirty(dquot);
4938                 return ext4_write_dquot(dquot);
4939         } else {
4940                 return dquot_mark_dquot_dirty(dquot);
4941         }
4942 }
4943
4944 static int ext4_write_info(struct super_block *sb, int type)
4945 {
4946         int ret, err;
4947         handle_t *handle;
4948
4949         /* Data block + inode block */
4950         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
4951         if (IS_ERR(handle))
4952                 return PTR_ERR(handle);
4953         ret = dquot_commit_info(sb, type);
4954         err = ext4_journal_stop(handle);
4955         if (!ret)
4956                 ret = err;
4957         return ret;
4958 }
4959
4960 /*
4961  * Turn on quotas during mount time - we need to find
4962  * the quota file and such...
4963  */
4964 static int ext4_quota_on_mount(struct super_block *sb, int type)
4965 {
4966         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4967                                         EXT4_SB(sb)->s_jquota_fmt, type);
4968 }
4969
4970 /*
4971  * Standard function to be called on quota_on
4972  */
4973 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4974                          struct path *path)
4975 {
4976         int err;
4977
4978         if (!test_opt(sb, QUOTA))
4979                 return -EINVAL;
4980
4981         /* Quotafile not on the same filesystem? */
4982         if (path->dentry->d_sb != sb)
4983                 return -EXDEV;
4984         /* Journaling quota? */
4985         if (EXT4_SB(sb)->s_qf_names[type]) {
4986                 /* Quotafile not in fs root? */
4987                 if (path->dentry->d_parent != sb->s_root)
4988                         ext4_msg(sb, KERN_WARNING,
4989                                 "Quota file not on filesystem root. "
4990                                 "Journaled quota will not work");
4991         }
4992
4993         /*
4994          * When we journal data on quota file, we have to flush journal to see
4995          * all updates to the file when we bypass pagecache...
4996          */
4997         if (EXT4_SB(sb)->s_journal &&
4998             ext4_should_journal_data(d_inode(path->dentry))) {
4999                 /*
5000                  * We don't need to lock updates but journal_flush() could
5001                  * otherwise be livelocked...
5002                  */
5003                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5004                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5005                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5006                 if (err)
5007                         return err;
5008         }
5009
5010         return dquot_quota_on(sb, type, format_id, path);
5011 }
5012
5013 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5014                              unsigned int flags)
5015 {
5016         int err;
5017         struct inode *qf_inode;
5018         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5019                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5020                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5021         };
5022
5023         BUG_ON(!ext4_has_feature_quota(sb));
5024
5025         if (!qf_inums[type])
5026                 return -EPERM;
5027
5028         qf_inode = ext4_iget(sb, qf_inums[type]);
5029         if (IS_ERR(qf_inode)) {
5030                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5031                 return PTR_ERR(qf_inode);
5032         }
5033
5034         /* Don't account quota for quota files to avoid recursion */
5035         qf_inode->i_flags |= S_NOQUOTA;
5036         err = dquot_enable(qf_inode, type, format_id, flags);
5037         iput(qf_inode);
5038
5039         return err;
5040 }
5041
5042 /* Enable usage tracking for all quota types. */
5043 static int ext4_enable_quotas(struct super_block *sb)
5044 {
5045         int type, err = 0;
5046         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5047                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5048                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5049         };
5050
5051         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5052         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5053                 if (qf_inums[type]) {
5054                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5055                                                 DQUOT_USAGE_ENABLED);
5056                         if (err) {
5057                                 ext4_warning(sb,
5058                                         "Failed to enable quota tracking "
5059                                         "(type=%d, err=%d). Please run "
5060                                         "e2fsck to fix.", type, err);
5061                                 return err;
5062                         }
5063                 }
5064         }
5065         return 0;
5066 }
5067
5068 static int ext4_quota_off(struct super_block *sb, int type)
5069 {
5070         struct inode *inode = sb_dqopt(sb)->files[type];
5071         handle_t *handle;
5072
5073         /* Force all delayed allocation blocks to be allocated.
