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/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/cleancache.h>
  42 #include <asm/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"       /* Needed for trace points definition */
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53
  54 #define CREATE_TRACE_POINTS
  55 #include <trace/events/ext4.h>
  56
  57 static struct proc_dir_entry *ext4_proc_root;
  58 static struct kset *ext4_kset;
  59 static struct ext4_lazy_init *ext4_li_info;
  60 static struct mutex ext4_li_mtx;
  61 static struct ext4_features *ext4_feat;
  62 static int ext4_mballoc_ready;
  63
  64 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  65                              unsigned long journal_devnum);
  66 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  67 static int ext4_commit_super(struct super_block *sb, int sync);
  68 static void ext4_mark_recovery_complete(struct super_block *sb,
  69                                         struct ext4_super_block *es);
  70 static void ext4_clear_journal_err(struct super_block *sb,
  71                                    struct ext4_super_block *es);
  72 static int ext4_sync_fs(struct super_block *sb, int wait);
  73 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  74 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  75 static int ext4_unfreeze(struct super_block *sb);
  76 static int ext4_freeze(struct super_block *sb);
  77 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  78                        const char *dev_name, void *data);
  79 static inline int ext2_feature_set_ok(struct super_block *sb);
  80 static inline int ext3_feature_set_ok(struct super_block *sb);
  81 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  82 static void ext4_destroy_lazyinit_thread(void);
  83 static void ext4_unregister_li_request(struct super_block *sb);
  84 static void ext4_clear_request_list(void);
  85 static int ext4_reserve_clusters(struct ext4_sb_info *, ext4_fsblk_t);
  86
  87 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  88 static struct file_system_type ext2_fs_type = {
  89         .owner          = THIS_MODULE,
  90         .name           = "ext2",
  91         .mount          = ext4_mount,
  92         .kill_sb        = kill_block_super,
  93         .fs_flags       = FS_REQUIRES_DEV,
  94 };
  95 MODULE_ALIAS_FS("ext2");
  96 MODULE_ALIAS("ext2");
  97 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  98 #else
  99 #define IS_EXT2_SB(sb) (0)
 100 #endif
 101
 102
 103 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 104 static struct file_system_type ext3_fs_type = {
 105         .owner          = THIS_MODULE,
 106         .name           = "ext3",
 107         .mount          = ext4_mount,
 108         .kill_sb        = kill_block_super,
 109         .fs_flags       = FS_REQUIRES_DEV,
 110 };
 111 MODULE_ALIAS_FS("ext3");
 112 MODULE_ALIAS("ext3");
 113 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 114 #else
 115 #define IS_EXT3_SB(sb) (0)
 116 #endif
 117
 118 static int ext4_verify_csum_type(struct super_block *sb,
 119                                  struct ext4_super_block *es)
 120 {
 121         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 122                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 123                 return 1;
 124
 125         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 126 }
 127
 128 static __le32 ext4_superblock_csum(struct super_block *sb,
 129                                    struct ext4_super_block *es)
 130 {
 131         struct ext4_sb_info *sbi = EXT4_SB(sb);
 132         int offset = offsetof(struct ext4_super_block, s_checksum);
 133         __u32 csum;
 134
 135         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 136
 137         return cpu_to_le32(csum);
 138 }
 139
 140 static int ext4_superblock_csum_verify(struct super_block *sb,
 141                                        struct ext4_super_block *es)
 142 {
 143         if (!ext4_has_metadata_csum(sb))
 144                 return 1;
 145
 146         return es->s_checksum == ext4_superblock_csum(sb, es);
 147 }
 148
 149 void ext4_superblock_csum_set(struct super_block *sb)
 150 {
 151         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 152
 153         if (!ext4_has_metadata_csum(sb))
 154                 return;
 155
 156         es->s_checksum = ext4_superblock_csum(sb, es);
 157 }
 158
 159 void *ext4_kvmalloc(size_t size, gfp_t flags)
 160 {
 161         void *ret;
 162
 163         ret = kmalloc(size, flags | __GFP_NOWARN);
 164         if (!ret)
 165                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 166         return ret;
 167 }
 168
 169 void *ext4_kvzalloc(size_t size, gfp_t flags)
 170 {
 171         void *ret;
 172
 173         ret = kzalloc(size, flags | __GFP_NOWARN);
 174         if (!ret)
 175                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 176         return ret;
 177 }
 178
 179 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 180                                struct ext4_group_desc *bg)
 181 {
 182         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 183                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 184                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 185 }
 186
 187 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 188                                struct ext4_group_desc *bg)
 189 {
 190         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 191                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 192                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 193 }
 194
 195 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 196                               struct ext4_group_desc *bg)
 197 {
 198         return le32_to_cpu(bg->bg_inode_table_lo) |
 199                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 200                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 201 }
 202
 203 __u32 ext4_free_group_clusters(struct super_block *sb,
 204                                struct ext4_group_desc *bg)
 205 {
 206         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 207                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 208                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 209 }
 210
 211 __u32 ext4_free_inodes_count(struct super_block *sb,
 212                               struct ext4_group_desc *bg)
 213 {
