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|SLAB_ACCOUNT),
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 static char *quotatypes[] = INITQFNAMES;
1101 #define QTYPE2NAME(t) (quotatypes[t])
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         .get_projid     = ext4_get_projid,
1135 };
1136
1137 static const struct quotactl_ops ext4_qctl_operations = {
1138         .quota_on       = ext4_quota_on,
1139         .quota_off      = ext4_quota_off,
1140         .quota_sync     = dquot_quota_sync,
1141         .get_state      = dquot_get_state,
1142         .set_info       = dquot_set_dqinfo,
1143         .get_dqblk      = dquot_get_dqblk,
1144         .set_dqblk      = dquot_set_dqblk
1145 };
1146 #endif
1147
1148 static const struct super_operations ext4_sops = {
1149         .alloc_inode    = ext4_alloc_inode,
1150         .destroy_inode  = ext4_destroy_inode,
1151         .write_inode    = ext4_write_inode,
1152         .dirty_inode    = ext4_dirty_inode,
1153         .drop_inode     = ext4_drop_inode,
1154         .evict_inode    = ext4_evict_inode,
1155         .put_super      = ext4_put_super,
1156         .sync_fs        = ext4_sync_fs,
1157         .freeze_fs      = ext4_freeze,
1158         .unfreeze_fs    = ext4_unfreeze,
1159         .statfs         = ext4_statfs,
1160         .remount_fs     = ext4_remount,
1161         .show_options   = ext4_show_options,
1162 #ifdef CONFIG_QUOTA
1163         .quota_read     = ext4_quota_read,
1164         .quota_write    = ext4_quota_write,
1165         .get_dquots     = ext4_get_dquots,
1166 #endif
1167         .bdev_try_to_free_page = bdev_try_to_free_page,
1168 };
1169
1170 static const struct export_operations ext4_export_ops = {
1171         .fh_to_dentry = ext4_fh_to_dentry,
1172         .fh_to_parent = ext4_fh_to_parent,
1173         .get_parent = ext4_get_parent,
1174 };
1175
1176 enum {
1177         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1178         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1179         Opt_nouid32, Opt_debug, Opt_removed,
1180         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1181         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1182         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1183         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1184         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1185         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1186         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1187         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1188         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1189         Opt_usrquota, Opt_grpquota, Opt_i_version, Opt_dax,
1190         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1191         Opt_lazytime, Opt_nolazytime,
1192         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1193         Opt_inode_readahead_blks, Opt_journal_ioprio,
1194         Opt_dioread_nolock, Opt_dioread_lock,
1195         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1196         Opt_max_dir_size_kb, Opt_nojournal_checksum,
1197 };
1198
1199 static const match_table_t tokens = {
1200         {Opt_bsd_df, "bsddf"},
1201         {Opt_minix_df, "minixdf"},
1202         {Opt_grpid, "grpid"},
1203         {Opt_grpid, "bsdgroups"},
1204         {Opt_nogrpid, "nogrpid"},
1205         {Opt_nogrpid, "sysvgroups"},
1206         {Opt_resgid, "resgid=%u"},
1207         {Opt_resuid, "resuid=%u"},
1208         {Opt_sb, "sb=%u"},
1209         {Opt_err_cont, "errors=continue"},
1210         {Opt_err_panic, "errors=panic"},
1211         {Opt_err_ro, "errors=remount-ro"},
1212         {Opt_nouid32, "nouid32"},
1213         {Opt_debug, "debug"},
1214         {Opt_removed, "oldalloc"},
1215         {Opt_removed, "orlov"},
1216         {Opt_user_xattr, "user_xattr"},
1217         {Opt_nouser_xattr, "nouser_xattr"},
1218         {Opt_acl, "acl"},
1219         {Opt_noacl, "noacl"},
1220         {Opt_noload, "norecovery"},
1221         {Opt_noload, "noload"},
1222         {Opt_removed, "nobh"},
1223         {Opt_removed, "bh"},
1224         {Opt_commit, "commit=%u"},
1225         {Opt_min_batch_time, "min_batch_time=%u"},
1226         {Opt_max_batch_time, "max_batch_time=%u"},
1227         {Opt_journal_dev, "journal_dev=%u"},
1228         {Opt_journal_path, "journal_path=%s"},
1229         {Opt_journal_checksum, "journal_checksum"},
1230         {Opt_nojournal_checksum, "nojournal_checksum"},
1231         {Opt_journal_async_commit, "journal_async_commit"},
1232         {Opt_abort, "abort"},
1233         {Opt_data_journal, "data=journal"},
1234         {Opt_data_ordered, "data=ordered"},
1235         {Opt_data_writeback, "data=writeback"},
1236         {Opt_data_err_abort, "data_err=abort"},
1237         {Opt_data_err_ignore, "data_err=ignore"},
1238         {Opt_offusrjquota, "usrjquota="},
1239         {Opt_usrjquota, "usrjquota=%s"},
1240         {Opt_offgrpjquota, "grpjquota="},
1241         {Opt_grpjquota, "grpjquota=%s"},
1242         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1243         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1244         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1245         {Opt_grpquota, "grpquota"},
1246         {Opt_noquota, "noquota"},
1247         {Opt_quota, "quota"},
1248         {Opt_usrquota, "usrquota"},
1249         {Opt_barrier, "barrier=%u"},
1250         {Opt_barrier, "barrier"},
1251         {Opt_nobarrier, "nobarrier"},
1252         {Opt_i_version, "i_version"},
1253         {Opt_dax, "dax"},
1254         {Opt_stripe, "stripe=%u"},
1255         {Opt_delalloc, "delalloc"},
1256         {Opt_lazytime, "lazytime"},
1257         {Opt_nolazytime, "nolazytime"},
1258         {Opt_nodelalloc, "nodelalloc"},
1259         {Opt_removed, "mblk_io_submit"},
1260         {Opt_removed, "nomblk_io_submit"},
1261         {Opt_block_validity, "block_validity"},
1262         {Opt_noblock_validity, "noblock_validity"},
1263         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1264         {Opt_journal_ioprio, "journal_ioprio=%u"},
1265         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1266         {Opt_auto_da_alloc, "auto_da_alloc"},
1267         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1268         {Opt_dioread_nolock, "dioread_nolock"},
1269         {Opt_dioread_lock, "dioread_lock"},
1270         {Opt_discard, "discard"},
1271         {Opt_nodiscard, "nodiscard"},
1272         {Opt_init_itable, "init_itable=%u"},
1273         {Opt_init_itable, "init_itable"},
1274         {Opt_noinit_itable, "noinit_itable"},
1275         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1276         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1277         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1278         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1279         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1280         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1281         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1282         {Opt_err, NULL},
1283 };
1284
1285 static ext4_fsblk_t get_sb_block(void **data)
1286 {
1287         ext4_fsblk_t    sb_block;
1288         char            *options = (char *) *data;
1289
1290         if (!options || strncmp(options, "sb=", 3) != 0)
1291                 return 1;       /* Default location */
1292
1293         options += 3;
1294         /* TODO: use simple_strtoll with >32bit ext4 */
1295         sb_block = simple_strtoul(options, &options, 0);
1296         if (*options && *options != ',') {
1297                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1298                        (char *) *data);
1299                 return 1;
1300         }
1301         if (*options == ',')
1302                 options++;
1303         *data = (void *) options;
1304
1305         return sb_block;
1306 }
1307
1308 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1309 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1310         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1311
1312 #ifdef CONFIG_QUOTA
1313 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1314 {
1315         struct ext4_sb_info *sbi = EXT4_SB(sb);
1316         char *qname;
1317         int ret = -1;
1318
1319         if (sb_any_quota_loaded(sb) &&
1320                 !sbi->s_qf_names[qtype]) {
1321                 ext4_msg(sb, KERN_ERR,
1322                         "Cannot change journaled "
1323                         "quota options when quota turned on");
1324                 return -1;
1325         }
1326         if (ext4_has_feature_quota(sb)) {
1327                 ext4_msg(sb, KERN_ERR, "Cannot set journaled quota options "
1328                          "when QUOTA feature is enabled");
1329                 return -1;
1330         }
1331         qname = match_strdup(args);
1332         if (!qname) {
1333                 ext4_msg(sb, KERN_ERR,
1334                         "Not enough memory for storing quotafile name");
1335                 return -1;
1336         }
1337         if (sbi->s_qf_names[qtype]) {
1338                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1339                         ret = 1;
1340                 else
1341                         ext4_msg(sb, KERN_ERR,
1342                                  "%s quota file already specified",
1343                                  QTYPE2NAME(qtype));
1344                 goto errout;
1345         }
1346         if (strchr(qname, '/')) {
1347                 ext4_msg(sb, KERN_ERR,
1348                         "quotafile must be on filesystem root");
1349                 goto errout;
1350         }
1351         sbi->s_qf_names[qtype] = qname;
1352         set_opt(sb, QUOTA);
1353         return 1;
1354 errout:
1355         kfree(qname);
1356         return ret;
1357 }
1358
1359 static int clear_qf_name(struct super_block *sb, int qtype)
1360 {
1361
1362         struct ext4_sb_info *sbi = EXT4_SB(sb);
1363
1364         if (sb_any_quota_loaded(sb) &&
1365                 sbi->s_qf_names[qtype]) {
1366                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1367                         " when quota turned on");
1368                 return -1;
1369         }
1370         kfree(sbi->s_qf_names[qtype]);
1371         sbi->s_qf_names[qtype] = NULL;
1372         return 1;
1373 }
1374 #endif
1375
1376 #define MOPT_SET        0x0001
1377 #define MOPT_CLEAR      0x0002
1378 #define MOPT_NOSUPPORT  0x0004
1379 #define MOPT_EXPLICIT   0x0008
1380 #define MOPT_CLEAR_ERR  0x0010
1381 #define MOPT_GTE0       0x0020
1382 #ifdef CONFIG_QUOTA
1383 #define MOPT_Q          0
1384 #define MOPT_QFMT       0x0040
1385 #else
1386 #define MOPT_Q          MOPT_NOSUPPORT
1387 #define MOPT_QFMT       MOPT_NOSUPPORT
1388 #endif
1389 #define MOPT_DATAJ      0x0080
1390 #define MOPT_NO_EXT2    0x0100
1391 #define MOPT_NO_EXT3    0x0200
1392 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1393 #define MOPT_STRING     0x0400
1394
1395 static const struct mount_opts {
1396         int     token;
1397         int     mount_opt;
1398         int     flags;
1399 } ext4_mount_opts[] = {
1400         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1401         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1402         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1403         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1404         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1405         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1406         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1407          MOPT_EXT4_ONLY | MOPT_SET},
1408         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1409          MOPT_EXT4_ONLY | MOPT_CLEAR},
1410         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1411         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1412         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1413          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1414         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1415          MOPT_EXT4_ONLY | MOPT_CLEAR},
1416         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1417          MOPT_EXT4_ONLY | MOPT_CLEAR},
1418         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1419          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1420         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1421                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1422          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1423         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1424         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1425         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1426         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1427         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1428          MOPT_NO_EXT2 | MOPT_SET},
1429         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1430          MOPT_NO_EXT2 | MOPT_CLEAR},
1431         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1432         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1433         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1434         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1435         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1436         {Opt_commit, 0, MOPT_GTE0},
1437         {Opt_max_batch_time, 0, MOPT_GTE0},
1438         {Opt_min_batch_time, 0, MOPT_GTE0},
1439         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1440         {Opt_init_itable, 0, MOPT_GTE0},
1441         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1442         {Opt_stripe, 0, MOPT_GTE0},
1443         {Opt_resuid, 0, MOPT_GTE0},
1444         {Opt_resgid, 0, MOPT_GTE0},
1445         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1446         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1447         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1448         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1449         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1450         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1451          MOPT_NO_EXT2 | MOPT_DATAJ},
1452         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1453         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1454 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1455         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1456         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1457 #else
1458         {Opt_acl, 0, MOPT_NOSUPPORT},
1459         {Opt_noacl, 0, MOPT_NOSUPPORT},
1460 #endif
1461         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1462         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1463         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1464         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1465                                                         MOPT_SET | MOPT_Q},
1466         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1467                                                         MOPT_SET | MOPT_Q},
1468         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1469                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1470         {Opt_usrjquota, 0, MOPT_Q},
1471         {Opt_grpjquota, 0, MOPT_Q},
1472         {Opt_offusrjquota, 0, MOPT_Q},
1473         {Opt_offgrpjquota, 0, MOPT_Q},
1474         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1475         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1476         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1477         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1478         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1479         {Opt_err, 0, 0}
1480 };
1481
1482 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1483                             substring_t *args, unsigned long *journal_devnum,
1484                             unsigned int *journal_ioprio, int is_remount)
1485 {
1486         struct ext4_sb_info *sbi = EXT4_SB(sb);
1487         const struct mount_opts *m;
1488         kuid_t uid;
1489         kgid_t gid;
1490         int arg = 0;
1491
1492 #ifdef CONFIG_QUOTA
1493         if (token == Opt_usrjquota)
1494                 return set_qf_name(sb, USRQUOTA, &args[0]);
1495         else if (token == Opt_grpjquota)
1496                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1497         else if (token == Opt_offusrjquota)
1498                 return clear_qf_name(sb, USRQUOTA);
1499         else if (token == Opt_offgrpjquota)
1500                 return clear_qf_name(sb, GRPQUOTA);
1501 #endif
1502         switch (token) {
1503         case Opt_noacl:
1504         case Opt_nouser_xattr:
1505                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1506                 break;
1507         case Opt_sb:
1508                 return 1;       /* handled by get_sb_block() */
1509         case Opt_removed:
1510                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1511                 return 1;
1512         case Opt_abort:
1513                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1514                 return 1;
1515         case Opt_i_version:
1516                 sb->s_flags |= MS_I_VERSION;
1517                 return 1;
1518         case Opt_lazytime:
1519                 sb->s_flags |= MS_LAZYTIME;
1520                 return 1;
1521         case Opt_nolazytime:
1522                 sb->s_flags &= ~MS_LAZYTIME;
1523                 return 1;
1524         }
1525
1526         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1527                 if (token == m->token)
1528                         break;
1529
1530         if (m->token == Opt_err) {
1531                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1532                          "or missing value", opt);
1533                 return -1;
1534         }
1535
1536         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1537                 ext4_msg(sb, KERN_ERR,
1538                          "Mount option \"%s\" incompatible with ext2", opt);
1539                 return -1;
1540         }
1541         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1542                 ext4_msg(sb, KERN_ERR,
1543                          "Mount option \"%s\" incompatible with ext3", opt);
1544                 return -1;
1545         }
1546
1547         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1548                 return -1;
1549         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1550                 return -1;
1551         if (m->flags & MOPT_EXPLICIT) {
1552                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1553                         set_opt2(sb, EXPLICIT_DELALLOC);
1554                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1555                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1556                 } else
1557                         return -1;
1558         }
1559         if (m->flags & MOPT_CLEAR_ERR)
1560                 clear_opt(sb, ERRORS_MASK);
1561         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1562                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1563                          "options when quota turned on");
1564                 return -1;
1565         }
1566
1567         if (m->flags & MOPT_NOSUPPORT) {
1568                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1569         } else if (token == Opt_commit) {
1570                 if (arg == 0)
1571                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1572                 sbi->s_commit_interval = HZ * arg;
1573         } else if (token == Opt_max_batch_time) {
1574                 sbi->s_max_batch_time = arg;
1575         } else if (token == Opt_min_batch_time) {
1576                 sbi->s_min_batch_time = arg;
1577         } else if (token == Opt_inode_readahead_blks) {
1578                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1579                         ext4_msg(sb, KERN_ERR,
1580                                  "EXT4-fs: inode_readahead_blks must be "
1581                                  "0 or a power of 2 smaller than 2^31");
1582                         return -1;
1583                 }
1584                 sbi->s_inode_readahead_blks = arg;
1585         } else if (token == Opt_init_itable) {
1586                 set_opt(sb, INIT_INODE_TABLE);
1587                 if (!args->from)
1588                         arg = EXT4_DEF_LI_WAIT_MULT;
1589                 sbi->s_li_wait_mult = arg;
1590         } else if (token == Opt_max_dir_size_kb) {
1591                 sbi->s_max_dir_size_kb = arg;
1592         } else if (token == Opt_stripe) {
1593                 sbi->s_stripe = arg;
1594         } else if (token == Opt_resuid) {
1595                 uid = make_kuid(current_user_ns(), arg);
1596                 if (!uid_valid(uid)) {
1597                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1598                         return -1;
1599                 }
1600                 sbi->s_resuid = uid;
1601         } else if (token == Opt_resgid) {
1602                 gid = make_kgid(current_user_ns(), arg);
1603                 if (!gid_valid(gid)) {
1604                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1605                         return -1;
1606                 }
1607                 sbi->s_resgid = gid;
1608         } else if (token == Opt_journal_dev) {
1609                 if (is_remount) {
1610                         ext4_msg(sb, KERN_ERR,
1611                                  "Cannot specify journal on remount");
1612                         return -1;
1613                 }
1614                 *journal_devnum = arg;
1615         } else if (token == Opt_journal_path) {
1616                 char *journal_path;
1617                 struct inode *journal_inode;
1618                 struct path path;
1619                 int error;
1620
1621                 if (is_remount) {
1622                         ext4_msg(sb, KERN_ERR,
1623                                  "Cannot specify journal on remount");
1624                         return -1;
1625                 }
1626                 journal_path = match_strdup(&args[0]);
1627                 if (!journal_path) {
1628                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1629                                 "journal device string");
1630                         return -1;
1631                 }
1632
1633                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1634                 if (error) {
1635                         ext4_msg(sb, KERN_ERR, "error: could not find "
1636                                 "journal device path: error %d", error);
1637                         kfree(journal_path);
1638                         return -1;
1639                 }
1640
1641                 journal_inode = d_inode(path.dentry);
1642                 if (!S_ISBLK(journal_inode->i_mode)) {
1643                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1644                                 "is not a block device", journal_path);
1645                         path_put(&path);
1646                         kfree(journal_path);
1647                         return -1;
1648                 }
1649
1650                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1651                 path_put(&path);
1652                 kfree(journal_path);
1653         } else if (token == Opt_journal_ioprio) {
1654                 if (arg > 7) {
1655                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1656                                  " (must be 0-7)");
1657                         return -1;
1658                 }
1659                 *journal_ioprio =
1660                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1661         } else if (token == Opt_test_dummy_encryption) {
1662 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1663                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1664                 ext4_msg(sb, KERN_WARNING,
1665                          "Test dummy encryption mode enabled");
1666 #else
1667                 ext4_msg(sb, KERN_WARNING,
1668                          "Test dummy encryption mount option ignored");
1669 #endif
1670         } else if (m->flags & MOPT_DATAJ) {
1671                 if (is_remount) {
1672                         if (!sbi->s_journal)
1673                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1674                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1675                                 ext4_msg(sb, KERN_ERR,
1676                                          "Cannot change data mode on remount");
1677                                 return -1;
1678                         }
1679                 } else {
1680                         clear_opt(sb, DATA_FLAGS);
1681                         sbi->s_mount_opt |= m->mount_opt;
1682                 }
1683 #ifdef CONFIG_QUOTA
1684         } else if (m->flags & MOPT_QFMT) {
1685                 if (sb_any_quota_loaded(sb) &&
1686                     sbi->s_jquota_fmt != m->mount_opt) {
1687                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1688                                  "quota options when quota turned on");
1689                         return -1;
1690                 }
1691                 if (ext4_has_feature_quota(sb)) {
1692                         ext4_msg(sb, KERN_ERR,
1693                                  "Cannot set journaled quota options "
1694                                  "when QUOTA feature is enabled");
1695                         return -1;
1696                 }
1697                 sbi->s_jquota_fmt = m->mount_opt;
1698 #endif
1699         } else if (token == Opt_dax) {
1700 #ifdef CONFIG_FS_DAX
1701                 ext4_msg(sb, KERN_WARNING,
1702                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1703                         sbi->s_mount_opt |= m->mount_opt;
1704 #else
1705                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1706                 return -1;
1707 #endif
1708         } else {
1709                 if (!args->from)
1710                         arg = 1;
1711                 if (m->flags & MOPT_CLEAR)
1712                         arg = !arg;
1713                 else if (unlikely(!(m->flags & MOPT_SET))) {
1714                         ext4_msg(sb, KERN_WARNING,
1715                                  "buggy handling of option %s", opt);
1716                         WARN_ON(1);
1717                         return -1;
1718                 }
1719                 if (arg != 0)
1720                         sbi->s_mount_opt |= m->mount_opt;
1721                 else
1722                         sbi->s_mount_opt &= ~m->mount_opt;
1723         }
1724         return 1;
1725 }
1726
1727 static int parse_options(char *options, struct super_block *sb,
1728                          unsigned long *journal_devnum,
1729                          unsigned int *journal_ioprio,
1730                          int is_remount)
1731 {
1732         struct ext4_sb_info *sbi = EXT4_SB(sb);
1733         char *p;
1734         substring_t args[MAX_OPT_ARGS];
1735         int token;
1736
1737         if (!options)
1738                 return 1;
1739
1740         while ((p = strsep(&options, ",")) != NULL) {
1741                 if (!*p)
1742                         continue;
1743                 /*
1744                  * Initialize args struct so we know whether arg was
1745                  * found; some options take optional arguments.