5074          * Caller already holds s_umount sem */
5075         if (test_opt(sb, DELALLOC))
5076                 sync_filesystem(sb);
5077
5078         if (!inode)
5079                 goto out;
5080
5081         /* Update modification times of quota files when userspace can
5082          * start looking at them */
5083         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5084         if (IS_ERR(handle))
5085                 goto out;
5086         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5087         ext4_mark_inode_dirty(handle, inode);
5088         ext4_journal_stop(handle);
5089
5090 out:
5091         return dquot_quota_off(sb, type);
5092 }
5093
5094 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5095  * acquiring the locks... As quota files are never truncated and quota code
5096  * itself serializes the operations (and no one else should touch the files)
5097  * we don't have to be afraid of races */
5098 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5099                                size_t len, loff_t off)
5100 {
5101         struct inode *inode = sb_dqopt(sb)->files[type];
5102         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5103         int offset = off & (sb->s_blocksize - 1);
5104         int tocopy;
5105         size_t toread;
5106         struct buffer_head *bh;
5107         loff_t i_size = i_size_read(inode);
5108
5109         if (off > i_size)
5110                 return 0;
5111         if (off+len > i_size)
5112                 len = i_size-off;
5113         toread = len;
5114         while (toread > 0) {
5115                 tocopy = sb->s_blocksize - offset < toread ?
5116                                 sb->s_blocksize - offset : toread;
5117                 bh = ext4_bread(NULL, inode, blk, 0);
5118                 if (IS_ERR(bh))
5119                         return PTR_ERR(bh);
5120                 if (!bh)        /* A hole? */
5121                         memset(data, 0, tocopy);
5122                 else
5123                         memcpy(data, bh->b_data+offset, tocopy);
5124                 brelse(bh);
5125                 offset = 0;
5126                 toread -= tocopy;
5127                 data += tocopy;
5128                 blk++;
5129         }
5130         return len;
5131 }
5132
5133 /* Write to quotafile (we know the transaction is already started and has
5134  * enough credits) */
5135 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5136                                 const char *data, size_t len, loff_t off)
5137 {
5138         struct inode *inode = sb_dqopt(sb)->files[type];
5139         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5140         int err, offset = off & (sb->s_blocksize - 1);
5141         int retries = 0;
5142         struct buffer_head *bh;
5143         handle_t *handle = journal_current_handle();
5144
5145         if (EXT4_SB(sb)->s_journal && !handle) {
5146                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5147                         " cancelled because transaction is not started",
5148                         (unsigned long long)off, (unsigned long long)len);
5149                 return -EIO;
5150         }
5151         /*
5152          * Since we account only one data block in transaction credits,
5153          * then it is impossible to cross a block boundary.
5154          */
5155         if (sb->s_blocksize - offset < len) {
5156                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5157                         " cancelled because not block aligned",
5158                         (unsigned long long)off, (unsigned long long)len);
5159                 return -EIO;
5160         }
5161
5162         do {
5163                 bh = ext4_bread(handle, inode, blk,
5164                                 EXT4_GET_BLOCKS_CREATE |
5165                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5166         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5167                  ext4_should_retry_alloc(inode->i_sb, &retries));
5168         if (IS_ERR(bh))
5169                 return PTR_ERR(bh);
5170         if (!bh)
5171                 goto out;
5172         BUFFER_TRACE(bh, "get write access");
5173         err = ext4_journal_get_write_access(handle, bh);
5174         if (err) {
5175                 brelse(bh);
5176                 return err;
5177         }
5178         lock_buffer(bh);
5179         memcpy(bh->b_data+offset, data, len);
5180         flush_dcache_page(bh->b_page);
5181         unlock_buffer(bh);
5182         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5183         brelse(bh);
5184 out:
5185         if (inode->i_size < off + len) {
5186                 i_size_write(inode, off + len);
5187                 EXT4_I(inode)->i_disksize = inode->i_size;
5188                 ext4_mark_inode_dirty(handle, inode);
5189         }
5190         return len;
5191 }
5192
5193 #endif
5194