 214         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 215                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 216                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 217 }
 218
 219 __u32 ext4_used_dirs_count(struct super_block *sb,
 220                               struct ext4_group_desc *bg)
 221 {
 222         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 223                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 224                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 225 }
 226
 227 __u32 ext4_itable_unused_count(struct super_block *sb,
 228                               struct ext4_group_desc *bg)
 229 {
 230         return le16_to_cpu(bg->bg_itable_unused_lo) |
 231                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 232                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 233 }
 234
 235 void ext4_block_bitmap_set(struct super_block *sb,
 236                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 237 {
 238         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 239         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 240                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 241 }
 242
 243 void ext4_inode_bitmap_set(struct super_block *sb,
 244                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 245 {
 246         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 247         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 248                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 249 }
 250
 251 void ext4_inode_table_set(struct super_block *sb,
 252                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 253 {
 254         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 255         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 256                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 257 }
 258
 259 void ext4_free_group_clusters_set(struct super_block *sb,
 260                                   struct ext4_group_desc *bg, __u32 count)
 261 {
 262         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 263         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 264                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 265 }
 266
 267 void ext4_free_inodes_set(struct super_block *sb,
 268                           struct ext4_group_desc *bg, __u32 count)
 269 {
 270         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 271         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 272                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 273 }
 274
 275 void ext4_used_dirs_set(struct super_block *sb,
 276                           struct ext4_group_desc *bg, __u32 count)
 277 {
 278         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 279         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 280                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 281 }
 282
 283 void ext4_itable_unused_set(struct super_block *sb,
 284                           struct ext4_group_desc *bg, __u32 count)
 285 {
 286         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 287         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 288                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 289 }
 290
 291
 292 static void __save_error_info(struct super_block *sb, const char *func,
 293                             unsigned int line)
 294 {
 295         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 296
 297         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 298         if (bdev_read_only(sb->s_bdev))
 299                 return;
 300         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 301         es->s_last_error_time = cpu_to_le32(get_seconds());
 302         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 303         es->s_last_error_line = cpu_to_le32(line);
 304         if (!es->s_first_error_time) {
 305                 es->s_first_error_time = es->s_last_error_time;
 306                 strncpy(es->s_first_error_func, func,
 307                         sizeof(es->s_first_error_func));
 308                 es->s_first_error_line = cpu_to_le32(line);
 309                 es->s_first_error_ino = es->s_last_error_ino;
 310                 es->s_first_error_block = es->s_last_error_block;
 311         }
 312         /*
 313          * Start the daily error reporting function if it hasn't been
 314          * started already
 315          */
 316         if (!es->s_error_count)
 317                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 318         le32_add_cpu(&es->s_error_count, 1);
 319 }
 320
 321 static void save_error_info(struct super_block *sb, const char *func,
 322                             unsigned int line)
 323 {
 324         __save_error_info(sb, func, line);
 325         ext4_commit_super(sb, 1);
 326 }
 327
 328 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 329 {
 330         struct super_block              *sb = journal->j_private;
 331         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 332         int                             error = is_journal_aborted(journal);
 333         struct ext4_journal_cb_entry    *jce;
 334
 335         BUG_ON(txn->t_state == T_FINISHED);
 336         spin_lock(&sbi->s_md_lock);
 337         while (!list_empty(&txn->t_private_list)) {
 338                 jce = list_entry(txn->t_private_list.next,
 339                                  struct ext4_journal_cb_entry, jce_list);
 340                 list_del_init(&jce->jce_list);
 341                 spin_unlock(&sbi->s_md_lock);
 342                 jce->jce_func(sb, jce, error);
 343                 spin_lock(&sbi->s_md_lock);
 344         }
 345         spin_unlock(&sbi->s_md_lock);
 346 }
 347
 348 /* Deal with the reporting of failure conditions on a filesystem such as
 349  * inconsistencies detected or read IO failures.
 350  *
 351  * On ext2, we can store the error state of the filesystem in the
 352  * superblock.  That is not possible on ext4, because we may have other
 353  * write ordering constraints on the superblock which prevent us from
 354  * writing it out straight away; and given that the journal is about to
 355  * be aborted, we can't rely on the current, or future, transactions to
 356  * write out the superblock safely.
 357  *
 358  * We'll just use the jbd2_journal_abort() error code to record an error in
 359  * the journal instead.  On recovery, the journal will complain about
 360  * that error until we've noted it down and cleared it.