1746                  */
1747                 args[0].to = args[0].from = NULL;
1748                 token = match_token(p, tokens, args);
1749                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1750                                      journal_ioprio, is_remount) < 0)
1751                         return 0;
1752         }
1753 #ifdef CONFIG_QUOTA
1754         if (ext4_has_feature_quota(sb) &&
1755             (test_opt(sb, USRQUOTA) || test_opt(sb, GRPQUOTA))) {
1756                 ext4_msg(sb, KERN_ERR, "Cannot set quota options when QUOTA "
1757                          "feature is enabled");
1758                 return 0;
1759         }
1760         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1761                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1762                         clear_opt(sb, USRQUOTA);
1763
1764                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1765                         clear_opt(sb, GRPQUOTA);
1766
1767                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1768                         ext4_msg(sb, KERN_ERR, "old and new quota "
1769                                         "format mixing");
1770                         return 0;
1771                 }
1772
1773                 if (!sbi->s_jquota_fmt) {
1774                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1775                                         "not specified");
1776                         return 0;
1777                 }
1778         }
1779 #endif
1780         if (test_opt(sb, DIOREAD_NOLOCK)) {
1781                 int blocksize =
1782                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1783
1784                 if (blocksize < PAGE_CACHE_SIZE) {
1785                         ext4_msg(sb, KERN_ERR, "can't mount with "
1786                                  "dioread_nolock if block size != PAGE_SIZE");
1787                         return 0;
1788                 }
1789         }
1790         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
1791             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
1792                 ext4_msg(sb, KERN_ERR, "can't mount with journal_async_commit "
1793                          "in data=ordered mode");
1794                 return 0;
1795         }
1796         return 1;
1797 }
1798
1799 static inline void ext4_show_quota_options(struct seq_file *seq,
1800                                            struct super_block *sb)
1801 {
1802 #if defined(CONFIG_QUOTA)
1803         struct ext4_sb_info *sbi = EXT4_SB(sb);
1804
1805         if (sbi->s_jquota_fmt) {
1806                 char *fmtname = "";
1807
1808                 switch (sbi->s_jquota_fmt) {
1809                 case QFMT_VFS_OLD:
1810                         fmtname = "vfsold";
1811                         break;
1812                 case QFMT_VFS_V0:
1813                         fmtname = "vfsv0";
1814                         break;
1815                 case QFMT_VFS_V1:
1816                         fmtname = "vfsv1";
1817                         break;
1818                 }
1819                 seq_printf(seq, ",jqfmt=%s", fmtname);
1820         }
1821
1822         if (sbi->s_qf_names[USRQUOTA])
1823                 seq_show_option(seq, "usrjquota", sbi->s_qf_names[USRQUOTA]);
1824
1825         if (sbi->s_qf_names[GRPQUOTA])
1826                 seq_show_option(seq, "grpjquota", sbi->s_qf_names[GRPQUOTA]);
1827 #endif
1828 }
1829
1830 static const char *token2str(int token)
1831 {
1832         const struct match_token *t;
1833
1834         for (t = tokens; t->token != Opt_err; t++)
1835                 if (t->token == token && !strchr(t->pattern, '='))
1836                         break;
1837         return t->pattern;
1838 }
1839
1840 /*
1841  * Show an option if
1842  *  - it's set to a non-default value OR
1843  *  - if the per-sb default is different from the global default
1844  */
1845 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1846                               int nodefs)
1847 {
1848         struct ext4_sb_info *sbi = EXT4_SB(sb);
1849         struct ext4_super_block *es = sbi->s_es;
1850         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1851         const struct mount_opts *m;
1852         char sep = nodefs ? '\n' : ',';
1853
1854 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1855 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1856
1857         if (sbi->s_sb_block != 1)
1858                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1859
1860         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1861                 int want_set = m->flags & MOPT_SET;
1862                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1863                     (m->flags & MOPT_CLEAR_ERR))
1864                         continue;
1865                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1866                         continue; /* skip if same as the default */
1867                 if ((want_set &&
1868                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1869                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1870                         continue; /* select Opt_noFoo vs Opt_Foo */
1871                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1872         }
1873
1874         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1875             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1876                 SEQ_OPTS_PRINT("resuid=%u",
1877                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1878         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1879             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1880                 SEQ_OPTS_PRINT("resgid=%u",
1881                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1882         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1883         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1884                 SEQ_OPTS_PUTS("errors=remount-ro");
1885         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1886                 SEQ_OPTS_PUTS("errors=continue");
1887         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1888                 SEQ_OPTS_PUTS("errors=panic");
1889         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1890                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1891         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1892                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1893         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1894                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1895         if (sb->s_flags & MS_I_VERSION)
1896                 SEQ_OPTS_PUTS("i_version");
1897         if (nodefs || sbi->s_stripe)
1898                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1899         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1900                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1901                         SEQ_OPTS_PUTS("data=journal");
1902                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1903                         SEQ_OPTS_PUTS("data=ordered");
1904                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1905                         SEQ_OPTS_PUTS("data=writeback");
1906         }
1907         if (nodefs ||
1908             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1909                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1910                                sbi->s_inode_readahead_blks);
1911
1912         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1913                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1914                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1915         if (nodefs || sbi->s_max_dir_size_kb)
1916                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1917
1918         ext4_show_quota_options(seq, sb);
1919         return 0;
1920 }
1921
1922 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1923 {
1924         return _ext4_show_options(seq, root->d_sb, 0);
1925 }
1926
1927 int ext4_seq_options_show(struct seq_file *seq, void *offset)
1928 {
1929         struct super_block *sb = seq->private;
1930         int rc;
1931
1932         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1933         rc = _ext4_show_options(seq, sb, 1);
1934         seq_puts(seq, "\n");
1935         return rc;
1936 }
1937
1938 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1939                             int read_only)
1940 {
1941         struct ext4_sb_info *sbi = EXT4_SB(sb);
1942         int res = 0;
1943
1944         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1945                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1946                          "forcing read-only mode");
1947                 res = MS_RDONLY;
1948         }
1949         if (read_only)
1950                 goto done;
1951         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1952                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1953                          "running e2fsck is recommended");
1954         else if (sbi->s_mount_state & EXT4_ERROR_FS)
1955                 ext4_msg(sb, KERN_WARNING,
1956                          "warning: mounting fs with errors, "
1957                          "running e2fsck is recommended");
1958         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1959                  le16_to_cpu(es->s_mnt_count) >=
1960                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1961                 ext4_msg(sb, KERN_WARNING,
1962                          "warning: maximal mount count reached, "
1963                          "running e2fsck is recommended");
1964         else if (le32_to_cpu(es->s_checkinterval) &&
1965                 (le32_to_cpu(es->s_lastcheck) +
1966                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1967                 ext4_msg(sb, KERN_WARNING,
1968                          "warning: checktime reached, "
1969                          "running e2fsck is recommended");
1970         if (!sbi->s_journal)
1971                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1972         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1973                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1974         le16_add_cpu(&es->s_mnt_count, 1);
1975         es->s_mtime = cpu_to_le32(get_seconds());
1976         ext4_update_dynamic_rev(sb);
1977         if (sbi->s_journal)
1978                 ext4_set_feature_journal_needs_recovery(sb);
1979
1980         ext4_commit_super(sb, 1);
1981 done:
1982         if (test_opt(sb, DEBUG))
1983                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1984                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1985                         sb->s_blocksize,
1986                         sbi->s_groups_count,
1987                         EXT4_BLOCKS_PER_GROUP(sb),
1988                         EXT4_INODES_PER_GROUP(sb),
1989                         sbi->s_mount_opt, sbi->s_mount_opt2);
1990
1991         cleancache_init_fs(sb);
1992         return res;
1993 }
1994
1995 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1996 {
1997         struct ext4_sb_info *sbi = EXT4_SB(sb);
1998         struct flex_groups *new_groups;
1999         int size;
2000
2001         if (!sbi->s_log_groups_per_flex)
2002                 return 0;
2003
2004         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2005         if (size <= sbi->s_flex_groups_allocated)
2006                 return 0;
2007
2008         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
2009         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
2010         if (!new_groups) {
2011                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
2012                          size / (int) sizeof(struct flex_groups));
2013                 return -ENOMEM;
2014         }
2015
2016         if (sbi->s_flex_groups) {
2017                 memcpy(new_groups, sbi->s_flex_groups,
2018                        (sbi->s_flex_groups_allocated *
2019                         sizeof(struct flex_groups)));
2020                 kvfree(sbi->s_flex_groups);
2021         }
2022         sbi->s_flex_groups = new_groups;
2023         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
2024         return 0;
2025 }
2026
2027 static int ext4_fill_flex_info(struct super_block *sb)
2028 {
2029         struct ext4_sb_info *sbi = EXT4_SB(sb);
2030         struct ext4_group_desc *gdp = NULL;
2031         ext4_group_t flex_group;
2032         int i, err;
2033
2034         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2035         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2036                 sbi->s_log_groups_per_flex = 0;
2037                 return 1;
2038         }
2039
2040         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2041         if (err)
2042                 goto failed;
2043
2044         for (i = 0; i < sbi->s_groups_count; i++) {
2045                 gdp = ext4_get_group_desc(sb, i, NULL);
2046
2047                 flex_group = ext4_flex_group(sbi, i);
2048                 atomic_add(ext4_free_inodes_count(sb, gdp),
2049                            &sbi->s_flex_groups[flex_group].free_inodes);
2050                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2051                              &sbi->s_flex_groups[flex_group].free_clusters);
2052                 atomic_add(ext4_used_dirs_count(sb, gdp),
2053                            &sbi->s_flex_groups[flex_group].used_dirs);
2054         }
2055
2056         return 1;
2057 failed:
2058         return 0;
2059 }
2060
2061 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2062                                    struct ext4_group_desc *gdp)
2063 {
2064         int offset;
2065         __u16 crc = 0;
2066         __le32 le_group = cpu_to_le32(block_group);
2067         struct ext4_sb_info *sbi = EXT4_SB(sb);
2068
2069         if (ext4_has_metadata_csum(sbi->s_sb)) {
2070                 /* Use new metadata_csum algorithm */
2071                 __le16 save_csum;
2072                 __u32 csum32;
2073
2074                 save_csum = gdp->bg_checksum;
2075                 gdp->bg_checksum = 0;
2076                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2077                                      sizeof(le_group));
2078                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
2079                                      sbi->s_desc_size);
2080                 gdp->bg_checksum = save_csum;
2081
2082                 crc = csum32 & 0xFFFF;
2083                 goto out;
2084         }
2085
2086         /* old crc16 code */
2087         if (!ext4_has_feature_gdt_csum(sb))
2088                 return 0;
2089
2090         offset = offsetof(struct ext4_group_desc, bg_checksum);
2091
2092         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2093         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2094         crc = crc16(crc, (__u8 *)gdp, offset);
2095         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2096         /* for checksum of struct ext4_group_desc do the rest...*/
2097         if (ext4_has_feature_64bit(sb) &&
2098             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2099                 crc = crc16(crc, (__u8 *)gdp + offset,
2100                             le16_to_cpu(sbi->s_es->s_desc_size) -
2101                                 offset);
2102
2103 out:
2104         return cpu_to_le16(crc);
2105 }
2106
2107 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2108                                 struct ext4_group_desc *gdp)
2109 {
2110         if (ext4_has_group_desc_csum(sb) &&
2111             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2112                 return 0;
2113
2114         return 1;
2115 }
2116
2117 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2118                               struct ext4_group_desc *gdp)
2119 {
2120         if (!ext4_has_group_desc_csum(sb))
2121                 return;
2122         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2123 }
2124
2125 /* Called at mount-time, super-block is locked */
2126 static int ext4_check_descriptors(struct super_block *sb,
2127                                   ext4_group_t *first_not_zeroed)
2128 {
2129         struct ext4_sb_info *sbi = EXT4_SB(sb);
2130         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2131         ext4_fsblk_t last_block;
2132         ext4_fsblk_t block_bitmap;
2133         ext4_fsblk_t inode_bitmap;
2134         ext4_fsblk_t inode_table;
2135         int flexbg_flag = 0;
2136         ext4_group_t i, grp = sbi->s_groups_count;
2137
2138         if (ext4_has_feature_flex_bg(sb))
2139                 flexbg_flag = 1;
2140
2141         ext4_debug("Checking group descriptors");
2142
2143         for (i = 0; i < sbi->s_groups_count; i++) {
2144                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2145
2146                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2147                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2148                 else
2149                         last_block = first_block +
2150                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2151
2152                 if ((grp == sbi->s_groups_count) &&
2153                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2154                         grp = i;
2155
2156                 block_bitmap = ext4_block_bitmap(sb, gdp);
2157                 if (block_bitmap < first_block || block_bitmap > last_block) {
2158                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2159                                "Block bitmap for group %u not in group "
2160                                "(block %llu)!", i, block_bitmap);
2161                         return 0;
2162                 }
2163                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2164                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2165                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2166                                "Inode bitmap for group %u not in group "
2167                                "(block %llu)!", i, inode_bitmap);
2168                         return 0;
2169                 }
2170                 inode_table = ext4_inode_table(sb, gdp);
2171                 if (inode_table < first_block ||
2172                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2173                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2174                                "Inode table for group %u not in group "
2175                                "(block %llu)!", i, inode_table);
2176                         return 0;
2177                 }
2178                 ext4_lock_group(sb, i);
2179                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2180                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2181                                  "Checksum for group %u failed (%u!=%u)",
2182                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2183                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2184                         if (!(sb->s_flags & MS_RDONLY)) {
2185                                 ext4_unlock_group(sb, i);
2186                                 return 0;
2187                         }
2188                 }
2189                 ext4_unlock_group(sb, i);
2190                 if (!flexbg_flag)
2191                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2192         }
2193         if (NULL != first_not_zeroed)
2194                 *first_not_zeroed = grp;
2195         return 1;
2196 }
2197
2198 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2199  * the superblock) which were deleted from all directories, but held open by
2200  * a process at the time of a crash.  We walk the list and try to delete these
2201  * inodes at recovery time (only with a read-write filesystem).