5195 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5196                        const char *dev_name, void *data)
5197 {
5198         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5199 }
5200
5201 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5202 static inline void register_as_ext2(void)
5203 {
5204         int err = register_filesystem(&ext2_fs_type);
5205         if (err)
5206                 printk(KERN_WARNING
5207                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5208 }
5209
5210 static inline void unregister_as_ext2(void)
5211 {
5212         unregister_filesystem(&ext2_fs_type);
5213 }
5214
5215 static inline int ext2_feature_set_ok(struct super_block *sb)
5216 {
5217         if (ext4_has_unknown_ext2_incompat_features(sb))
5218                 return 0;
5219         if (sb->s_flags & MS_RDONLY)
5220                 return 1;
5221         if (ext4_has_unknown_ext2_ro_compat_features(sb))
5222                 return 0;
5223         return 1;
5224 }
5225 #else
5226 static inline void register_as_ext2(void) { }
5227 static inline void unregister_as_ext2(void) { }
5228 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5229 #endif
5230
5231 static inline void register_as_ext3(void)
5232 {
5233         int err = register_filesystem(&ext3_fs_type);
5234         if (err)
5235                 printk(KERN_WARNING
5236                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5237 }
5238
5239 static inline void unregister_as_ext3(void)
5240 {
5241         unregister_filesystem(&ext3_fs_type);
5242 }
5243
5244 static inline int ext3_feature_set_ok(struct super_block *sb)
5245 {
5246         if (ext4_has_unknown_ext3_incompat_features(sb))
5247                 return 0;
5248         if (!ext4_has_feature_journal(sb))
5249                 return 0;
5250         if (sb->s_flags & MS_RDONLY)
5251                 return 1;
5252         if (ext4_has_unknown_ext3_ro_compat_features(sb))
5253                 return 0;
5254         return 1;
5255 }
5256
5257 static struct file_system_type ext4_fs_type = {
5258         .owner          = THIS_MODULE,
5259         .name           = "ext4",
5260         .mount          = ext4_mount,
5261         .kill_sb        = kill_block_super,
5262         .fs_flags       = FS_REQUIRES_DEV,
5263 };
5264 MODULE_ALIAS_FS("ext4");
5265
5266 /* Shared across all ext4 file systems */
5267 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5268 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5269
5270 static int __init ext4_init_fs(void)
5271 {
5272         int i, err;
5273
5274         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
5275         ext4_li_info = NULL;
5276         mutex_init(&ext4_li_mtx);
5277
5278         /* Build-time check for flags consistency */
5279         ext4_check_flag_values();
5280
5281         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5282                 mutex_init(&ext4__aio_mutex[i]);
5283                 init_waitqueue_head(&ext4__ioend_wq[i]);
5284         }
5285
5286         err = ext4_init_es();
5287         if (err)
5288                 return err;
5289
5290         err = ext4_init_pageio();
5291         if (err)
5292                 goto out5;
5293
5294         err = ext4_init_system_zone();
5295         if (err)
5296                 goto out4;
5297
5298         err = ext4_init_sysfs();
5299         if (err)
5300                 goto out3;
5301
5302         err = ext4_init_mballoc();
5303         if (err)
5304                 goto out2;
5305         else
5306                 ext4_mballoc_ready = 1;
5307         err = init_inodecache();
5308         if (err)
5309                 goto out1;
5310         register_as_ext3();
5311         register_as_ext2();
5312         err = register_filesystem(&ext4_fs_type);
5313         if (err)
5314                 goto out;
5315
5316         return 0;
5317 out:
5318         unregister_as_ext2();
5319         unregister_as_ext3();
5320         destroy_inodecache();
5321 out1:
5322         ext4_mballoc_ready = 0;
5323         ext4_exit_mballoc();
5324 out2:
5325         ext4_exit_sysfs();
5326 out3:
5327         ext4_exit_system_zone();
5328 out4:
5329         ext4_exit_pageio();
5330 out5:
5331         ext4_exit_es();
5332
5333         return err;
5334 }
5335
5336 static void __exit ext4_exit_fs(void)
5337 {
5338         ext4_exit_crypto();
5339         ext4_destroy_lazyinit_thread();
5340         unregister_as_ext2();
5341         unregister_as_ext3();
5342         unregister_filesystem(&ext4_fs_type);
5343         destroy_inodecache();
5344         ext4_exit_mballoc();
5345         ext4_exit_sysfs();
5346         ext4_exit_system_zone();
5347         ext4_exit_pageio();
5348         ext4_exit_es();
5349 }
5350
5351 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5352 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5353 MODULE_LICENSE("GPL");
5354 module_init(ext4_init_fs)
5355 module_exit(ext4_exit_fs)