 361  */
 362
 363 static void ext4_handle_error(struct super_block *sb)
 364 {
 365         if (sb->s_flags & MS_RDONLY)
 366                 return;
 367
 368         if (!test_opt(sb, ERRORS_CONT)) {
 369                 journal_t *journal = EXT4_SB(sb)->s_journal;
 370
 371                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 372                 if (journal)
 373                         jbd2_journal_abort(journal, -EIO);
 374         }
 375         if (test_opt(sb, ERRORS_RO)) {
 376                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 377                 /*
 378                  * Make sure updated value of ->s_mount_flags will be visible
 379                  * before ->s_flags update
 380                  */
 381                 smp_wmb();
 382                 sb->s_flags |= MS_RDONLY;
 383         }
 384         if (test_opt(sb, ERRORS_PANIC))
 385                 panic("EXT4-fs (device %s): panic forced after error\n",
 386                         sb->s_id);
 387 }
 388
 389 #define ext4_error_ratelimit(sb)                                        \
 390                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 391                              "EXT4-fs error")
 392
 393 void __ext4_error(struct super_block *sb, const char *function,
 394                   unsigned int line, const char *fmt, ...)
 395 {
 396         struct va_format vaf;
 397         va_list args;
 398
 399         if (ext4_error_ratelimit(sb)) {
 400                 va_start(args, fmt);
 401                 vaf.fmt = fmt;
 402                 vaf.va = &args;
 403                 printk(KERN_CRIT
 404                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 405                        sb->s_id, function, line, current->comm, &vaf);
 406                 va_end(args);
 407         }
 408         save_error_info(sb, function, line);
 409         ext4_handle_error(sb);
 410 }
 411
 412 void __ext4_error_inode(struct inode *inode, const char *function,
 413                         unsigned int line, ext4_fsblk_t block,
 414                         const char *fmt, ...)
 415 {
 416         va_list args;
 417         struct va_format vaf;
 418         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 419
 420         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 421         es->s_last_error_block = cpu_to_le64(block);
 422         if (ext4_error_ratelimit(inode->i_sb)) {
 423                 va_start(args, fmt);
 424                 vaf.fmt = fmt;
 425                 vaf.va = &args;
 426                 if (block)
 427                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 428                                "inode #%lu: block %llu: comm %s: %pV\n",
 429                                inode->i_sb->s_id, function, line, inode->i_ino,
 430                                block, current->comm, &vaf);
 431                 else
 432                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 433                                "inode #%lu: comm %s: %pV\n",
 434                                inode->i_sb->s_id, function, line, inode->i_ino,
 435                                current->comm, &vaf);
 436                 va_end(args);
 437         }
 438         save_error_info(inode->i_sb, function, line);
 439         ext4_handle_error(inode->i_sb);
 440 }
 441
 442 void __ext4_error_file(struct file *file, const char *function,
 443                        unsigned int line, ext4_fsblk_t block,
 444                        const char *fmt, ...)
 445 {
 446         va_list args;
 447         struct va_format vaf;
 448         struct ext4_super_block *es;
 449         struct inode *inode = file_inode(file);
 450         char pathname[80], *path;
 451
 452         es = EXT4_SB(inode->i_sb)->s_es;
 453         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 454         if (ext4_error_ratelimit(inode->i_sb)) {
 455                 path = file_path(file, pathname, sizeof(pathname));
 456                 if (IS_ERR(path))
 457                         path = "(unknown)";
 458                 va_start(args, fmt);
 459                 vaf.fmt = fmt;
 460                 vaf.va = &args;
 461                 if (block)
 462                         printk(KERN_CRIT
 463                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 464                                "block %llu: comm %s: path %s: %pV\n",
 465                                inode->i_sb->s_id, function, line, inode->i_ino,
 466                                block, current->comm, path, &vaf);
 467                 else
 468                         printk(KERN_CRIT
 469                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 470                                "comm %s: path %s: %pV\n",
 471                                inode->i_sb->s_id, function, line, inode->i_ino,
 472                                current->comm, path, &vaf);
 473                 va_end(args);
 474         }
 475         save_error_info(inode->i_sb, function, line);
 476         ext4_handle_error(inode->i_sb);
 477 }
 478
 479 const char *ext4_decode_error(struct super_block *sb, int errno,
 480                               char nbuf[16])
 481 {
 482         char *errstr = NULL;
 483
 484         switch (errno) {
 485         case -EIO:
 486                 errstr = "IO failure";
 487                 break;
 488         case -ENOMEM:
 489                 errstr = "Out of memory";
 490                 break;
 491         case -EROFS:
 492                 if (!sb || (EXT4_SB(sb)->s_journal &&