2202  *
2203  * In order to keep the orphan inode chain consistent during traversal (in
2204  * case of crash during recovery), we link each inode into the superblock
2205  * orphan list_head and handle it the same way as an inode deletion during
2206  * normal operation (which journals the operations for us).
2207  *
2208  * We only do an iget() and an iput() on each inode, which is very safe if we
2209  * accidentally point at an in-use or already deleted inode.  The worst that
2210  * can happen in this case is that we get a "bit already cleared" message from
2211  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2212  * e2fsck was run on this filesystem, and it must have already done the orphan
2213  * inode cleanup for us, so we can safely abort without any further action.
2214  */
2215 static void ext4_orphan_cleanup(struct super_block *sb,
2216                                 struct ext4_super_block *es)
2217 {
2218         unsigned int s_flags = sb->s_flags;
2219         int nr_orphans = 0, nr_truncates = 0;
2220 #ifdef CONFIG_QUOTA
2221         int i;
2222 #endif
2223         if (!es->s_last_orphan) {
2224                 jbd_debug(4, "no orphan inodes to clean up\n");
2225                 return;
2226         }
2227
2228         if (bdev_read_only(sb->s_bdev)) {
2229                 ext4_msg(sb, KERN_ERR, "write access "
2230                         "unavailable, skipping orphan cleanup");
2231                 return;
2232         }
2233
2234         /* Check if feature set would not allow a r/w mount */
2235         if (!ext4_feature_set_ok(sb, 0)) {
2236                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2237                          "unknown ROCOMPAT features");
2238                 return;
2239         }
2240
2241         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2242                 /* don't clear list on RO mount w/ errors */
2243                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2244                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2245                                   "clearing orphan list.\n");
2246                         es->s_last_orphan = 0;
2247                 }
2248                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2249                 return;
2250         }
2251
2252         if (s_flags & MS_RDONLY) {
2253                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2254                 sb->s_flags &= ~MS_RDONLY;
2255         }
2256 #ifdef CONFIG_QUOTA
2257         /* Needed for iput() to work correctly and not trash data */
2258         sb->s_flags |= MS_ACTIVE;
2259         /* Turn on quotas so that they are updated correctly */
2260         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2261                 if (EXT4_SB(sb)->s_qf_names[i]) {
2262                         int ret = ext4_quota_on_mount(sb, i);
2263                         if (ret < 0)
2264                                 ext4_msg(sb, KERN_ERR,
2265                                         "Cannot turn on journaled "
2266                                         "quota: error %d", ret);
2267                 }
2268         }
2269 #endif
2270
2271         while (es->s_last_orphan) {
2272                 struct inode *inode;
2273
2274                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2275                 if (IS_ERR(inode)) {
2276                         es->s_last_orphan = 0;
2277                         break;
2278                 }
2279
2280                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2281                 dquot_initialize(inode);
2282                 if (inode->i_nlink) {
2283                         if (test_opt(sb, DEBUG))
2284                                 ext4_msg(sb, KERN_DEBUG,
2285                                         "%s: truncating inode %lu to %lld bytes",
2286                                         __func__, inode->i_ino, inode->i_size);
2287                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2288                                   inode->i_ino, inode->i_size);
2289                         mutex_lock(&inode->i_mutex);
2290                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2291                         ext4_truncate(inode);
2292                         mutex_unlock(&inode->i_mutex);
2293                         nr_truncates++;
2294                 } else {
2295                         if (test_opt(sb, DEBUG))
2296                                 ext4_msg(sb, KERN_DEBUG,
2297                                         "%s: deleting unreferenced inode %lu",
2298                                         __func__, inode->i_ino);
2299                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2300                                   inode->i_ino);
2301                         nr_orphans++;
2302                 }
2303                 iput(inode);  /* The delete magic happens here! */
2304         }
2305
2306 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2307
2308         if (nr_orphans)
2309                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2310                        PLURAL(nr_orphans));
2311         if (nr_truncates)
2312                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2313                        PLURAL(nr_truncates));
2314 #ifdef CONFIG_QUOTA
2315         /* Turn quotas off */
2316         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2317                 if (sb_dqopt(sb)->files[i])
2318                         dquot_quota_off(sb, i);
2319         }
2320 #endif
2321         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2322 }
2323
2324 /*
2325  * Maximal extent format file size.
2326  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2327  * extent format containers, within a sector_t, and within i_blocks
2328  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2329  * so that won't be a limiting factor.
2330  *
2331  * However there is other limiting factor. We do store extents in the form
2332  * of starting block and length, hence the resulting length of the extent
2333  * covering maximum file size must fit into on-disk format containers as
2334  * well. Given that length is always by 1 unit bigger than max unit (because
2335  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2336  *
2337  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2338  */
2339 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2340 {
2341         loff_t res;
2342         loff_t upper_limit = MAX_LFS_FILESIZE;
2343
2344         /* small i_blocks in vfs inode? */
2345         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2346                 /*
2347                  * CONFIG_LBDAF is not enabled implies the inode
2348                  * i_block represent total blocks in 512 bytes
2349                  * 32 == size of vfs inode i_blocks * 8
2350                  */
2351                 upper_limit = (1LL << 32) - 1;
2352
2353                 /* total blocks in file system block size */
2354                 upper_limit >>= (blkbits - 9);
2355                 upper_limit <<= blkbits;
2356         }
2357
2358         /*
2359          * 32-bit extent-start container, ee_block. We lower the maxbytes
2360          * by one fs block, so ee_len can cover the extent of maximum file
2361          * size
2362          */
2363         res = (1LL << 32) - 1;
2364         res <<= blkbits;
2365
2366         /* Sanity check against vm- & vfs- imposed limits */
2367         if (res > upper_limit)
2368                 res = upper_limit;
2369
2370         return res;
2371 }
2372
2373 /*
2374  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2375  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2376  * We need to be 1 filesystem block less than the 2^48 sector limit.
2377  */
2378 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2379 {
2380         loff_t res = EXT4_NDIR_BLOCKS;
2381         int meta_blocks;
2382         loff_t upper_limit;
2383         /* This is calculated to be the largest file size for a dense, block
2384          * mapped file such that the file's total number of 512-byte sectors,
2385          * including data and all indirect blocks, does not exceed (2^48 - 1).
2386          *
2387          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2388          * number of 512-byte sectors of the file.
2389          */
2390
2391         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2392                 /*
2393                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2394                  * the inode i_block field represents total file blocks in
2395                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2396                  */
2397                 upper_limit = (1LL << 32) - 1;
2398
2399                 /* total blocks in file system block size */
2400                 upper_limit >>= (bits - 9);
2401
2402         } else {
2403                 /*
2404                  * We use 48 bit ext4_inode i_blocks
2405                  * With EXT4_HUGE_FILE_FL set the i_blocks
2406                  * represent total number of blocks in
2407                  * file system block size
2408                  */
2409                 upper_limit = (1LL << 48) - 1;
2410
2411         }
2412
2413         /* indirect blocks */
2414         meta_blocks = 1;
2415         /* double indirect blocks */
2416         meta_blocks += 1 + (1LL << (bits-2));
2417         /* tripple indirect blocks */
2418         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2419
2420         upper_limit -= meta_blocks;
2421         upper_limit <<= bits;
2422
2423         res += 1LL << (bits-2);
2424         res += 1LL << (2*(bits-2));
2425         res += 1LL << (3*(bits-2));
2426         res <<= bits;
2427         if (res > upper_limit)
2428                 res = upper_limit;
2429
2430         if (res > MAX_LFS_FILESIZE)
2431                 res = MAX_LFS_FILESIZE;
2432
2433         return res;
2434 }
2435
2436 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2437                                    ext4_fsblk_t logical_sb_block, int nr)
2438 {
2439         struct ext4_sb_info *sbi = EXT4_SB(sb);
2440         ext4_group_t bg, first_meta_bg;
2441         int has_super = 0;
2442
2443         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2444
2445         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2446                 return logical_sb_block + nr + 1;
2447         bg = sbi->s_desc_per_block * nr;
2448         if (ext4_bg_has_super(sb, bg))
2449                 has_super = 1;
2450
2451         /*
2452          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2453          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2454          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2455          * compensate.
2456          */
2457         if (sb->s_blocksize == 1024 && nr == 0 &&
2458             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2459                 has_super++;
2460
2461         return (has_super + ext4_group_first_block_no(sb, bg));
2462 }
2463
2464 /**
2465  * ext4_get_stripe_size: Get the stripe size.
2466  * @sbi: In memory super block info
2467  *
2468  * If we have specified it via mount option, then
2469  * use the mount option value. If the value specified at mount time is
2470  * greater than the blocks per group use the super block value.
2471  * If the super block value is greater than blocks per group return 0.
2472  * Allocator needs it be less than blocks per group.
2473  *
2474  */
2475 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2476 {
2477         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2478         unsigned long stripe_width =
2479                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2480         int ret;
2481
2482         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2483                 ret = sbi->s_stripe;
2484         else if (stripe_width <= sbi->s_blocks_per_group)
2485                 ret = stripe_width;
2486         else if (stride <= sbi->s_blocks_per_group)
2487                 ret = stride;
2488         else
2489                 ret = 0;
2490
2491         /*
2492          * If the stripe width is 1, this makes no sense and
2493          * we set it to 0 to turn off stripe handling code.
2494          */
2495         if (ret <= 1)
2496                 ret = 0;
2497
2498         return ret;
2499 }
2500
2501 /*
2502  * Check whether this filesystem can be mounted based on
2503  * the features present and the RDONLY/RDWR mount requested.
2504  * Returns 1 if this filesystem can be mounted as requested,
2505  * 0 if it cannot be.
2506  */
2507 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2508 {
2509         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2510                 ext4_msg(sb, KERN_ERR,
2511                         "Couldn't mount because of "
2512                         "unsupported optional features (%x)",
2513                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2514                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2515                 return 0;
2516         }
2517
2518         if (readonly)
2519                 return 1;
2520
2521         if (ext4_has_feature_readonly(sb)) {
2522                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2523                 sb->s_flags |= MS_RDONLY;
2524                 return 1;
2525         }
2526
2527         /* Check that feature set is OK for a read-write mount */
2528         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2529                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2530                          "unsupported optional features (%x)",
2531                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2532                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2533                 return 0;
2534         }
2535         /*
2536          * Large file size enabled file system can only be mounted
2537          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2538          */
2539         if (ext4_has_feature_huge_file(sb)) {
2540                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2541                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2542                                  "cannot be mounted RDWR without "
2543                                  "CONFIG_LBDAF");
2544                         return 0;
2545                 }
2546         }
2547         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2548                 ext4_msg(sb, KERN_ERR,
2549                          "Can't support bigalloc feature without "
2550                          "extents feature\n");
2551                 return 0;
2552         }
2553
2554 #ifndef CONFIG_QUOTA
2555         if (ext4_has_feature_quota(sb) && !readonly) {
2556                 ext4_msg(sb, KERN_ERR,
2557                          "Filesystem with quota feature cannot be mounted RDWR "
2558                          "without CONFIG_QUOTA");
2559                 return 0;
2560         }
2561         if (ext4_has_feature_project(sb) && !readonly) {
2562                 ext4_msg(sb, KERN_ERR,
2563                          "Filesystem with project quota feature cannot be mounted RDWR "
2564                          "without CONFIG_QUOTA");
2565                 return 0;
2566         }
2567 #endif  /* CONFIG_QUOTA */
2568         return 1;
2569 }
2570
2571 /*
2572  * This function is called once a day if we have errors logged
2573  * on the file system
2574  */
2575 static void print_daily_error_info(unsigned long arg)
2576 {
2577         struct super_block *sb = (struct super_block *) arg;
2578         struct ext4_sb_info *sbi;
2579         struct ext4_super_block *es;
2580
2581         sbi = EXT4_SB(sb);
2582         es = sbi->s_es;
2583
2584         if (es->s_error_count)
2585                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2586                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2587                          le32_to_cpu(es->s_error_count));
2588         if (es->s_first_error_time) {
2589                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2590                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2591                        (int) sizeof(es->s_first_error_func),
2592                        es->s_first_error_func,
2593                        le32_to_cpu(es->s_first_error_line));
2594                 if (es->s_first_error_ino)
2595                         printk(": inode %u",
2596                                le32_to_cpu(es->s_first_error_ino));
2597                 if (es->s_first_error_block)
2598                         printk(": block %llu", (unsigned long long)
2599                                le64_to_cpu(es->s_first_error_block));
2600                 printk("\n");
2601         }
2602         if (es->s_last_error_time) {
2603                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2604                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2605                        (int) sizeof(es->s_last_error_func),
2606                        es->s_last_error_func,
2607                        le32_to_cpu(es->s_last_error_line));
2608                 if (es->s_last_error_ino)
2609                         printk(": inode %u",
2610                                le32_to_cpu(es->s_last_error_ino));
2611                 if (es->s_last_error_block)
2612                         printk(": block %llu", (unsigned long long)
2613                                le64_to_cpu(es->s_last_error_block));
2614                 printk("\n");
2615         }
2616         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2617 }
2618
2619 /* Find next suitable group and run ext4_init_inode_table */
2620 static int ext4_run_li_request(struct ext4_li_request *elr)
2621 {
2622         struct ext4_group_desc *gdp = NULL;
2623         ext4_group_t group, ngroups;
2624         struct super_block *sb;
2625         unsigned long timeout = 0;
2626         int ret = 0;
2627
2628         sb = elr->lr_super;
2629         ngroups = EXT4_SB(sb)->s_groups_count;
2630
2631         sb_start_write(sb);
2632         for (group = elr->lr_next_group; group < ngroups; group++) {
2633                 gdp = ext4_get_group_desc(sb, group, NULL);
2634                 if (!gdp) {
2635                         ret = 1;
2636                         break;
2637                 }
2638
2639                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2640                         break;
2641         }
2642
2643         if (group >= ngroups)
2644                 ret = 1;
2645
2646         if (!ret) {
2647                 timeout = jiffies;
2648                 ret = ext4_init_inode_table(sb, group,
2649                                             elr->lr_timeout ? 0 : 1);
2650                 if (elr->lr_timeout == 0) {
2651                         timeout = (jiffies - timeout) *
2652                                   elr->lr_sbi->s_li_wait_mult;
2653                         elr->lr_timeout = timeout;
2654                 }
2655                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2656                 elr->lr_next_group = group + 1;
2657         }
2658         sb_end_write(sb);
2659
2660         return ret;
2661 }
2662
2663 /*
2664  * Remove lr_request from the list_request and free the
2665  * request structure. Should be called with li_list_mtx held
2666  */
2667 static void ext4_remove_li_request(struct ext4_li_request *elr)
2668 {
2669         struct ext4_sb_info *sbi;
2670
2671         if (!elr)
2672                 return;
2673
2674         sbi = elr->lr_sbi;
2675
2676         list_del(&elr->lr_request);
2677         sbi->s_li_request = NULL;
2678         kfree(elr);
2679 }
2680
2681 static void ext4_unregister_li_request(struct super_block *sb)
2682 {
2683         mutex_lock(&ext4_li_mtx);
2684         if (!ext4_li_info) {
2685                 mutex_unlock(&ext4_li_mtx);
2686                 return;
2687         }
2688
2689         mutex_lock(&ext4_li_info->li_list_mtx);
2690         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2691         mutex_unlock(&ext4_li_info->li_list_mtx);
2692         mutex_unlock(&ext4_li_mtx);
2693 }
2694
2695 static struct task_struct *ext4_lazyinit_task;
2696
2697 /*
2698  * This is the function where ext4lazyinit thread lives. It walks
2699  * through the request list searching for next scheduled filesystem.
2700  * When such a fs is found, run the lazy initialization request
2701  * (ext4_rn_li_request) and keep track of the time spend in this
2702  * function. Based on that time we compute next schedule time of
2703  * the request. When walking through the list is complete, compute
2704  * next waking time and put itself into sleep.
2705  */
2706 static int ext4_lazyinit_thread(void *arg)
2707 {
2708         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2709         struct list_head *pos, *n;
2710         struct ext4_li_request *elr;
2711         unsigned long next_wakeup, cur;
2712
2713         BUG_ON(NULL == eli);
2714
2715 cont_thread:
2716         while (true) {
2717                 next_wakeup = MAX_JIFFY_OFFSET;
2718
2719                 mutex_lock(&eli->li_list_mtx);
2720                 if (list_empty(&eli->li_request_list)) {
2721                         mutex_unlock(&eli->li_list_mtx);
2722                         goto exit_thread;
2723                 }
2724
2725                 list_for_each_safe(pos, n, &eli->li_request_list) {
2726                         elr = list_entry(pos, struct ext4_li_request,
2727                                          lr_request);
2728
2729                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2730                                 if (ext4_run_li_request(elr) != 0) {
2731                                         /* error, remove the lazy_init job */
2732                                         ext4_remove_li_request(elr);
2733                                         continue;
2734                                 }
2735                         }
2736
2737                         if (time_before(elr->lr_next_sched, next_wakeup))
2738                                 next_wakeup = elr->lr_next_sched;
2739                 }
2740                 mutex_unlock(&eli->li_list_mtx);
2741
2742                 try_to_freeze();
2743
2744                 cur = jiffies;
2745                 if ((time_after_eq(cur, next_wakeup)) ||
2746                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2747                         cond_resched();
2748                         continue;
2749                 }
2750
2751                 schedule_timeout_interruptible(next_wakeup - cur);
2752
2753                 if (kthread_should_stop()) {
2754                         ext4_clear_request_list();
2755                         goto exit_thread;
2756                 }
2757         }
2758
2759 exit_thread:
2760         /*
2761          * It looks like the request list is empty, but we need
2762          * to check it under the li_list_mtx lock, to prevent any
2763          * additions into it, and of course we should lock ext4_li_mtx
2764          * to atomically free the list and ext4_li_info, because at
2765          * this point another ext4 filesystem could be registering
2766          * new one.
2767          */
2768         mutex_lock(&ext4_li_mtx);
2769         mutex_lock(&eli->li_list_mtx);
2770         if (!list_empty(&eli->li_request_list)) {
2771                 mutex_unlock(&eli->li_list_mtx);
2772                 mutex_unlock(&ext4_li_mtx);
2773                 goto cont_thread;
2774         }
2775         mutex_unlock(&eli->li_list_mtx);
2776         kfree(ext4_li_info);
2777         ext4_li_info = NULL;
2778         mutex_unlock(&ext4_li_mtx);
2779
2780         return 0;
2781 }
2782
2783 static void ext4_clear_request_list(void)
2784 {
2785         struct list_head *pos, *n;
2786         struct ext4_li_request *elr;
2787
2788         mutex_lock(&ext4_li_info->li_list_mtx);
2789         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2790                 elr = list_entry(pos, struct ext4_li_request,
2791                                  lr_request);
2792                 ext4_remove_li_request(elr);
2793         }
2794         mutex_unlock(&ext4_li_info->li_list_mtx);
2795 }
2796
2797 static int ext4_run_lazyinit_thread(void)
2798 {
2799         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2800                                          ext4_li_info, "ext4lazyinit");
2801         if (IS_ERR(ext4_lazyinit_task)) {
2802                 int err = PTR_ERR(ext4_lazyinit_task);
2803                 ext4_clear_request_list();
2804                 kfree(ext4_li_info);
2805                 ext4_li_info = NULL;
2806                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2807                                  "initialization thread\n",
2808                                  err);
2809                 return err;
2810         }
2811         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2812         return 0;
2813 }
2814
2815 /*
2816  * Check whether it make sense to run itable init. thread or not.
2817  * If there is at least one uninitialized inode table, return
2818  * corresponding group number, else the loop goes through all
2819  * groups and return total number of groups.
2820  */
2821 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2822 {
2823         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2824         struct ext4_group_desc *gdp = NULL;
2825
2826         for (group = 0; group < ngroups; group++) {
2827                 gdp = ext4_get_group_desc(sb, group, NULL);
2828                 if (!gdp)
2829                         continue;
2830
2831                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2832                         break;
2833         }
2834
2835         return group;
2836 }
2837
2838 static int ext4_li_info_new(void)
2839 {
2840         struct ext4_lazy_init *eli = NULL;
2841
2842         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2843         if (!eli)
2844                 return -ENOMEM;
2845
2846         INIT_LIST_HEAD(&eli->li_request_list);
2847         mutex_init(&eli->li_list_mtx);
2848
2849         eli->li_state |= EXT4_LAZYINIT_QUIT;
2850
2851         ext4_li_info = eli;
2852
2853         return 0;
2854 }
2855
2856 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2857                                             ext4_group_t start)
2858 {
2859         struct ext4_sb_info *sbi = EXT4_SB(sb);
2860         struct ext4_li_request *elr;
2861
2862         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2863         if (!elr)
2864                 return NULL;
2865
2866         elr->lr_super = sb;
2867         elr->lr_sbi = sbi;
2868         elr->lr_next_group = start;
2869
2870         /*
2871          * Randomize first schedule time of the request to
2872          * spread the inode table initialization requests
2873          * better.
2874          */
2875         elr->lr_next_sched = jiffies + (prandom_u32() %
2876                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
2877         return elr;
2878 }
2879
2880 int ext4_register_li_request(struct super_block *sb,
2881                              ext4_group_t first_not_zeroed)
2882 {
2883         struct ext4_sb_info *sbi = EXT4_SB(sb);
2884         struct ext4_li_request *elr = NULL;
2885         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2886         int ret = 0;
2887
2888         mutex_lock(&ext4_li_mtx);
2889         if (sbi->s_li_request != NULL) {
2890                 /*
2891                  * Reset timeout so it can be computed again, because
2892                  * s_li_wait_mult might have changed.
2893                  */
2894                 sbi->s_li_request->lr_timeout = 0;
2895                 goto out;
2896         }
2897
2898         if (first_not_zeroed == ngroups ||
2899             (sb->s_flags & MS_RDONLY) ||
2900             !test_opt(sb, INIT_INODE_TABLE))
2901                 goto out;
2902
2903         elr = ext4_li_request_new(sb, first_not_zeroed);
2904         if (!elr) {
2905                 ret = -ENOMEM;
2906                 goto out;
2907         }
2908
2909         if (NULL == ext4_li_info) {
2910                 ret = ext4_li_info_new();
2911                 if (ret)
2912                         goto out;
2913         }
2914
2915         mutex_lock(&ext4_li_info->li_list_mtx);
2916         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2917         mutex_unlock(&ext4_li_info->li_list_mtx);
2918
2919         sbi->s_li_request = elr;
2920         /*
2921          * set elr to NULL here since it has been inserted to
2922          * the request_list and the removal and free of it is
2923          * handled by ext4_clear_request_list from now on.
2924          */
2925         elr = NULL;
2926
2927         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
2928                 ret = ext4_run_lazyinit_thread();
2929                 if (ret)
2930                         goto out;
2931         }
2932 out:
2933         mutex_unlock(&ext4_li_mtx);
2934         if (ret)
2935                 kfree(elr);
2936         return ret;
2937 }
2938
2939 /*
2940  * We do not need to lock anything since this is called on
2941  * module unload.
2942  */
2943 static void ext4_destroy_lazyinit_thread(void)
2944 {
2945         /*
2946          * If thread exited earlier
2947          * there's nothing to be done.
2948          */
2949         if (!ext4_li_info || !ext4_lazyinit_task)
2950                 return;
2951
2952         kthread_stop(ext4_lazyinit_task);
2953 }
2954
2955 static int set_journal_csum_feature_set(struct super_block *sb)
2956 {
2957         int ret = 1;
2958         int compat, incompat;
2959         struct ext4_sb_info *sbi = EXT4_SB(sb);
2960
2961         if (ext4_has_metadata_csum(sb)) {
2962                 /* journal checksum v3 */
2963                 compat = 0;
2964                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
2965         } else {
2966                 /* journal checksum v1 */
2967                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
2968                 incompat = 0;
2969         }
2970
2971         jbd2_journal_clear_features(sbi->s_journal,
2972                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2973                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
2974                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
2975         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2976                 ret = jbd2_journal_set_features(sbi->s_journal,
2977                                 compat, 0,
2978                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
2979                                 incompat);
2980         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2981                 ret = jbd2_journal_set_features(sbi->s_journal,
2982                                 compat, 0,
2983                                 incompat);
2984                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2985                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2986         } else {
2987                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2988                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2989         }
2990
2991         return ret;
2992 }
2993
2994 /*
2995  * Note: calculating the overhead so we can be compatible with
2996  * historical BSD practice is quite difficult in the face of
2997  * clusters/bigalloc.  This is because multiple metadata blocks from
2998  * different block group can end up in the same allocation cluster.
2999  * Calculating the exact overhead in the face of clustered allocation
3000  * requires either O(all block bitmaps) in memory or O(number of block
3001  * groups**2) in time.  We will still calculate the superblock for
3002  * older file systems --- and if we come across with a bigalloc file
3003  * system with zero in s_overhead_clusters the estimate will be close to
3004  * correct especially for very large cluster sizes --- but for newer
3005  * file systems, it's better to calculate this figure once at mkfs
3006  * time, and store it in the superblock.  If the superblock value is
3007  * present (even for non-bigalloc file systems), we will use it.
3008  */
3009 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3010                           char *buf)
3011 {
3012         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3013         struct ext4_group_desc  *gdp;
3014         ext4_fsblk_t            first_block, last_block, b;
3015         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3016         int                     s, j, count = 0;
3017
3018         if (!ext4_has_feature_bigalloc(sb))
3019                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3020                         sbi->s_itb_per_group + 2);
3021
3022         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3023                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3024         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3025         for (i = 0; i < ngroups; i++) {
3026                 gdp = ext4_get_group_desc(sb, i, NULL);
3027                 b = ext4_block_bitmap(sb, gdp);
3028                 if (b >= first_block && b <= last_block) {
3029                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3030                         count++;
3031                 }
3032                 b = ext4_inode_bitmap(sb, gdp);
3033                 if (b >= first_block && b <= last_block) {
3034                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3035                         count++;
3036                 }
3037                 b = ext4_inode_table(sb, gdp);
3038                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3039                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3040                                 int c = EXT4_B2C(sbi, b - first_block);
3041                                 ext4_set_bit(c, buf);
3042                                 count++;
3043                         }
3044                 if (i != grp)
3045                         continue;
3046                 s = 0;
3047                 if (ext4_bg_has_super(sb, grp)) {
3048                         ext4_set_bit(s++, buf);
3049                         count++;
3050                 }
3051                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3052                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3053                         count++;
3054                 }
3055         }
3056         if (!count)
3057                 return 0;
3058         return EXT4_CLUSTERS_PER_GROUP(sb) -
3059                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3060 }
3061
3062 /*
3063  * Compute the overhead and stash it in sbi->s_overhead
3064  */
3065 int ext4_calculate_overhead(struct super_block *sb)
3066 {
3067         struct ext4_sb_info *sbi = EXT4_SB(sb);
3068         struct ext4_super_block *es = sbi->s_es;
3069         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3070         ext4_fsblk_t overhead = 0;
3071         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3072
3073         if (!buf)
3074                 return -ENOMEM;
3075
3076         /*
3077          * Compute the overhead (FS structures).  This is constant
3078          * for a given filesystem unless the number of block groups
3079          * changes so we cache the previous value until it does.
3080          */
3081
3082         /*
3083          * All of the blocks before first_data_block are overhead
3084          */
3085         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3086
3087         /*
3088          * Add the overhead found in each block group
3089          */
3090         for (i = 0; i < ngroups; i++) {
3091                 int blks;
3092
3093                 blks = count_overhead(sb, i, buf);
3094                 overhead += blks;
3095                 if (blks)
3096                         memset(buf, 0, PAGE_SIZE);
3097                 cond_resched();
3098         }
3099         /* Add the internal journal blocks as well */
3100         if (sbi->s_journal && !sbi->journal_bdev)
3101                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3102
3103         sbi->s_overhead = overhead;
3104         smp_wmb();
3105         free_page((unsigned long) buf);
3106         return 0;
3107 }
3108
3109 static void ext4_set_resv_clusters(struct super_block *sb)
3110 {
3111         ext4_fsblk_t resv_clusters;
3112         struct ext4_sb_info *sbi = EXT4_SB(sb);
3113
3114         /*
3115          * There's no need to reserve anything when we aren't using extents.
3116          * The space estimates are exact, there are no unwritten extents,
3117          * hole punching doesn't need new metadata... This is needed especially
3118          * to keep ext2/3 backward compatibility.
3119          */
3120         if (!ext4_has_feature_extents(sb))
3121                 return;
3122         /*
3123          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3124          * This should cover the situations where we can not afford to run
3125          * out of space like for example punch hole, or converting
3126          * unwritten extents in delalloc path. In most cases such
3127          * allocation would require 1, or 2 blocks, higher numbers are
3128          * very rare.
3129          */
3130         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3131                          sbi->s_cluster_bits);
3132
3133         do_div(resv_clusters, 50);
3134         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3135
3136         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3137 }
3138
3139 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3140 {
3141         char *orig_data = kstrdup(data, GFP_KERNEL);
3142         struct buffer_head *bh;
3143         struct ext4_super_block *es = NULL;
3144         struct ext4_sb_info *sbi;
3145         ext4_fsblk_t block;
3146         ext4_fsblk_t sb_block = get_sb_block(&data);
3147         ext4_fsblk_t logical_sb_block;
3148         unsigned long offset = 0;
3149         unsigned long journal_devnum = 0;
3150         unsigned long def_mount_opts;
3151         struct inode *root;
3152         const char *descr;
3153         int ret = -ENOMEM;
3154         int blocksize, clustersize;
3155         unsigned int db_count;
3156         unsigned int i;
3157         int needs_recovery, has_huge_files, has_bigalloc;
3158         __u64 blocks_count;
3159         int err = 0;
3160         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3161         ext4_group_t first_not_zeroed;
3162
3163         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3164         if (!sbi)
3165                 goto out_free_orig;
3166
3167         sbi->s_blockgroup_lock =
3168                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3169         if (!sbi->s_blockgroup_lock) {
3170                 kfree(sbi);
3171                 goto out_free_orig;
3172         }
3173         sb->s_fs_info = sbi;
3174         sbi->s_sb = sb;
3175         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3176         sbi->s_sb_block = sb_block;
3177         if (sb->s_bdev->bd_part)
3178                 sbi->s_sectors_written_start =
3179                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3180
3181         /* Cleanup superblock name */
3182         strreplace(sb->s_id, '/', '!');
3183
3184         /* -EINVAL is default */
3185         ret = -EINVAL;
3186         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3187         if (!blocksize) {
3188                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3189                 goto out_fail;
3190         }
3191
3192         /*
3193          * The ext4 superblock will not be buffer aligned for other than 1kB
3194          * block sizes.  We need to calculate the offset from buffer start.
3195          */
3196         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3197                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3198                 offset = do_div(logical_sb_block, blocksize);
3199         } else {
3200                 logical_sb_block = sb_block;
3201         }
3202
3203         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3204                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3205                 goto out_fail;
3206         }
3207         /*
3208          * Note: s_es must be initialized as soon as possible because
3209          *       some ext4 macro-instructions depend on its value
3210          */
3211         es = (struct ext4_super_block *) (bh->b_data + offset);
3212         sbi->s_es = es;
3213         sb->s_magic = le16_to_cpu(es->s_magic);
3214         if (sb->s_magic != EXT4_SUPER_MAGIC)
3215                 goto cantfind_ext4;
3216         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3217
3218         /* Warn if metadata_csum and gdt_csum are both set. */
3219         if (ext4_has_feature_metadata_csum(sb) &&
3220             ext4_has_feature_gdt_csum(sb))
3221                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3222                              "redundant flags; please run fsck.");
3223
3224         /* Check for a known checksum algorithm */
3225         if (!ext4_verify_csum_type(sb, es)) {
3226                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3227                          "unknown checksum algorithm.");
3228                 silent = 1;
3229                 goto cantfind_ext4;
3230         }
3231
3232         /* Load the checksum driver */
3233         if (ext4_has_feature_metadata_csum(sb)) {
3234                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3235                 if (IS_ERR(sbi->s_chksum_driver)) {
3236                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3237                         ret = PTR_ERR(sbi->s_chksum_driver);
3238                         sbi->s_chksum_driver = NULL;
3239                         goto failed_mount;
3240                 }
3241         }
3242
3243         /* Check superblock checksum */
3244         if (!ext4_superblock_csum_verify(sb, es)) {
3245                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3246                          "invalid superblock checksum.  Run e2fsck?");
3247                 silent = 1;
3248                 ret = -EFSBADCRC;
3249                 goto cantfind_ext4;
3250         }
3251
3252         /* Precompute checksum seed for all metadata */
3253         if (ext4_has_feature_csum_seed(sb))
3254                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3255         else if (ext4_has_metadata_csum(sb))
3256                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3257                                                sizeof(es->s_uuid));
3258
3259         /* Set defaults before we parse the mount options */
3260         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3261         set_opt(sb, INIT_INODE_TABLE);
3262         if (def_mount_opts & EXT4_DEFM_DEBUG)
3263                 set_opt(sb, DEBUG);
3264         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3265                 set_opt(sb, GRPID);
3266         if (def_mount_opts & EXT4_DEFM_UID16)
3267                 set_opt(sb, NO_UID32);
3268         /* xattr user namespace & acls are now defaulted on */
3269         set_opt(sb, XATTR_USER);
3270 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3271         set_opt(sb, POSIX_ACL);
3272 #endif
3273         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3274         if (ext4_has_metadata_csum(sb))
3275                 set_opt(sb, JOURNAL_CHECKSUM);
3276
3277         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3278                 set_opt(sb, JOURNAL_DATA);
3279         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3280                 set_opt(sb, ORDERED_DATA);
3281         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3282                 set_opt(sb, WRITEBACK_DATA);
3283
3284         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3285                 set_opt(sb, ERRORS_PANIC);
3286         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3287                 set_opt(sb, ERRORS_CONT);
3288         else
3289                 set_opt(sb, ERRORS_RO);
3290         /* block_validity enabled by default; disable with noblock_validity */
3291         set_opt(sb, BLOCK_VALIDITY);
3292         if (def_mount_opts & EXT4_DEFM_DISCARD)
3293                 set_opt(sb, DISCARD);
3294
3295         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3296         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3297         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3298         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3299         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3300
3301         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3302                 set_opt(sb, BARRIER);
3303
3304         /*
3305          * enable delayed allocation by default
3306          * Use -o nodelalloc to turn it off
3307          */
3308         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3309             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3310                 set_opt(sb, DELALLOC);
3311
3312         /*
3313          * set default s_li_wait_mult for lazyinit, for the case there is
3314          * no mount option specified.