 493                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 494                         errstr = "Journal has aborted";
 495                 else
 496                         errstr = "Readonly filesystem";
 497                 break;
 498         default:
 499                 /* If the caller passed in an extra buffer for unknown
 500                  * errors, textualise them now.  Else we just return
 501                  * NULL. */
 502                 if (nbuf) {
 503                         /* Check for truncated error codes... */
 504                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 505                                 errstr = nbuf;
 506                 }
 507                 break;
 508         }
 509
 510         return errstr;
 511 }
 512
 513 /* __ext4_std_error decodes expected errors from journaling functions
 514  * automatically and invokes the appropriate error response.  */
 515
 516 void __ext4_std_error(struct super_block *sb, const char *function,
 517                       unsigned int line, int errno)
 518 {
 519         char nbuf[16];
 520         const char *errstr;
 521
 522         /* Special case: if the error is EROFS, and we're not already
 523          * inside a transaction, then there's really no point in logging
 524          * an error. */
 525         if (errno == -EROFS && journal_current_handle() == NULL &&
 526             (sb->s_flags & MS_RDONLY))
 527                 return;
 528
 529         if (ext4_error_ratelimit(sb)) {
 530                 errstr = ext4_decode_error(sb, errno, nbuf);
 531                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 532                        sb->s_id, function, line, errstr);
 533         }
 534
 535         save_error_info(sb, function, line);
 536         ext4_handle_error(sb);
 537 }
 538
 539 /*
 540  * ext4_abort is a much stronger failure handler than ext4_error.  The
 541  * abort function may be used to deal with unrecoverable failures such
 542  * as journal IO errors or ENOMEM at a critical moment in log management.
 543  *
 544  * We unconditionally force the filesystem into an ABORT|READONLY state,
 545  * unless the error response on the fs has been set to panic in which
 546  * case we take the easy way out and panic immediately.
 547  */
 548
 549 void __ext4_abort(struct super_block *sb, const char *function,
 550                 unsigned int line, const char *fmt, ...)
 551 {
 552         va_list args;
 553
 554         save_error_info(sb, function, line);
 555         va_start(args, fmt);
 556         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 557                function, line);
 558         vprintk(fmt, args);
 559         printk("\n");
 560         va_end(args);
 561
 562         if ((sb->s_flags & MS_RDONLY) == 0) {
 563                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 564                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 565                 /*
 566                  * Make sure updated value of ->s_mount_flags will be visible
 567                  * before ->s_flags update
 568                  */
 569                 smp_wmb();
 570                 sb->s_flags |= MS_RDONLY;
 571                 if (EXT4_SB(sb)->s_journal)
 572                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 573                 save_error_info(sb, function, line);
 574         }
 575         if (test_opt(sb, ERRORS_PANIC))
 576                 panic("EXT4-fs panic from previous error\n");
 577 }
 578
 579 void __ext4_msg(struct super_block *sb,
 580                 const char *prefix, const char *fmt, ...)
 581 {
 582         struct va_format vaf;
 583         va_list args;
 584
 585         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 586                 return;
 587
 588         va_start(args, fmt);
 589         vaf.fmt = fmt;
 590         vaf.va = &args;
 591         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 592         va_end(args);
 593 }
 594
 595 #define ext4_warning_ratelimit(sb)                                      \
 596                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
 597                              "EXT4-fs warning")
 598
 599 void __ext4_warning(struct super_block *sb, const char *function,
 600                     unsigned int line, const char *fmt, ...)
 601 {
 602         struct va_format vaf;
 603         va_list args;
 604
 605         if (!ext4_warning_ratelimit(sb))
 606                 return;
 607
 608         va_start(args, fmt);
 609         vaf.fmt = fmt;
 610         vaf.va = &args;
 611         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 612                sb->s_id, function, line, &vaf);
 613         va_end(args);
 614 }
 615
 616 void __ext4_warning_inode(const struct inode *inode, const char *function,
 617                           unsigned int line, const char *fmt, ...)
 618 {
 619         struct va_format vaf;
 620         va_list args;
 621
 622         if (!ext4_warning_ratelimit(inode->i_sb))
 623                 return;
 624
 625         va_start(args, fmt);
 626         vaf.fmt = fmt;
 627         vaf.va = &args;
 628         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 629                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 630                function, line, inode->i_ino, current->comm, &vaf);
 631         va_end(args);
 632 }
 633
 634 void __ext4_grp_locked_error(const char *function, unsigned int line,
 635                              struct super_block *sb, ext4_group_t grp,
 636                              unsigned long ino, ext4_fsblk_t block,
 637                              const char *fmt, ...)
 638 __releases(bitlock)
 639 __acquires(bitlock)
 640 {
 641         struct va_format vaf;
 642         va_list args;
 643         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 644
 645         es->s_last_error_ino = cpu_to_le32(ino);
 646         es->s_last_error_block = cpu_to_le64(block);
 647         __save_error_info(sb, function, line);
 648
 649         if (ext4_error_ratelimit(sb)) {
 650                 va_start(args, fmt);
 651                 vaf.fmt = fmt;
 652                 vaf.va = &args;
 653                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 654                        sb->s_id, function, line, grp);
 655                 if (ino)
 656                         printk(KERN_CONT "inode %lu: ", ino);
 657                 if (block)
 658                         printk(KERN_CONT "block %llu:",
 659                                (unsigned long long) block);
 660                 printk(KERN_CONT "%pV\n", &vaf);
 661                 va_end(args);
 662         }
 663
 664         if (test_opt(sb, ERRORS_CONT)) {
 665                 ext4_commit_super(sb, 0);
 666                 return;
 667         }
 668
 669         ext4_unlock_group(sb, grp);
 670         ext4_handle_error(sb);
 671         /*
 672          * We only get here in the ERRORS_RO case; relocking the group
 673          * may be dangerous, but nothing bad will happen since the
 674          * filesystem will have already been marked read/only and the
 675          * journal has been aborted.  We return 1 as a hint to callers
 676          * who might what to use the return value from
 677          * ext4_grp_locked_error() to distinguish between the
 678          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 679          * aggressively from the ext4 function in question, with a
 680          * more appropriate error code.
 681          */
 682         ext4_lock_group(sb, grp);
 683         return;
 684 }
 685
 686 void ext4_update_dynamic_rev(struct super_block *sb)
 687 {
 688         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 689
 690         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 691                 return;
 692
 693         ext4_warning(sb,
 694                      "updating to rev %d because of new feature flag, "
 695                      "running e2fsck is recommended",
 696                      EXT4_DYNAMIC_REV);
 697
 698         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 699         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 700         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 701         /* leave es->s_feature_*compat flags alone */
 702         /* es->s_uuid will be set by e2fsck if empty */
 703
 704         /*
 705          * The rest of the superblock fields should be zero, and if not it
 706          * means they are likely already in use, so leave them alone.  We
 707          * can leave it up to e2fsck to clean up any inconsistencies there.
 708          */
 709 }
 710
 711 /*
 712  * Open the external journal device
 713  */
 714 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 715 {
 716         struct block_device *bdev;
 717         char b[BDEVNAME_SIZE];
 718
 719         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 720         if (IS_ERR(bdev))
 721                 goto fail;
 722         return bdev;
 723
 724 fail:
 725         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 726                         __bdevname(dev, b), PTR_ERR(bdev));
 727         return NULL;
 728 }
 729
 730 /*
 731  * Release the journal device
 732  */
 733 static void ext4_blkdev_put(struct block_device *bdev)
 734 {
 735         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 736 }
 737
 738 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 739 {
 740         struct block_device *bdev;
 741         bdev = sbi->journal_bdev;
 742         if (bdev) {
 743                 ext4_blkdev_put(bdev);
 744                 sbi->journal_bdev = NULL;
 745         }
 746 }
 747
 748 static inline struct inode *orphan_list_entry(struct list_head *l)
 749 {
 750         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 751 }
 752
 753 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 754 {
 755         struct list_head *l;
 756
 757         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 758                  le32_to_cpu(sbi->s_es->s_last_orphan));
 759
 760         printk(KERN_ERR "sb_info orphan list:\n");
 761         list_for_each(l, &sbi->s_orphan) {
 762                 struct inode *inode = orphan_list_entry(l);
 763                 printk(KERN_ERR "  "
 764                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 765                        inode->i_sb->s_id, inode->i_ino, inode,
 766                        inode->i_mode, inode->i_nlink,
 767                        NEXT_ORPHAN(inode));
 768         }
 769 }
 770
 771 static void ext4_put_super(struct super_block *sb)
 772 {
 773         struct ext4_sb_info *sbi = EXT4_SB(sb);
 774         struct ext4_super_block *es = sbi->s_es;
 775         int i, err;
 776
 777         ext4_unregister_li_request(sb);