3315          */
3316         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3317
3318         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3319                            &journal_devnum, &journal_ioprio, 0)) {
3320                 ext4_msg(sb, KERN_WARNING,
3321                          "failed to parse options in superblock: %s",
3322                          sbi->s_es->s_mount_opts);
3323         }
3324         sbi->s_def_mount_opt = sbi->s_mount_opt;
3325         if (!parse_options((char *) data, sb, &journal_devnum,
3326                            &journal_ioprio, 0))
3327                 goto failed_mount;
3328
3329         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3330                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3331                             "with data=journal disables delayed "
3332                             "allocation and O_DIRECT support!\n");
3333                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3334                         ext4_msg(sb, KERN_ERR, "can't mount with "
3335                                  "both data=journal and delalloc");
3336                         goto failed_mount;
3337                 }
3338                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3339                         ext4_msg(sb, KERN_ERR, "can't mount with "
3340                                  "both data=journal and dioread_nolock");
3341                         goto failed_mount;
3342                 }
3343                 if (test_opt(sb, DAX)) {
3344                         ext4_msg(sb, KERN_ERR, "can't mount with "
3345                                  "both data=journal and dax");
3346                         goto failed_mount;
3347                 }
3348                 if (test_opt(sb, DELALLOC))
3349                         clear_opt(sb, DELALLOC);
3350         } else {
3351                 sb->s_iflags |= SB_I_CGROUPWB;
3352         }
3353
3354         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3355                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3356
3357         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3358             (ext4_has_compat_features(sb) ||
3359              ext4_has_ro_compat_features(sb) ||
3360              ext4_has_incompat_features(sb)))
3361                 ext4_msg(sb, KERN_WARNING,
3362                        "feature flags set on rev 0 fs, "
3363                        "running e2fsck is recommended");
3364
3365         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3366                 set_opt2(sb, HURD_COMPAT);
3367                 if (ext4_has_feature_64bit(sb)) {
3368                         ext4_msg(sb, KERN_ERR,
3369                                  "The Hurd can't support 64-bit file systems");
3370                         goto failed_mount;
3371                 }
3372         }
3373
3374         if (IS_EXT2_SB(sb)) {
3375                 if (ext2_feature_set_ok(sb))
3376                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3377                                  "using the ext4 subsystem");
3378                 else {
3379                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3380                                  "to feature incompatibilities");
3381                         goto failed_mount;
3382                 }
3383         }
3384
3385         if (IS_EXT3_SB(sb)) {
3386                 if (ext3_feature_set_ok(sb))
3387                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3388                                  "using the ext4 subsystem");
3389                 else {
3390                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3391                                  "to feature incompatibilities");
3392                         goto failed_mount;
3393                 }
3394         }
3395
3396         /*
3397          * Check feature flags regardless of the revision level, since we
3398          * previously didn't change the revision level when setting the flags,
3399          * so there is a chance incompat flags are set on a rev 0 filesystem.
3400          */
3401         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3402                 goto failed_mount;
3403
3404         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3405         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3406             blocksize > EXT4_MAX_BLOCK_SIZE) {
3407                 ext4_msg(sb, KERN_ERR,
3408                        "Unsupported filesystem blocksize %d", blocksize);
3409                 goto failed_mount;
3410         }
3411
3412         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3413                 if (blocksize != PAGE_SIZE) {
3414                         ext4_msg(sb, KERN_ERR,
3415                                         "error: unsupported blocksize for dax");
3416                         goto failed_mount;
3417                 }
3418                 if (!sb->s_bdev->bd_disk->fops->direct_access) {
3419                         ext4_msg(sb, KERN_ERR,
3420                                         "error: device does not support dax");
3421                         goto failed_mount;
3422                 }
3423         }
3424
3425         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3426                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3427                          es->s_encryption_level);
3428                 goto failed_mount;
3429         }
3430
3431         if (sb->s_blocksize != blocksize) {
3432                 /* Validate the filesystem blocksize */
3433                 if (!sb_set_blocksize(sb, blocksize)) {
3434                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3435                                         blocksize);
3436                         goto failed_mount;
3437                 }
3438
3439                 brelse(bh);
3440                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3441                 offset = do_div(logical_sb_block, blocksize);
3442                 bh = sb_bread_unmovable(sb, logical_sb_block);
3443                 if (!bh) {
3444                         ext4_msg(sb, KERN_ERR,
3445                                "Can't read superblock on 2nd try");
3446                         goto failed_mount;
3447                 }
3448                 es = (struct ext4_super_block *)(bh->b_data + offset);
3449                 sbi->s_es = es;
3450                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3451                         ext4_msg(sb, KERN_ERR,
3452                                "Magic mismatch, very weird!");
3453                         goto failed_mount;
3454                 }
3455         }
3456
3457         has_huge_files = ext4_has_feature_huge_file(sb);
3458         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3459                                                       has_huge_files);
3460         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3461
3462         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3463                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3464                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3465         } else {
3466                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3467                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3468                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3469                     (!is_power_of_2(sbi->s_inode_size)) ||
3470                     (sbi->s_inode_size > blocksize)) {
3471                         ext4_msg(sb, KERN_ERR,
3472                                "unsupported inode size: %d",
3473                                sbi->s_inode_size);
3474                         goto failed_mount;
3475                 }
3476                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3477                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3478         }
3479
3480         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3481         if (ext4_has_feature_64bit(sb)) {
3482                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3483                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3484                     !is_power_of_2(sbi->s_desc_size)) {
3485                         ext4_msg(sb, KERN_ERR,
3486                                "unsupported descriptor size %lu",
3487                                sbi->s_desc_size);
3488                         goto failed_mount;
3489                 }
3490         } else
3491                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3492
3493         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3494         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3495         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3496                 goto cantfind_ext4;
3497
3498         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3499         if (sbi->s_inodes_per_block == 0)
3500                 goto cantfind_ext4;
3501         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3502                                         sbi->s_inodes_per_block;
3503         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3504         sbi->s_sbh = bh;
3505         sbi->s_mount_state = le16_to_cpu(es->s_state);
3506         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3507         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3508
3509         for (i = 0; i < 4; i++)
3510                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3511         sbi->s_def_hash_version = es->s_def_hash_version;
3512         if (ext4_has_feature_dir_index(sb)) {
3513                 i = le32_to_cpu(es->s_flags);
3514                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3515                         sbi->s_hash_unsigned = 3;
3516                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3517 #ifdef __CHAR_UNSIGNED__
3518                         if (!(sb->s_flags & MS_RDONLY))
3519                                 es->s_flags |=
3520                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3521                         sbi->s_hash_unsigned = 3;
3522 #else
3523                         if (!(sb->s_flags & MS_RDONLY))
3524                                 es->s_flags |=
3525                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3526 #endif
3527                 }
3528         }
3529
3530         /* Handle clustersize */
3531         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3532         has_bigalloc = ext4_has_feature_bigalloc(sb);
3533         if (has_bigalloc) {
3534                 if (clustersize < blocksize) {
3535                         ext4_msg(sb, KERN_ERR,
3536                                  "cluster size (%d) smaller than "
3537                                  "block size (%d)", clustersize, blocksize);
3538                         goto failed_mount;
3539                 }
3540                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3541                         le32_to_cpu(es->s_log_block_size);
3542                 sbi->s_clusters_per_group =
3543                         le32_to_cpu(es->s_clusters_per_group);
3544                 if (sbi->s_clusters_per_group > blocksize * 8) {
3545                         ext4_msg(sb, KERN_ERR,
3546                                  "#clusters per group too big: %lu",
3547                                  sbi->s_clusters_per_group);
3548                         goto failed_mount;
3549                 }
3550                 if (sbi->s_blocks_per_group !=
3551                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3552                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3553                                  "clusters per group (%lu) inconsistent",
3554                                  sbi->s_blocks_per_group,
3555                                  sbi->s_clusters_per_group);
3556                         goto failed_mount;
3557                 }
3558         } else {
3559                 if (clustersize != blocksize) {
3560                         ext4_warning(sb, "fragment/cluster size (%d) != "
3561                                      "block size (%d)", clustersize,
3562                                      blocksize);
3563                         clustersize = blocksize;
3564                 }
3565                 if (sbi->s_blocks_per_group > blocksize * 8) {
3566                         ext4_msg(sb, KERN_ERR,
3567                                  "#blocks per group too big: %lu",
3568                                  sbi->s_blocks_per_group);
3569                         goto failed_mount;
3570                 }
3571                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3572                 sbi->s_cluster_bits = 0;
3573         }
3574         sbi->s_cluster_ratio = clustersize / blocksize;
3575
3576         if (sbi->s_inodes_per_group > blocksize * 8) {
3577                 ext4_msg(sb, KERN_ERR,
3578                        "#inodes per group too big: %lu",
3579                        sbi->s_inodes_per_group);
3580                 goto failed_mount;
3581         }
3582
3583         /* Do we have standard group size of clustersize * 8 blocks ? */
3584         if (sbi->s_blocks_per_group == clustersize << 3)
3585                 set_opt2(sb, STD_GROUP_SIZE);
3586
3587         /*
3588          * Test whether we have more sectors than will fit in sector_t,
3589          * and whether the max offset is addressable by the page cache.
3590          */
3591         err = generic_check_addressable(sb->s_blocksize_bits,
3592                                         ext4_blocks_count(es));
3593         if (err) {
3594                 ext4_msg(sb, KERN_ERR, "filesystem"
3595                          " too large to mount safely on this system");
3596                 if (sizeof(sector_t) < 8)
3597                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3598                 goto failed_mount;
3599         }
3600
3601         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3602                 goto cantfind_ext4;
3603
3604         /* check blocks count against device size */
3605         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3606         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3607                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3608                        "exceeds size of device (%llu blocks)",
3609                        ext4_blocks_count(es), blocks_count);
3610                 goto failed_mount;
3611         }
3612
3613         /*
3614          * It makes no sense for the first data block to be beyond the end
3615          * of the filesystem.
3616          */
3617         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3618                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3619                          "block %u is beyond end of filesystem (%llu)",
3620                          le32_to_cpu(es->s_first_data_block),
3621                          ext4_blocks_count(es));
3622                 goto failed_mount;
3623         }
3624         blocks_count = (ext4_blocks_count(es) -
3625                         le32_to_cpu(es->s_first_data_block) +
3626                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3627         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3628         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3629                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3630                        "(block count %llu, first data block %u, "
3631                        "blocks per group %lu)", sbi->s_groups_count,
3632                        ext4_blocks_count(es),
3633                        le32_to_cpu(es->s_first_data_block),
3634                        EXT4_BLOCKS_PER_GROUP(sb));
3635                 goto failed_mount;
3636         }
3637         sbi->s_groups_count = blocks_count;
3638         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3639                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3640         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3641                    EXT4_DESC_PER_BLOCK(sb);
3642         sbi->s_group_desc = ext4_kvmalloc(db_count *
3643                                           sizeof(struct buffer_head *),
3644                                           GFP_KERNEL);
3645         if (sbi->s_group_desc == NULL) {
3646                 ext4_msg(sb, KERN_ERR, "not enough memory");
3647                 ret = -ENOMEM;
3648                 goto failed_mount;
3649         }
3650
3651         bgl_lock_init(sbi->s_blockgroup_lock);
3652
3653         for (i = 0; i < db_count; i++) {
3654                 block = descriptor_loc(sb, logical_sb_block, i);
3655                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
3656                 if (!sbi->s_group_desc[i]) {
3657                         ext4_msg(sb, KERN_ERR,
3658                                "can't read group descriptor %d", i);
3659                         db_count = i;
3660                         goto failed_mount2;
3661                 }
3662         }
3663         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3664                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3665                 ret = -EFSCORRUPTED;
3666                 goto failed_mount2;
3667         }
3668
3669         sbi->s_gdb_count = db_count;
3670         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3671         spin_lock_init(&sbi->s_next_gen_lock);
3672
3673         setup_timer(&sbi->s_err_report, print_daily_error_info,
3674                 (unsigned long) sb);
3675
3676         /* Register extent status tree shrinker */
3677         if (ext4_es_register_shrinker(sbi))
3678                 goto failed_mount3;
3679
3680         sbi->s_stripe = ext4_get_stripe_size(sbi);
3681         sbi->s_extent_max_zeroout_kb = 32;
3682
3683         /*
3684          * set up enough so that it can read an inode
3685          */
3686         sb->s_op = &ext4_sops;
3687         sb->s_export_op = &ext4_export_ops;
3688         sb->s_xattr = ext4_xattr_handlers;
3689 #ifdef CONFIG_QUOTA
3690         sb->dq_op = &ext4_quota_operations;
3691         if (ext4_has_feature_quota(sb))
3692                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
3693         else
3694                 sb->s_qcop = &ext4_qctl_operations;
3695         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
3696 #endif
3697         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3698
3699         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3700         mutex_init(&sbi->s_orphan_lock);
3701
3702         sb->s_root = NULL;
3703
3704         needs_recovery = (es->s_last_orphan != 0 ||
3705                           ext4_has_feature_journal_needs_recovery(sb));
3706
3707         if (ext4_has_feature_mmp(sb) && !(sb->s_flags & MS_RDONLY))
3708                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3709                         goto failed_mount3a;
3710
3711         /*
3712          * The first inode we look at is the journal inode.  Don't try
3713          * root first: it may be modified in the journal!
3714          */
3715         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
3716                 if (ext4_load_journal(sb, es, journal_devnum))
3717                         goto failed_mount3a;
3718         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3719                    ext4_has_feature_journal_needs_recovery(sb)) {
3720                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3721                        "suppressed and not mounted read-only");
3722                 goto failed_mount_wq;
3723         } else {
3724                 /* Nojournal mode, all journal mount options are illegal */
3725                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
3726                         ext4_msg(sb, KERN_ERR, "can't mount with "
3727                                  "journal_checksum, fs mounted w/o journal");
3728                         goto failed_mount_wq;
3729                 }
3730                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3731                         ext4_msg(sb, KERN_ERR, "can't mount with "
3732                                  "journal_async_commit, fs mounted w/o journal");
3733                         goto failed_mount_wq;
3734                 }
3735                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
3736                         ext4_msg(sb, KERN_ERR, "can't mount with "
3737                                  "commit=%lu, fs mounted w/o journal",
3738                                  sbi->s_commit_interval / HZ);
3739                         goto failed_mount_wq;
3740                 }
3741                 if (EXT4_MOUNT_DATA_FLAGS &
3742                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
3743                         ext4_msg(sb, KERN_ERR, "can't mount with "
3744                                  "data=, fs mounted w/o journal");
3745                         goto failed_mount_wq;
3746                 }
3747                 sbi->s_def_mount_opt &= EXT4_MOUNT_JOURNAL_CHECKSUM;
3748                 clear_opt(sb, JOURNAL_CHECKSUM);
3749                 clear_opt(sb, DATA_FLAGS);
3750                 sbi->s_journal = NULL;
3751                 needs_recovery = 0;
3752                 goto no_journal;
3753         }
3754
3755         if (ext4_has_feature_64bit(sb) &&
3756             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3757                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3758                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3759                 goto failed_mount_wq;
3760         }
3761
3762         if (!set_journal_csum_feature_set(sb)) {
3763                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3764                          "feature set");
3765                 goto failed_mount_wq;
3766         }
3767
3768         /* We have now updated the journal if required, so we can
3769          * validate the data journaling mode. */
3770         switch (test_opt(sb, DATA_FLAGS)) {
3771         case 0:
3772                 /* No mode set, assume a default based on the journal
3773                  * capabilities: ORDERED_DATA if the journal can
3774                  * cope, else JOURNAL_DATA
3775                  */
3776                 if (jbd2_journal_check_available_features
3777                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3778                         set_opt(sb, ORDERED_DATA);
3779                 else
3780                         set_opt(sb, JOURNAL_DATA);
3781                 break;
3782
3783         case EXT4_MOUNT_ORDERED_DATA:
3784         case EXT4_MOUNT_WRITEBACK_DATA:
3785                 if (!jbd2_journal_check_available_features
3786                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3787                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3788                                "requested data journaling mode");
3789                         goto failed_mount_wq;
3790                 }
3791         default:
3792                 break;
3793         }
3794         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3795
3796         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3797
3798 no_journal:
3799         if (ext4_mballoc_ready) {
3800                 sbi->s_mb_cache = ext4_xattr_create_cache(sb->s_id);
3801                 if (!sbi->s_mb_cache) {
3802                         ext4_msg(sb, KERN_ERR, "Failed to create an mb_cache");
3803                         goto failed_mount_wq;
3804                 }
3805         }
3806
3807         if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
3808             (blocksize != PAGE_CACHE_SIZE)) {
3809                 ext4_msg(sb, KERN_ERR,
3810                          "Unsupported blocksize for fs encryption");
3811                 goto failed_mount_wq;
3812         }
3813
3814         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !(sb->s_flags & MS_RDONLY) &&
3815             !ext4_has_feature_encrypt(sb)) {
3816                 ext4_set_feature_encrypt(sb);
3817                 ext4_commit_super(sb, 1);
3818         }
3819
3820         /*
3821          * Get the # of file system overhead blocks from the
3822          * superblock if present.
3823          */
3824         if (es->s_overhead_clusters)
3825                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3826         else {
3827                 err = ext4_calculate_overhead(sb);
3828                 if (err)
3829                         goto failed_mount_wq;
3830         }
3831
3832         /*
3833          * The maximum number of concurrent works can be high and
3834          * concurrency isn't really necessary.  Limit it to 1.
3835          */
3836         EXT4_SB(sb)->rsv_conversion_wq =
3837                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3838         if (!EXT4_SB(sb)->rsv_conversion_wq) {
3839                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
3840                 ret = -ENOMEM;
3841                 goto failed_mount4;
3842         }
3843
3844         /*
3845          * The jbd2_journal_load will have done any necessary log recovery,
3846          * so we can safely mount the rest of the filesystem now.
3847          */
3848
3849         root = ext4_iget(sb, EXT4_ROOT_INO);
3850         if (IS_ERR(root)) {
3851                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3852                 ret = PTR_ERR(root);
3853                 root = NULL;
3854                 goto failed_mount4;
3855         }
3856         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3857                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3858                 iput(root);
3859                 goto failed_mount4;
3860         }
3861         sb->s_root = d_make_root(root);
3862         if (!sb->s_root) {
3863                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3864                 ret = -ENOMEM;
3865                 goto failed_mount4;
3866         }
3867
3868         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3869                 sb->s_flags |= MS_RDONLY;
3870
3871         /* determine the minimum size of new large inodes, if present */
3872         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3873                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3874                                                      EXT4_GOOD_OLD_INODE_SIZE;
3875                 if (ext4_has_feature_extra_isize(sb)) {
3876                         if (sbi->s_want_extra_isize <
3877                             le16_to_cpu(es->s_want_extra_isize))
3878                                 sbi->s_want_extra_isize =
3879                                         le16_to_cpu(es->s_want_extra_isize);
3880                         if (sbi->s_want_extra_isize <
3881                             le16_to_cpu(es->s_min_extra_isize))
3882                                 sbi->s_want_extra_isize =
3883                                         le16_to_cpu(es->s_min_extra_isize);
3884                 }
3885         }
3886         /* Check if enough inode space is available */
3887         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3888                                                         sbi->s_inode_size) {
3889                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3890                                                        EXT4_GOOD_OLD_INODE_SIZE;
3891                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3892                          "available");
3893         }
3894
3895         ext4_set_resv_clusters(sb);
3896
3897         err = ext4_setup_system_zone(sb);
3898         if (err) {
3899                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3900                          "zone (%d)", err);
3901                 goto failed_mount4a;
3902         }
3903
3904         ext4_ext_init(sb);
3905         err = ext4_mb_init(sb);
3906         if (err) {
3907                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3908                          err);
3909                 goto failed_mount5;
3910         }
3911
3912         block = ext4_count_free_clusters(sb);
3913         ext4_free_blocks_count_set(sbi->s_es,
3914                                    EXT4_C2B(sbi, block));
3915         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
3916                                   GFP_KERNEL);
3917         if (!err) {
3918                 unsigned long freei = ext4_count_free_inodes(sb);
3919                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
3920                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
3921                                           GFP_KERNEL);
3922         }
3923         if (!err)
3924                 err = percpu_counter_init(&sbi->s_dirs_counter,
3925                                           ext4_count_dirs(sb), GFP_KERNEL);
3926         if (!err)
3927                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
3928                                           GFP_KERNEL);
3929         if (err) {
3930                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3931                 goto failed_mount6;
3932         }
3933
3934         if (ext4_has_feature_flex_bg(sb))
3935                 if (!ext4_fill_flex_info(sb)) {
3936                         ext4_msg(sb, KERN_ERR,
3937                                "unable to initialize "
3938                                "flex_bg meta info!");
3939                         goto failed_mount6;
3940                 }
3941
3942         err = ext4_register_li_request(sb, first_not_zeroed);
3943         if (err)
3944                 goto failed_mount6;
3945
3946         err = ext4_register_sysfs(sb);
3947         if (err)
3948                 goto failed_mount7;
3949
3950 #ifdef CONFIG_QUOTA
3951         /* Enable quota usage during mount. */
3952         if (ext4_has_feature_quota(sb) && !(sb->s_flags & MS_RDONLY)) {
3953                 err = ext4_enable_quotas(sb);
3954                 if (err)
3955                         goto failed_mount8;
3956         }
3957 #endif  /* CONFIG_QUOTA */
3958
3959         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3960         ext4_orphan_cleanup(sb, es);
3961         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3962         if (needs_recovery) {
3963                 ext4_msg(sb, KERN_INFO, "recovery complete");
3964                 ext4_mark_recovery_complete(sb, es);
3965         }
3966         if (EXT4_SB(sb)->s_journal) {
3967                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3968                         descr = " journalled data mode";
3969                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3970                         descr = " ordered data mode";
3971                 else
3972                         descr = " writeback data mode";
3973         } else
3974                 descr = "out journal";
3975
3976         if (test_opt(sb, DISCARD)) {
3977                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
3978                 if (!blk_queue_discard(q))
3979                         ext4_msg(sb, KERN_WARNING,
3980                                  "mounting with \"discard\" option, but "
3981                                  "the device does not support discard");
3982         }
3983
3984         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
3985                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3986                          "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3987                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3988
3989         if (es->s_error_count)
3990                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3991
3992         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
3993         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
3994         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
3995         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
3996
3997         kfree(orig_data);
3998         return 0;
3999
4000 cantfind_ext4:
4001         if (!silent)
4002                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4003         goto failed_mount;
4004
4005 #ifdef CONFIG_QUOTA
4006 failed_mount8:
4007         ext4_unregister_sysfs(sb);
4008 #endif
4009 failed_mount7:
4010         ext4_unregister_li_request(sb);
4011 failed_mount6:
4012         ext4_mb_release(sb);
4013         if (sbi->s_flex_groups)
4014                 kvfree(sbi->s_flex_groups);
4015         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4016         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4017         percpu_counter_destroy(&sbi->s_dirs_counter);
4018         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4019 failed_mount5:
4020         ext4_ext_release(sb);
4021         ext4_release_system_zone(sb);
4022 failed_mount4a:
4023         dput(sb->s_root);
4024         sb->s_root = NULL;
4025 failed_mount4:
4026         ext4_msg(sb, KERN_ERR, "mount failed");
4027         if (EXT4_SB(sb)->rsv_conversion_wq)
4028                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4029 failed_mount_wq:
4030         if (sbi->s_journal) {
4031                 jbd2_journal_destroy(sbi->s_journal);
4032                 sbi->s_journal = NULL;
4033         }
4034 failed_mount3a:
4035         ext4_es_unregister_shrinker(sbi);
4036 failed_mount3:
4037         del_timer_sync(&sbi->s_err_report);
4038         if (sbi->s_mmp_tsk)
4039                 kthread_stop(sbi->s_mmp_tsk);
4040 failed_mount2:
4041         for (i = 0; i < db_count; i++)
4042                 brelse(sbi->s_group_desc[i]);
4043         kvfree(sbi->s_group_desc);
4044 failed_mount:
4045         if (sbi->s_chksum_driver)
4046                 crypto_free_shash(sbi->s_chksum_driver);
4047 #ifdef CONFIG_QUOTA
4048         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4049                 kfree(sbi->s_qf_names[i]);
4050 #endif
4051         ext4_blkdev_remove(sbi);
4052         brelse(bh);
4053 out_fail:
4054         sb->s_fs_info = NULL;
4055         kfree(sbi->s_blockgroup_lock);
4056         kfree(sbi);
4057 out_free_orig:
4058         kfree(orig_data);
4059         return err ? err : ret;
4060 }
4061
4062 /*
4063  * Setup any per-fs journal parameters now.  We'll do this both on
4064  * initial mount, once the journal has been initialised but before we've
4065  * done any recovery; and again on any subsequent remount.
4066  */
4067 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4068 {
4069         struct ext4_sb_info *sbi = EXT4_SB(sb);
4070
4071         journal->j_commit_interval = sbi->s_commit_interval;
4072         journal->j_min_batch_time = sbi->s_min_batch_time;
4073         journal->j_max_batch_time = sbi->s_max_batch_time;
4074
4075         write_lock(&journal->j_state_lock);
4076         if (test_opt(sb, BARRIER))
4077                 journal->j_flags |= JBD2_BARRIER;
4078         else
4079                 journal->j_flags &= ~JBD2_BARRIER;
4080         if (test_opt(sb, DATA_ERR_ABORT))
4081                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4082         else
4083                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4084         write_unlock(&journal->j_state_lock);
4085 }
4086
4087 static journal_t *ext4_get_journal(struct super_block *sb,
4088                                    unsigned int journal_inum)
4089 {
4090         struct inode *journal_inode;
4091         journal_t *journal;
4092
4093         BUG_ON(!ext4_has_feature_journal(sb));
4094
4095         /* First, test for the existence of a valid inode on disk.  Bad
4096          * things happen if we iget() an unused inode, as the subsequent
4097          * iput() will try to delete it. */
4098
4099         journal_inode = ext4_iget(sb, journal_inum);
4100         if (IS_ERR(journal_inode)) {
4101                 ext4_msg(sb, KERN_ERR, "no journal found");
4102                 return NULL;
4103         }
4104         if (!journal_inode->i_nlink) {
4105                 make_bad_inode(journal_inode);
4106                 iput(journal_inode);
4107                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4108                 return NULL;
4109         }
4110
4111         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4112                   journal_inode, journal_inode->i_size);
4113         if (!S_ISREG(journal_inode->i_mode)) {
4114                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4115                 iput(journal_inode);
4116                 return NULL;
4117         }
4118
4119         journal = jbd2_journal_init_inode(journal_inode);
4120         if (!journal) {
4121                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4122                 iput(journal_inode);
4123                 return NULL;
4124         }
4125         journal->j_private = sb;
4126         ext4_init_journal_params(sb, journal);
4127         return journal;
4128 }
4129
4130 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4131                                        dev_t j_dev)
4132 {
4133         struct buffer_head *bh;
4134         journal_t *journal;
4135         ext4_fsblk_t start;
4136         ext4_fsblk_t len;
4137         int hblock, blocksize;
4138         ext4_fsblk_t sb_block;
4139         unsigned long offset;
4140         struct ext4_super_block *es;
4141         struct block_device *bdev;
4142
4143         BUG_ON(!ext4_has_feature_journal(sb));
4144
4145         bdev = ext4_blkdev_get(j_dev, sb);
4146         if (bdev == NULL)
4147                 return NULL;
4148
4149         blocksize = sb->s_blocksize;
4150         hblock = bdev_logical_block_size(bdev);
4151         if (blocksize < hblock) {
4152                 ext4_msg(sb, KERN_ERR,
4153                         "blocksize too small for journal device");
4154                 goto out_bdev;
4155         }
4156
4157         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4158         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4159         set_blocksize(bdev, blocksize);
4160         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4161                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4162                        "external journal");
4163                 goto out_bdev;
4164         }
4165
4166         es = (struct ext4_super_block *) (bh->b_data + offset);
4167         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4168             !(le32_to_cpu(es->s_feature_incompat) &
4169               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4170                 ext4_msg(sb, KERN_ERR, "external journal has "
4171                                         "bad superblock");
4172                 brelse(bh);
4173                 goto out_bdev;
4174         }
4175
4176         if ((le32_to_cpu(es->s_feature_ro_compat) &
4177              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4178             es->s_checksum != ext4_superblock_csum(sb, es)) {
4179                 ext4_msg(sb, KERN_ERR, "external journal has "
4180                                        "corrupt superblock");
4181                 brelse(bh);
4182                 goto out_bdev;
4183         }
4184
4185         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4186                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4187                 brelse(bh);
4188                 goto out_bdev;
4189         }
4190
4191         len = ext4_blocks_count(es);
4192         start = sb_block + 1;
4193         brelse(bh);     /* we're done with the superblock */
4194
4195         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4196                                         start, len, blocksize);
4197         if (!journal) {
4198                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4199                 goto out_bdev;
4200         }
4201         journal->j_private = sb;
4202         ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4203         wait_on_buffer(journal->j_sb_buffer);
4204         if (!buffer_uptodate(journal->j_sb_buffer)) {
4205                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4206                 goto out_journal;
4207         }
4208         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4209                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4210                                         "user (unsupported) - %d",
4211                         be32_to_cpu(journal->j_superblock->s_nr_users));
4212                 goto out_journal;
4213         }
4214         EXT4_SB(sb)->journal_bdev = bdev;
4215         ext4_init_journal_params(sb, journal);
4216         return journal;
4217
4218 out_journal:
4219         jbd2_journal_destroy(journal);
4220 out_bdev:
4221         ext4_blkdev_put(bdev);
4222         return NULL;
4223 }
4224
4225 static int ext4_load_journal(struct super_block *sb,
4226                              struct ext4_super_block *es,
4227                              unsigned long journal_devnum)
4228 {
4229         journal_t *journal;
4230         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4231         dev_t journal_dev;
4232         int err = 0;
4233         int really_read_only;
4234
4235         BUG_ON(!ext4_has_feature_journal(sb));
4236
4237         if (journal_devnum &&
4238             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4239                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4240                         "numbers have changed");
4241                 journal_dev = new_decode_dev(journal_devnum);
4242         } else
4243                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4244
4245         really_read_only = bdev_read_only(sb->s_bdev);
4246
4247         /*
4248          * Are we loading a blank journal or performing recovery after a
4249          * crash?  For recovery, we need to check in advance whether we
4250          * can get read-write access to the device.
4251          */
4252         if (ext4_has_feature_journal_needs_recovery(sb)) {
4253                 if (sb->s_flags & MS_RDONLY) {
4254                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4255                                         "required on readonly filesystem");
4256                         if (really_read_only) {
4257                                 ext4_msg(sb, KERN_ERR, "write access "
4258                                         "unavailable, cannot proceed");
4259                                 return -EROFS;
4260                         }
4261                         ext4_msg(sb, KERN_INFO, "write access will "
4262                                "be enabled during recovery");
4263                 }
4264         }
4265
4266         if (journal_inum && journal_dev) {
4267                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4268                        "and inode journals!");
4269                 return -EINVAL;
4270         }
4271
4272         if (journal_inum) {
4273                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4274                         return -EINVAL;
4275         } else {
4276                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4277                         return -EINVAL;
4278         }
4279
4280         if (!(journal->j_flags & JBD2_BARRIER))
4281                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4282
4283         if (!ext4_has_feature_journal_needs_recovery(sb))
4284                 err = jbd2_journal_wipe(journal, !really_read_only);
4285         if (!err) {
4286                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4287                 if (save)
4288                         memcpy(save, ((char *) es) +
4289                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4290                 err = jbd2_journal_load(journal);
4291                 if (save)
4292                         memcpy(((char *) es) + EXT4_S_ERR_START,
4293                                save, EXT4_S_ERR_LEN);
4294                 kfree(save);
4295         }
4296
4297         if (err) {
4298                 ext4_msg(sb, KERN_ERR, "error loading journal");
4299                 jbd2_journal_destroy(journal);
4300                 return err;
4301         }
4302
4303         EXT4_SB(sb)->s_journal = journal;
4304         ext4_clear_journal_err(sb, es);
4305
4306         if (!really_read_only && journal_devnum &&
4307             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4308                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4309
4310                 /* Make sure we flush the recovery flag to disk. */
4311                 ext4_commit_super(sb, 1);
4312         }
4313
4314         return 0;
4315 }
4316
4317 static int ext4_commit_super(struct super_block *sb, int sync)
4318 {
4319         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4320         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4321         int error = 0;
4322
4323         if (!sbh || block_device_ejected(sb))
4324                 return error;
4325         if (buffer_write_io_error(sbh)) {
4326                 /*
4327                  * Oh, dear.  A previous attempt to write the
4328                  * superblock failed.  This could happen because the
4329                  * USB device was yanked out.  Or it could happen to
4330                  * be a transient write error and maybe the block will
4331                  * be remapped.  Nothing we can do but to retry the
4332                  * write and hope for the best.