 778         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 779
 780         flush_workqueue(sbi->rsv_conversion_wq);
 781         destroy_workqueue(sbi->rsv_conversion_wq);
 782
 783         if (sbi->s_journal) {
 784                 err = jbd2_journal_destroy(sbi->s_journal);
 785                 sbi->s_journal = NULL;
 786                 if (err < 0)
 787                         ext4_abort(sb, "Couldn't clean up the journal");
 788         }
 789
 790         ext4_es_unregister_shrinker(sbi);
 791         del_timer_sync(&sbi->s_err_report);
 792         ext4_release_system_zone(sb);
 793         ext4_mb_release(sb);
 794         ext4_ext_release(sb);
 795         ext4_xattr_put_super(sb);
 796
 797         if (!(sb->s_flags & MS_RDONLY)) {
 798                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 799                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 800         }
 801         if (!(sb->s_flags & MS_RDONLY))
 802                 ext4_commit_super(sb, 1);
 803
 804         if (sbi->s_proc) {
 805                 remove_proc_entry("options", sbi->s_proc);
 806                 remove_proc_entry(sb->s_id, ext4_proc_root);
 807         }
 808         kobject_del(&sbi->s_kobj);
 809
 810         for (i = 0; i < sbi->s_gdb_count; i++)
 811                 brelse(sbi->s_group_desc[i]);
 812         kvfree(sbi->s_group_desc);
 813         kvfree(sbi->s_flex_groups);
 814         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 815         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 816         percpu_counter_destroy(&sbi->s_dirs_counter);
 817         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 818         brelse(sbi->s_sbh);
 819 #ifdef CONFIG_QUOTA
 820         for (i = 0; i < EXT4_MAXQUOTAS; i++)
 821                 kfree(sbi->s_qf_names[i]);
 822 #endif
 823
 824         /* Debugging code just in case the in-memory inode orphan list
 825          * isn't empty.  The on-disk one can be non-empty if we've
 826          * detected an error and taken the fs readonly, but the
 827          * in-memory list had better be clean by this point. */
 828         if (!list_empty(&sbi->s_orphan))
 829                 dump_orphan_list(sb, sbi);
 830         J_ASSERT(list_empty(&sbi->s_orphan));
 831
 832         sync_blockdev(sb->s_bdev);
 833         invalidate_bdev(sb->s_bdev);
 834         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 835                 /*
 836                  * Invalidate the journal device's buffers.  We don't want them
 837                  * floating about in memory - the physical journal device may
 838                  * hotswapped, and it breaks the `ro-after' testing code.
 839                  */
 840                 sync_blockdev(sbi->journal_bdev);
 841                 invalidate_bdev(sbi->journal_bdev);
 842                 ext4_blkdev_remove(sbi);
 843         }
 844         if (sbi->s_mb_cache) {
 845                 ext4_xattr_destroy_cache(sbi->s_mb_cache);
 846                 sbi->s_mb_cache = NULL;
 847         }
 848         if (sbi->s_mmp_tsk)
 849                 kthread_stop(sbi->s_mmp_tsk);
 850         sb->s_fs_info = NULL;
 851         /*
 852          * Now that we are completely done shutting down the
 853          * superblock, we need to actually destroy the kobject.
 854          */
 855         kobject_put(&sbi->s_kobj);
 856         wait_for_completion(&sbi->s_kobj_unregister);
 857         if (sbi->s_chksum_driver)
 858                 crypto_free_shash(sbi->s_chksum_driver);
 859         kfree(sbi->s_blockgroup_lock);
 860         kfree(sbi);
 861 }
 862
 863 static struct kmem_cache *ext4_inode_cachep;
 864
 865 /*
 866  * Called inside transaction, so use GFP_NOFS
 867  */
 868 static struct inode *ext4_alloc_inode(struct super_block *sb)
 869 {
 870         struct ext4_inode_info *ei;
 871
 872         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 873         if (!ei)
 874                 return NULL;
 875
 876         ei->vfs_inode.i_version = 1;
 877         spin_lock_init(&ei->i_raw_lock);
 878         INIT_LIST_HEAD(&ei->i_prealloc_list);
 879         spin_lock_init(&ei->i_prealloc_lock);
 880         ext4_es_init_tree(&ei->i_es_tree);
 881         rwlock_init(&ei->i_es_lock);
 882         INIT_LIST_HEAD(&ei->i_es_list);
 883         ei->i_es_all_nr = 0;
 884         ei->i_es_shk_nr = 0;
 885         ei->i_es_shrink_lblk = 0;
 886         ei->i_reserved_data_blocks = 0;
 887         ei->i_reserved_meta_blocks = 0;
 888         ei->i_allocated_meta_blocks = 0;
 889         ei->i_da_metadata_calc_len = 0;
 890         ei->i_da_metadata_calc_last_lblock = 0;
 891         spin_lock_init(&(ei->i_block_reservation_lock));
 892 #ifdef CONFIG_QUOTA
 893         ei->i_reserved_quota = 0;
 894         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
 895 #endif
 896         ei->jinode = NULL;
 897         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
 898         spin_lock_init(&ei->i_completed_io_lock);
 899         ei->i_sync_tid = 0;
 900         ei->i_datasync_tid = 0;
 901         atomic_set(&ei->i_ioend_count, 0);