4333                  */
4334                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4335                        "superblock detected");
4336                 clear_buffer_write_io_error(sbh);
4337                 set_buffer_uptodate(sbh);
4338         }
4339         /*
4340          * If the file system is mounted read-only, don't update the
4341          * superblock write time.  This avoids updating the superblock
4342          * write time when we are mounting the root file system
4343          * read/only but we need to replay the journal; at that point,
4344          * for people who are east of GMT and who make their clock
4345          * tick in localtime for Windows bug-for-bug compatibility,
4346          * the clock is set in the future, and this will cause e2fsck
4347          * to complain and force a full file system check.
4348          */
4349         if (!(sb->s_flags & MS_RDONLY))
4350                 es->s_wtime = cpu_to_le32(get_seconds());
4351         if (sb->s_bdev->bd_part)
4352                 es->s_kbytes_written =
4353                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4354                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4355                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4356         else
4357                 es->s_kbytes_written =
4358                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4359         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4360                 ext4_free_blocks_count_set(es,
4361                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4362                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4363         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4364                 es->s_free_inodes_count =
4365                         cpu_to_le32(percpu_counter_sum_positive(
4366                                 &EXT4_SB(sb)->s_freeinodes_counter));
4367         BUFFER_TRACE(sbh, "marking dirty");
4368         ext4_superblock_csum_set(sb);
4369         mark_buffer_dirty(sbh);
4370         if (sync) {
4371                 error = __sync_dirty_buffer(sbh,
4372                         test_opt(sb, BARRIER) ? WRITE_FUA : WRITE_SYNC);
4373                 if (error)
4374                         return error;
4375
4376                 error = buffer_write_io_error(sbh);
4377                 if (error) {
4378                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4379                                "superblock");
4380                         clear_buffer_write_io_error(sbh);
4381                         set_buffer_uptodate(sbh);
4382                 }
4383         }
4384         return error;
4385 }
4386
4387 /*
4388  * Have we just finished recovery?  If so, and if we are mounting (or
4389  * remounting) the filesystem readonly, then we will end up with a
4390  * consistent fs on disk.  Record that fact.
4391  */
4392 static void ext4_mark_recovery_complete(struct super_block *sb,
4393                                         struct ext4_super_block *es)
4394 {
4395         journal_t *journal = EXT4_SB(sb)->s_journal;
4396
4397         if (!ext4_has_feature_journal(sb)) {
4398                 BUG_ON(journal != NULL);
4399                 return;
4400         }
4401         jbd2_journal_lock_updates(journal);
4402         if (jbd2_journal_flush(journal) < 0)
4403                 goto out;
4404
4405         if (ext4_has_feature_journal_needs_recovery(sb) &&
4406             sb->s_flags & MS_RDONLY) {
4407                 ext4_clear_feature_journal_needs_recovery(sb);
4408                 ext4_commit_super(sb, 1);
4409         }
4410
4411 out:
4412         jbd2_journal_unlock_updates(journal);
4413 }
4414
4415 /*
4416  * If we are mounting (or read-write remounting) a filesystem whose journal
4417  * has recorded an error from a previous lifetime, move that error to the
4418  * main filesystem now.
4419  */
4420 static void ext4_clear_journal_err(struct super_block *sb,
4421                                    struct ext4_super_block *es)
4422 {
4423         journal_t *journal;
4424         int j_errno;
4425         const char *errstr;
4426
4427         BUG_ON(!ext4_has_feature_journal(sb));
4428
4429         journal = EXT4_SB(sb)->s_journal;
4430
4431         /*
4432          * Now check for any error status which may have been recorded in the
4433          * journal by a prior ext4_error() or ext4_abort()
4434          */
4435
4436         j_errno = jbd2_journal_errno(journal);
4437         if (j_errno) {
4438                 char nbuf[16];
4439
4440                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4441                 ext4_warning(sb, "Filesystem error recorded "
4442                              "from previous mount: %s", errstr);
4443                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4444
4445                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4446                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4447                 ext4_commit_super(sb, 1);
4448
4449                 jbd2_journal_clear_err(journal);
4450                 jbd2_journal_update_sb_errno(journal);
4451         }
4452 }
4453
4454 /*
4455  * Force the running and committing transactions to commit,
4456  * and wait on the commit.
4457  */
4458 int ext4_force_commit(struct super_block *sb)
4459 {
4460         journal_t *journal;
4461
4462         if (sb->s_flags & MS_RDONLY)
4463                 return 0;
4464
4465         journal = EXT4_SB(sb)->s_journal;
4466         return ext4_journal_force_commit(journal);
4467 }
4468
4469 static int ext4_sync_fs(struct super_block *sb, int wait)
4470 {
4471         int ret = 0;
4472         tid_t target;
4473         bool needs_barrier = false;
4474         struct ext4_sb_info *sbi = EXT4_SB(sb);
4475
4476         trace_ext4_sync_fs(sb, wait);
4477         flush_workqueue(sbi->rsv_conversion_wq);
4478         /*
4479          * Writeback quota in non-journalled quota case - journalled quota has
4480          * no dirty dquots
4481          */
4482         dquot_writeback_dquots(sb, -1);
4483         /*
4484          * Data writeback is possible w/o journal transaction, so barrier must
4485          * being sent at the end of the function. But we can skip it if
4486          * transaction_commit will do it for us.
4487          */
4488         if (sbi->s_journal) {
4489                 target = jbd2_get_latest_transaction(sbi->s_journal);
4490                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4491                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4492                         needs_barrier = true;
4493
4494                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4495                         if (wait)
4496                                 ret = jbd2_log_wait_commit(sbi->s_journal,
4497                                                            target);
4498                 }
4499         } else if (wait && test_opt(sb, BARRIER))
4500                 needs_barrier = true;
4501         if (needs_barrier) {
4502                 int err;
4503                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4504                 if (!ret)
4505                         ret = err;
4506         }
4507
4508         return ret;
4509 }
4510
4511 /*
4512  * LVM calls this function before a (read-only) snapshot is created.  This
4513  * gives us a chance to flush the journal completely and mark the fs clean.
4514  *
4515  * Note that only this function cannot bring a filesystem to be in a clean
4516  * state independently. It relies on upper layer to stop all data & metadata
4517  * modifications.
4518  */
4519 static int ext4_freeze(struct super_block *sb)
4520 {
4521         int error = 0;
4522         journal_t *journal;
4523
4524         if (sb->s_flags & MS_RDONLY)
4525                 return 0;
4526
4527         journal = EXT4_SB(sb)->s_journal;
4528
4529         if (journal) {
4530                 /* Now we set up the journal barrier. */
4531                 jbd2_journal_lock_updates(journal);
4532
4533                 /*
4534                  * Don't clear the needs_recovery flag if we failed to
4535                  * flush the journal.
4536                  */
4537                 error = jbd2_journal_flush(journal);
4538                 if (error < 0)
4539                         goto out;
4540
4541                 /* Journal blocked and flushed, clear needs_recovery flag. */
4542                 ext4_clear_feature_journal_needs_recovery(sb);
4543         }
4544
4545         error = ext4_commit_super(sb, 1);
4546 out:
4547         if (journal)
4548                 /* we rely on upper layer to stop further updates */
4549                 jbd2_journal_unlock_updates(journal);
4550         return error;
4551 }
4552
4553 /*
4554  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4555  * flag here, even though the filesystem is not technically dirty yet.
4556  */
4557 static int ext4_unfreeze(struct super_block *sb)
4558 {
4559         if (sb->s_flags & MS_RDONLY)
4560                 return 0;
4561
4562         if (EXT4_SB(sb)->s_journal) {
4563                 /* Reset the needs_recovery flag before the fs is unlocked. */
4564                 ext4_set_feature_journal_needs_recovery(sb);
4565         }
4566
4567         ext4_commit_super(sb, 1);
4568         return 0;
4569 }
4570
4571 /*
4572  * Structure to save mount options for ext4_remount's benefit
4573  */
4574 struct ext4_mount_options {
4575         unsigned long s_mount_opt;
4576         unsigned long s_mount_opt2;
4577         kuid_t s_resuid;
4578         kgid_t s_resgid;
4579         unsigned long s_commit_interval;
4580         u32 s_min_batch_time, s_max_batch_time;
4581 #ifdef CONFIG_QUOTA
4582         int s_jquota_fmt;
4583         char *s_qf_names[EXT4_MAXQUOTAS];
4584 #endif
4585 };
4586
4587 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4588 {
4589         struct ext4_super_block *es;
4590         struct ext4_sb_info *sbi = EXT4_SB(sb);
4591         unsigned long old_sb_flags;
4592         struct ext4_mount_options old_opts;
4593         int enable_quota = 0;
4594         ext4_group_t g;
4595         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4596         int err = 0;
4597 #ifdef CONFIG_QUOTA
4598         int i, j;
4599 #endif
4600         char *orig_data = kstrdup(data, GFP_KERNEL);
4601
4602         /* Store the original options */
4603         old_sb_flags = sb->s_flags;
4604         old_opts.s_mount_opt = sbi->s_mount_opt;
4605         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4606         old_opts.s_resuid = sbi->s_resuid;
4607         old_opts.s_resgid = sbi->s_resgid;
4608         old_opts.s_commit_interval = sbi->s_commit_interval;
4609         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4610         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4611 #ifdef CONFIG_QUOTA
4612         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4613         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4614                 if (sbi->s_qf_names[i]) {
4615                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4616                                                          GFP_KERNEL);
4617                         if (!old_opts.s_qf_names[i]) {
4618                                 for (j = 0; j < i; j++)
4619                                         kfree(old_opts.s_qf_names[j]);
4620                                 kfree(orig_data);
4621                                 return -ENOMEM;
4622                         }
4623                 } else
4624                         old_opts.s_qf_names[i] = NULL;
4625 #endif
4626         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4627                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4628
4629         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4630                 err = -EINVAL;
4631                 goto restore_opts;
4632         }
4633
4634         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
4635             test_opt(sb, JOURNAL_CHECKSUM)) {
4636                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
4637                          "during remount not supported; ignoring");
4638                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
4639         }
4640
4641         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4642                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4643                         ext4_msg(sb, KERN_ERR, "can't mount with "
4644                                  "both data=journal and delalloc");
4645                         err = -EINVAL;
4646                         goto restore_opts;
4647                 }
4648                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4649                         ext4_msg(sb, KERN_ERR, "can't mount with "
4650                                  "both data=journal and dioread_nolock");
4651                         err = -EINVAL;
4652                         goto restore_opts;
4653                 }
4654                 if (test_opt(sb, DAX)) {
4655                         ext4_msg(sb, KERN_ERR, "can't mount with "
4656                                  "both data=journal and dax");
4657                         err = -EINVAL;
4658                         goto restore_opts;
4659                 }
4660         }
4661
4662         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
4663                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
4664                         "dax flag with busy inodes while remounting");
4665                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
4666         }
4667
4668         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4669                 ext4_abort(sb, "Abort forced by user");
4670
4671         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4672                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4673
4674         es = sbi->s_es;
4675
4676         if (sbi->s_journal) {
4677                 ext4_init_journal_params(sb, sbi->s_journal);
4678                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4679         }
4680
4681         if (*flags & MS_LAZYTIME)
4682                 sb->s_flags |= MS_LAZYTIME;
4683
4684         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4685                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4686                         err = -EROFS;
4687                         goto restore_opts;
4688                 }
4689
4690                 if (*flags & MS_RDONLY) {
4691                         err = sync_filesystem(sb);
4692                         if (err < 0)
4693                                 goto restore_opts;
4694                         err = dquot_suspend(sb, -1);
4695                         if (err < 0)
4696                                 goto restore_opts;
4697
4698                         /*
4699                          * First of all, the unconditional stuff we have to do
4700                          * to disable replay of the journal when we next remount
4701                          */
4702                         sb->s_flags |= MS_RDONLY;
4703
4704                         /*
4705                          * OK, test if we are remounting a valid rw partition
4706                          * readonly, and if so set the rdonly flag and then
4707                          * mark the partition as valid again.
4708                          */
4709                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4710                             (sbi->s_mount_state & EXT4_VALID_FS))
4711                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4712
4713                         if (sbi->s_journal)
4714                                 ext4_mark_recovery_complete(sb, es);
4715                 } else {
4716                         /* Make sure we can mount this feature set readwrite */
4717                         if (ext4_has_feature_readonly(sb) ||
4718                             !ext4_feature_set_ok(sb, 0)) {
4719                                 err = -EROFS;
4720                                 goto restore_opts;
4721                         }
4722                         /*
4723                          * Make sure the group descriptor checksums
4724                          * are sane.  If they aren't, refuse to remount r/w.
4725                          */
4726                         for (g = 0; g < sbi->s_groups_count; g++) {
4727                                 struct ext4_group_desc *gdp =
4728                                         ext4_get_group_desc(sb, g, NULL);
4729
4730                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4731                                         ext4_msg(sb, KERN_ERR,
4732                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4733                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
4734                                                le16_to_cpu(gdp->bg_checksum));
4735                                         err = -EFSBADCRC;
4736                                         goto restore_opts;
4737                                 }
4738                         }
4739
4740                         /*
4741                          * If we have an unprocessed orphan list hanging
4742                          * around from a previously readonly bdev mount,
4743                          * require a full umount/remount for now.
4744                          */
4745                         if (es->s_last_orphan) {
4746                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4747                                        "remount RDWR because of unprocessed "
4748                                        "orphan inode list.  Please "
4749                                        "umount/remount instead");
4750                                 err = -EINVAL;
4751                                 goto restore_opts;
4752                         }
4753
4754                         /*
4755                          * Mounting a RDONLY partition read-write, so reread
4756                          * and store the current valid flag.  (It may have
4757                          * been changed by e2fsck since we originally mounted
4758                          * the partition.)
4759                          */
4760                         if (sbi->s_journal)
4761                                 ext4_clear_journal_err(sb, es);
4762                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4763                         if (!ext4_setup_super(sb, es, 0))
4764                                 sb->s_flags &= ~MS_RDONLY;
4765                         if (ext4_has_feature_mmp(sb))
4766                                 if (ext4_multi_mount_protect(sb,
4767                                                 le64_to_cpu(es->s_mmp_block))) {
4768                                         err = -EROFS;
4769                                         goto restore_opts;
4770                                 }
4771                         enable_quota = 1;
4772                 }
4773         }
4774
4775         /*
4776          * Reinitialize lazy itable initialization thread based on
4777          * current settings
4778          */
4779         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4780                 ext4_unregister_li_request(sb);
4781         else {
4782                 ext4_group_t first_not_zeroed;
4783                 first_not_zeroed = ext4_has_uninit_itable(sb);
4784                 ext4_register_li_request(sb, first_not_zeroed);
4785         }
4786
4787         ext4_setup_system_zone(sb);
4788         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4789                 ext4_commit_super(sb, 1);
4790
4791 #ifdef CONFIG_QUOTA
4792         /* Release old quota file names */
4793         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4794                 kfree(old_opts.s_qf_names[i]);
4795         if (enable_quota) {
4796                 if (sb_any_quota_suspended(sb))
4797                         dquot_resume(sb, -1);
4798                 else if (ext4_has_feature_quota(sb)) {
4799                         err = ext4_enable_quotas(sb);
4800                         if (err)
4801                                 goto restore_opts;
4802                 }
4803         }
4804 #endif
4805
4806         *flags = (*flags & ~MS_LAZYTIME) | (sb->s_flags & MS_LAZYTIME);
4807         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4808         kfree(orig_data);
4809         return 0;
4810
4811 restore_opts:
4812         sb->s_flags = old_sb_flags;
4813         sbi->s_mount_opt = old_opts.s_mount_opt;
4814         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4815         sbi->s_resuid = old_opts.s_resuid;
4816         sbi->s_resgid = old_opts.s_resgid;
4817         sbi->s_commit_interval = old_opts.s_commit_interval;
4818         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4819         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4820 #ifdef CONFIG_QUOTA
4821         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4822         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
4823                 kfree(sbi->s_qf_names[i]);
4824                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4825         }
4826 #endif
4827         kfree(orig_data);
4828         return err;
4829 }
4830
4831 #ifdef CONFIG_QUOTA
4832 static int ext4_statfs_project(struct super_block *sb,
4833                                kprojid_t projid, struct kstatfs *buf)
4834 {
4835         struct kqid qid;
4836         struct dquot *dquot;
4837         u64 limit;
4838         u64 curblock;
4839
4840         qid = make_kqid_projid(projid);
4841         dquot = dqget(sb, qid);
4842         if (IS_ERR(dquot))
4843                 return PTR_ERR(dquot);
4844         spin_lock(&dq_data_lock);
4845
4846         limit = (dquot->dq_dqb.dqb_bsoftlimit ?