 902         atomic_set(&ei->i_unwritten, 0);
 903         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
 904 #ifdef CONFIG_EXT4_FS_ENCRYPTION
 905         ei->i_crypt_info = NULL;
 906 #endif
 907         return &ei->vfs_inode;
 908 }
 909
 910 static int ext4_drop_inode(struct inode *inode)
 911 {
 912         int drop = generic_drop_inode(inode);
 913
 914         trace_ext4_drop_inode(inode, drop);
 915         return drop;
 916 }
 917
 918 static void ext4_i_callback(struct rcu_head *head)
 919 {
 920         struct inode *inode = container_of(head, struct inode, i_rcu);
 921         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 922 }
 923
 924 static void ext4_destroy_inode(struct inode *inode)
 925 {
 926         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 927                 ext4_msg(inode->i_sb, KERN_ERR,
 928                          "Inode %lu (%p): orphan list check failed!",
 929                          inode->i_ino, EXT4_I(inode));
 930                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 931                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 932                                 true);
 933                 dump_stack();
 934         }
 935         call_rcu(&inode->i_rcu, ext4_i_callback);
 936 }
 937
 938 static void init_once(void *foo)
 939 {
 940         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 941
 942         INIT_LIST_HEAD(&ei->i_orphan);
 943         init_rwsem(&ei->xattr_sem);
 944         init_rwsem(&ei->i_data_sem);
 945         inode_init_once(&ei->vfs_inode);
 946 }
 947
 948 static int __init init_inodecache(void)
 949 {
 950         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 951                                              sizeof(struct ext4_inode_info),
 952                                              0, (SLAB_RECLAIM_ACCOUNT|
 953                                                 SLAB_MEM_SPREAD),
 954                                              init_once);
 955         if (ext4_inode_cachep == NULL)
 956                 return -ENOMEM;
 957         return 0;
 958 }
 959
 960 static void destroy_inodecache(void)
 961 {
 962         /*
 963          * Make sure all delayed rcu free inodes are flushed before we
 964          * destroy cache.
 965          */
 966         rcu_barrier();
 967         kmem_cache_destroy(ext4_inode_cachep);
 968 }
 969
 970 void ext4_clear_inode(struct inode *inode)
 971 {
 972         invalidate_inode_buffers(inode);
 973         clear_inode(inode);
 974         dquot_drop(inode);
 975         ext4_discard_preallocations(inode);
 976         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
 977         if (EXT4_I(inode)->jinode) {
 978                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 979                                                EXT4_I(inode)->jinode);
 980                 jbd2_free_inode(EXT4_I(inode)->jinode);
 981                 EXT4_I(inode)->jinode = NULL;
 982         }
 983 #ifdef CONFIG_EXT4_FS_ENCRYPTION
 984         if (EXT4_I(inode)->i_crypt_info)
 985                 ext4_free_encryption_info(inode, EXT4_I(inode)->i_crypt_info);
 986 #endif
 987 }
 988
 989 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 990                                         u64 ino, u32 generation)
 991 {
 992         struct inode *inode;
 993
 994         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 995                 return ERR_PTR(-ESTALE);
 996         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 997                 return ERR_PTR(-ESTALE);
 998
 999         /* iget isn't really right if the inode is currently unallocated!!
1000          *
1001          * ext4_read_inode will return a bad_inode if the inode had been
1002          * deleted, so we should be safe.
1003          *
1004          * Currently we don't know the generation for parent directory, so
1005          * a generation of 0 means "accept any"
1006          */
1007         inode = ext4_iget_normal(sb, ino);
1008         if (IS_ERR(inode))
1009                 return ERR_CAST(inode);
1010         if (generation && inode->i_generation != generation) {
1011                 iput(inode);
1012                 return ERR_PTR(-ESTALE);
1013         }
1014
1015         return inode;
1016 }
1017
1018 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1019                                         int fh_len, int fh_type)
1020 {
1021         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1022                                     ext4_nfs_get_inode);
1023 }
1024
1025 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1026                                         int fh_len, int fh_type)
1027 {
1028         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1029                                     ext4_nfs_get_inode);
1030 }
1031
1032 /*
1033  * Try to release metadata pages (indirect blocks, directories) which are
1034  * mapped via the block device.  Since these pages could have journal heads
1035  * which would prevent try_to_free_buffers() from freeing them, we must use
1036  * jbd2 layer's try_to_free_buffers() function to release them.