4847                  dquot->dq_dqb.dqb_bsoftlimit :
4848                  dquot->dq_dqb.dqb_bhardlimit) >> sb->s_blocksize_bits;
4849         if (limit && buf->f_blocks > limit) {
4850                 curblock = dquot->dq_dqb.dqb_curspace >> sb->s_blocksize_bits;
4851                 buf->f_blocks = limit;
4852                 buf->f_bfree = buf->f_bavail =
4853                         (buf->f_blocks > curblock) ?
4854                          (buf->f_blocks - curblock) : 0;
4855         }
4856
4857         limit = dquot->dq_dqb.dqb_isoftlimit ?
4858                 dquot->dq_dqb.dqb_isoftlimit :
4859                 dquot->dq_dqb.dqb_ihardlimit;
4860         if (limit && buf->f_files > limit) {
4861                 buf->f_files = limit;
4862                 buf->f_ffree =
4863                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
4864                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
4865         }
4866
4867         spin_unlock(&dq_data_lock);
4868         dqput(dquot);
4869         return 0;
4870 }
4871 #endif
4872
4873 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4874 {
4875         struct super_block *sb = dentry->d_sb;
4876         struct ext4_sb_info *sbi = EXT4_SB(sb);
4877         struct ext4_super_block *es = sbi->s_es;
4878         ext4_fsblk_t overhead = 0, resv_blocks;
4879         u64 fsid;
4880         s64 bfree;
4881         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
4882
4883         if (!test_opt(sb, MINIX_DF))
4884                 overhead = sbi->s_overhead;
4885
4886         buf->f_type = EXT4_SUPER_MAGIC;
4887         buf->f_bsize = sb->s_blocksize;
4888         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
4889         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4890                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4891         /* prevent underflow in case that few free space is available */
4892         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4893         buf->f_bavail = buf->f_bfree -
4894                         (ext4_r_blocks_count(es) + resv_blocks);
4895         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
4896                 buf->f_bavail = 0;
4897         buf->f_files = le32_to_cpu(es->s_inodes_count);
4898         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4899         buf->f_namelen = EXT4_NAME_LEN;
4900         fsid = le64_to_cpup((void *)es->s_uuid) ^
4901                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4902         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4903         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4904
4905 #ifdef CONFIG_QUOTA
4906         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
4907             sb_has_quota_limits_enabled(sb, PRJQUOTA))
4908                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
4909 #endif
4910         return 0;
4911 }
4912
4913 /* Helper function for writing quotas on sync - we need to start transaction
4914  * before quota file is locked for write. Otherwise the are possible deadlocks:
4915  * Process 1                         Process 2
4916  * ext4_create()                     quota_sync()
4917  *   jbd2_journal_start()                  write_dquot()
4918  *   dquot_initialize()                         down(dqio_mutex)
4919  *     down(dqio_mutex)                    jbd2_journal_start()
4920  *
4921  */
4922
4923 #ifdef CONFIG_QUOTA
4924
4925 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4926 {
4927         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
4928 }
4929
4930 static int ext4_write_dquot(struct dquot *dquot)
4931 {
4932         int ret, err;
4933         handle_t *handle;
4934         struct inode *inode;
4935
4936         inode = dquot_to_inode(dquot);
4937         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4938                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4939         if (IS_ERR(handle))
4940                 return PTR_ERR(handle);
4941         ret = dquot_commit(dquot);
4942         err = ext4_journal_stop(handle);
4943         if (!ret)
4944                 ret = err;
4945         return ret;
4946 }
4947
4948 static int ext4_acquire_dquot(struct dquot *dquot)
4949 {
4950         int ret, err;
4951         handle_t *handle;
4952
4953         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4954                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4955         if (IS_ERR(handle))
4956                 return PTR_ERR(handle);
4957         ret = dquot_acquire(dquot);
4958         err = ext4_journal_stop(handle);
4959         if (!ret)
4960                 ret = err;
4961         return ret;
4962 }
4963
4964 static int ext4_release_dquot(struct dquot *dquot)
4965 {
4966         int ret, err;
4967         handle_t *handle;
4968
4969         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4970                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4971         if (IS_ERR(handle)) {
4972                 /* Release dquot anyway to avoid endless cycle in dqput() */
4973                 dquot_release(dquot);
4974                 return PTR_ERR(handle);
4975         }
4976         ret = dquot_release(dquot);
4977         err = ext4_journal_stop(handle);
4978         if (!ret)
4979                 ret = err;
4980         return ret;
4981 }
4982
4983 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4984 {
4985         struct super_block *sb = dquot->dq_sb;
4986         struct ext4_sb_info *sbi = EXT4_SB(sb);
4987
4988         /* Are we journaling quotas? */
4989         if (ext4_has_feature_quota(sb) ||
4990             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
4991                 dquot_mark_dquot_dirty(dquot);
4992                 return ext4_write_dquot(dquot);
4993         } else {
4994                 return dquot_mark_dquot_dirty(dquot);
4995         }
4996 }
4997
4998 static int ext4_write_info(struct super_block *sb, int type)
4999 {
5000         int ret, err;
5001         handle_t *handle;
5002
5003         /* Data block + inode block */
5004         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5005         if (IS_ERR(handle))
5006                 return PTR_ERR(handle);
5007         ret = dquot_commit_info(sb, type);
5008         err = ext4_journal_stop(handle);
5009         if (!ret)
5010                 ret = err;
5011         return ret;
5012 }
5013
5014 /*
5015  * Turn on quotas during mount time - we need to find
5016  * the quota file and such...
5017  */
5018 static int ext4_quota_on_mount(struct super_block *sb, int type)
5019 {
5020         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5021                                         EXT4_SB(sb)->s_jquota_fmt, type);
5022 }
5023
5024 /*
5025  * Standard function to be called on quota_on
5026  */
5027 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5028                          struct path *path)
5029 {
5030         int err;
5031
5032         if (!test_opt(sb, QUOTA))
5033                 return -EINVAL;
5034
5035         /* Quotafile not on the same filesystem? */
5036         if (path->dentry->d_sb != sb)
5037                 return -EXDEV;
5038         /* Journaling quota? */
5039         if (EXT4_SB(sb)->s_qf_names[type]) {
5040                 /* Quotafile not in fs root? */
5041                 if (path->dentry->d_parent != sb->s_root)
5042                         ext4_msg(sb, KERN_WARNING,
5043                                 "Quota file not on filesystem root. "
5044                                 "Journaled quota will not work");
5045         }
5046
5047         /*
5048          * When we journal data on quota file, we have to flush journal to see
5049          * all updates to the file when we bypass pagecache...
5050          */
5051         if (EXT4_SB(sb)->s_journal &&
5052             ext4_should_journal_data(d_inode(path->dentry))) {
5053                 /*
5054                  * We don't need to lock updates but journal_flush() could
5055                  * otherwise be livelocked...
5056                  */
5057                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5058                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5059                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5060                 if (err)
5061                         return err;
5062         }
5063
5064         return dquot_quota_on(sb, type, format_id, path);
5065 }
5066
5067 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5068                              unsigned int flags)
5069 {
5070         int err;
5071         struct inode *qf_inode;
5072         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5073                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5074                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5075                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5076         };
5077
5078         BUG_ON(!ext4_has_feature_quota(sb));
5079
5080         if (!qf_inums[type])
5081                 return -EPERM;
5082
5083         qf_inode = ext4_iget(sb, qf_inums[type]);
5084         if (IS_ERR(qf_inode)) {
5085                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5086                 return PTR_ERR(qf_inode);
5087         }
5088
5089         /* Don't account quota for quota files to avoid recursion */
5090         qf_inode->i_flags |= S_NOQUOTA;
5091         err = dquot_enable(qf_inode, type, format_id, flags);
5092         iput(qf_inode);
5093
5094         return err;
5095 }
5096
5097 /* Enable usage tracking for all quota types. */
5098 static int ext4_enable_quotas(struct super_block *sb)
5099 {
5100         int type, err = 0;
5101         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5102                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5103                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5104                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5105         };
5106
5107         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5108         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5109                 if (qf_inums[type]) {
5110                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5111                                                 DQUOT_USAGE_ENABLED);
5112                         if (err) {
5113                                 ext4_warning(sb,
5114                                         "Failed to enable quota tracking "
5115                                         "(type=%d, err=%d). Please run "
5116                                         "e2fsck to fix.", type, err);
5117                                 return err;
5118                         }
5119                 }
5120         }
5121         return 0;
5122 }
5123
5124 static int ext4_quota_off(struct super_block *sb, int type)
5125 {
5126         struct inode *inode = sb_dqopt(sb)->files[type];
5127         handle_t *handle;
5128
5129         /* Force all delayed allocation blocks to be allocated.
5130          * Caller already holds s_umount sem */
5131         if (test_opt(sb, DELALLOC))
5132                 sync_filesystem(sb);
5133
5134         if (!inode)
5135                 goto out;
5136
5137         /* Update modification times of quota files when userspace can
5138          * start looking at them */
5139         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5140         if (IS_ERR(handle))
5141                 goto out;
5142         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5143         ext4_mark_inode_dirty(handle, inode);
5144         ext4_journal_stop(handle);
5145
5146 out:
5147         return dquot_quota_off(sb, type);
5148 }
5149
5150 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5151  * acquiring the locks... As quota files are never truncated and quota code
5152  * itself serializes the operations (and no one else should touch the files)
5153  * we don't have to be afraid of races */
5154 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5155                                size_t len, loff_t off)
5156 {
5157         struct inode *inode = sb_dqopt(sb)->files[type];
5158         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5159         int offset = off & (sb->s_blocksize - 1);
5160         int tocopy;
5161         size_t toread;
5162         struct buffer_head *bh;
5163         loff_t i_size = i_size_read(inode);
5164
5165         if (off > i_size)
5166                 return 0;
5167         if (off+len > i_size)
5168                 len = i_size-off;
5169         toread = len;
5170         while (toread > 0) {
5171                 tocopy = sb->s_blocksize - offset < toread ?
5172                                 sb->s_blocksize - offset : toread;
5173                 bh = ext4_bread(NULL, inode, blk, 0);
5174                 if (IS_ERR(bh))
5175                         return PTR_ERR(bh);
5176                 if (!bh)        /* A hole? */
5177                         memset(data, 0, tocopy);
5178                 else
5179                         memcpy(data, bh->b_data+offset, tocopy);
5180                 brelse(bh);
5181                 offset = 0;
5182                 toread -= tocopy;
5183                 data += tocopy;
5184                 blk++;
5185         }
5186         return len;
5187 }
5188
5189 /* Write to quotafile (we know the transaction is already started and has
5190  * enough credits) */
5191 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5192                                 const char *data, size_t len, loff_t off)
5193 {
5194         struct inode *inode = sb_dqopt(sb)->files[type];
5195         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5196         int err, offset = off & (sb->s_blocksize - 1);
5197         int retries = 0;
5198         struct buffer_head *bh;
5199         handle_t *handle = journal_current_handle();
5200
5201         if (EXT4_SB(sb)->s_journal && !handle) {
5202                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5203                         " cancelled because transaction is not started",
5204                         (unsigned long long)off, (unsigned long long)len);
5205                 return -EIO;
5206         }
5207         /*
5208          * Since we account only one data block in transaction credits,
5209          * then it is impossible to cross a block boundary.
5210          */
5211         if (sb->s_blocksize - offset < len) {
5212                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5213                         " cancelled because not block aligned",
5214                         (unsigned long long)off, (unsigned long long)len);
5215                 return -EIO;
5216         }
5217
5218         do {
5219                 bh = ext4_bread(handle, inode, blk,
5220                                 EXT4_GET_BLOCKS_CREATE |
5221                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5222         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5223                  ext4_should_retry_alloc(inode->i_sb, &retries));
5224         if (IS_ERR(bh))
5225                 return PTR_ERR(bh);
5226         if (!bh)
5227                 goto out;
5228         BUFFER_TRACE(bh, "get write access");
5229         err = ext4_journal_get_write_access(handle, bh);
5230         if (err) {
5231                 brelse(bh);
5232                 return err;
5233         }
5234         lock_buffer(bh);
5235         memcpy(bh->b_data+offset, data, len);
5236         flush_dcache_page(bh->b_page);
5237         unlock_buffer(bh);
5238         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5239         brelse(bh);
5240 out:
5241         if (inode->i_size < off + len) {
5242                 i_size_write(inode, off + len);
5243                 EXT4_I(inode)->i_disksize = inode->i_size;
5244                 ext4_mark_inode_dirty(handle, inode);
5245         }
5246         return len;
5247 }
5248
5249 #endif
5250
5251 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5252                        const char *dev_name, void *data)
5253 {
5254         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5255 }
5256
5257 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5258 static inline void register_as_ext2(void)
5259 {
5260         int err = register_filesystem(&ext2_fs_type);
5261         if (err)
5262                 printk(KERN_WARNING
5263                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5264 }
5265
5266 static inline void unregister_as_ext2(void)
5267 {
5268         unregister_filesystem(&ext2_fs_type);
5269 }
5270
5271 static inline int ext2_feature_set_ok(struct super_block *sb)
5272 {
5273         if (ext4_has_unknown_ext2_incompat_features(sb))
5274                 return 0;
5275         if (sb->s_flags & MS_RDONLY)
5276                 return 1;
5277         if (ext4_has_unknown_ext2_ro_compat_features(sb))
5278                 return 0;
5279         return 1;
5280 }
5281 #else
5282 static inline void register_as_ext2(void) { }
5283 static inline void unregister_as_ext2(void) { }
5284 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5285 #endif
5286
5287 static inline void register_as_ext3(void)
5288 {
5289         int err = register_filesystem(&ext3_fs_type);
5290         if (err)
5291                 printk(KERN_WARNING
5292                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5293 }
5294
5295 static inline void unregister_as_ext3(void)
5296 {
5297         unregister_filesystem(&ext3_fs_type);
5298 }
5299
5300 static inline int ext3_feature_set_ok(struct super_block *sb)
5301 {
5302         if (ext4_has_unknown_ext3_incompat_features(sb))
5303                 return 0;
5304         if (!ext4_has_feature_journal(sb))
5305                 return 0;
5306         if (sb->s_flags & MS_RDONLY)
5307                 return 1;
5308         if (ext4_has_unknown_ext3_ro_compat_features(sb))
5309                 return 0;
5310         return 1;
5311 }
5312
5313 static struct file_system_type ext4_fs_type = {
5314         .owner          = THIS_MODULE,
5315         .name           = "ext4",
5316         .mount          = ext4_mount,
5317         .kill_sb        = kill_block_super,
5318         .fs_flags       = FS_REQUIRES_DEV,
5319 };
5320 MODULE_ALIAS_FS("ext4");
5321
5322 /* Shared across all ext4 file systems */
5323 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5324 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5325
5326 static int __init ext4_init_fs(void)
5327 {
5328         int i, err;
5329
5330         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
5331         ext4_li_info = NULL;
5332         mutex_init(&ext4_li_mtx);
5333
5334         /* Build-time check for flags consistency */
5335         ext4_check_flag_values();
5336
5337         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5338                 mutex_init(&ext4__aio_mutex[i]);
5339                 init_waitqueue_head(&ext4__ioend_wq[i]);
5340         }
5341
5342         err = ext4_init_es();
5343         if (err)
5344                 return err;
5345
5346         err = ext4_init_pageio();
5347         if (err)
5348                 goto out5;
5349
5350         err = ext4_init_system_zone();
5351         if (err)
5352                 goto out4;
5353
5354         err = ext4_init_sysfs();
5355         if (err)
5356                 goto out3;
5357
5358         err = ext4_init_mballoc();
5359         if (err)
5360                 goto out2;
5361         else
5362                 ext4_mballoc_ready = 1;
5363         err = init_inodecache();
5364         if (err)
5365                 goto out1;
5366         register_as_ext3();
5367         register_as_ext2();
5368         err = register_filesystem(&ext4_fs_type);
5369         if (err)
5370                 goto out;
5371
5372         return 0;
5373 out:
5374         unregister_as_ext2();
5375         unregister_as_ext3();
5376         destroy_inodecache();
5377 out1:
5378         ext4_mballoc_ready = 0;
5379         ext4_exit_mballoc();
5380 out2:
5381         ext4_exit_sysfs();
5382 out3:
5383         ext4_exit_system_zone();
5384 out4:
5385         ext4_exit_pageio();
5386 out5:
5387         ext4_exit_es();
5388
5389         return err;
5390 }
5391
5392 static void __exit ext4_exit_fs(void)
5393 {
5394         ext4_exit_crypto();
5395         ext4_destroy_lazyinit_thread();
5396         unregister_as_ext2();
5397         unregister_as_ext3();
5398         unregister_filesystem(&ext4_fs_type);
5399         destroy_inodecache();
5400         ext4_exit_mballoc();
5401         ext4_exit_sysfs();
5402         ext4_exit_system_zone();
5403         ext4_exit_pageio();
5404         ext4_exit_es();
5405 }
5406
5407 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5408 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5409 MODULE_LICENSE("GPL");
5410 module_init(ext4_init_fs)
5411 module_exit(ext4_exit_fs)