1037  */
1038 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1039                                  gfp_t wait)
1040 {
1041         journal_t *journal = EXT4_SB(sb)->s_journal;
1042
1043         WARN_ON(PageChecked(page));
1044         if (!page_has_buffers(page))
1045                 return 0;
1046         if (journal)
1047                 return jbd2_journal_try_to_free_buffers(journal, page,
1048                                                         wait & ~__GFP_WAIT);
1049         return try_to_free_buffers(page);
1050 }
1051
1052 #ifdef CONFIG_QUOTA
1053 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1054 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1055
1056 static int ext4_write_dquot(struct dquot *dquot);
1057 static int ext4_acquire_dquot(struct dquot *dquot);
1058 static int ext4_release_dquot(struct dquot *dquot);
1059 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1060 static int ext4_write_info(struct super_block *sb, int type);
1061 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1062                          struct path *path);
1063 static int ext4_quota_off(struct super_block *sb, int type);
1064 static int ext4_quota_on_mount(struct super_block *sb, int type);
1065 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1066                                size_t len, loff_t off);
1067 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1068                                 const char *data, size_t len, loff_t off);
1069 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1070                              unsigned int flags);
1071 static int ext4_enable_quotas(struct super_block *sb);
1072
1073 static struct dquot **ext4_get_dquots(struct inode *inode)
1074 {
1075         return EXT4_I(inode)->i_dquot;
1076 }
1077
1078 static const struct dquot_operations ext4_quota_operations = {
1079         .get_reserved_space = ext4_get_reserved_space,
1080         .write_dquot    = ext4_write_dquot,
1081         .acquire_dquot  = ext4_acquire_dquot,
1082         .release_dquot  = ext4_release_dquot,
1083         .mark_dirty     = ext4_mark_dquot_dirty,
1084         .write_info     = ext4_write_info,
1085         .alloc_dquot    = dquot_alloc,
1086         .destroy_dquot  = dquot_destroy,
1087 };
1088
1089 static const struct quotactl_ops ext4_qctl_operations = {
1090         .quota_on       = ext4_quota_on,
1091         .quota_off      = ext4_quota_off,
1092         .quota_sync     = dquot_quota_sync,
1093         .get_state      = dquot_get_state,
1094         .set_info       = dquot_set_dqinfo,
1095         .get_dqblk      = dquot_get_dqblk,
1096         .set_dqblk      = dquot_set_dqblk
1097 };
1098 #endif
1099
1100 static const struct super_operations ext4_sops = {
1101         .alloc_inode    = ext4_alloc_inode,
1102         .destroy_inode  = ext4_destroy_inode,
1103         .write_inode    = ext4_write_inode,
1104         .dirty_inode    = ext4_dirty_inode,
1105         .drop_inode     = ext4_drop_inode,
1106         .evict_inode    = ext4_evict_inode,
1107         .put_super      = ext4_put_super,
1108         .sync_fs        = ext4_sync_fs,
1109         .freeze_fs      = ext4_freeze,
1110         .unfreeze_fs    = ext4_unfreeze,
1111         .statfs         = ext4_statfs,
1112         .remount_fs     = ext4_remount,
1113         .show_options   = ext4_show_options,
1114 #ifdef CONFIG_QUOTA
1115         .quota_read     = ext4_quota_read,
1116         .quota_write    = ext4_quota_write,
1117         .get_dquots     = ext4_get_dquots,
1118 #endif
1119         .bdev_try_to_free_page = bdev_try_to_free_page,
1120 };
1121
1122 static const struct export_operations ext4_export_ops = {
1123         .fh_to_dentry = ext4_fh_to_dentry,
1124         .fh_to_parent = ext4_fh_to_parent,
1125         .get_parent = ext4_get_parent,
1126 };
1127
1128 enum {
1129         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1130         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1131         Opt_nouid32, Opt_debug, Opt_removed,
1132         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1133         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1134         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1135         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1136         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1137         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1138         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1139         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1140         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1141         Opt_usrquota, Opt_grpquota, Opt_i_version, Opt_dax,
1142         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1143         Opt_lazytime, Opt_nolazytime,
1144         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1145         Opt_inode_readahead_blks, Opt_journal_ioprio,
1146         Opt_dioread_nolock, Opt_dioread_lock,
1147         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1148         Opt_max_dir_size_kb, Opt_nojournal_checksum,
1149 };
1150
1151 static const match_table_t tokens = {
1152         {Opt_bsd_df, "bsddf"},
1153         {Opt_minix_df, "minixdf"},
1154         {Opt_grpid, "grpid"},
1155         {Opt_grpid, "bsdgroups"},
1156         {Opt_nogrpid, "nogrpid"},
1157         {Opt_nogrpid, "sysvgroups"},
1158         {Opt_resgid, "resgid=%u"},
1159         {Opt_resuid, "resuid=%u"},
1160         {Opt_sb, "sb=%u"},
1161         {Opt_err_cont, "errors=continue"},
1162         {Opt_err_panic, "errors=panic"},
1163         {Opt_err_ro, "errors=remount-ro"},
1164         {Opt_nouid32, "nouid32"},
1165         {Opt_debug, "debug"},
1166         {Opt_removed, "oldalloc"},
1167         {Opt_removed, "orlov"},
1168         {Opt_user_xattr, "user_xattr"},
1169         {Opt_nouser_xattr, "nouser_xattr"},
1170         {Opt_acl, "acl"},
1171         {Opt_noacl, "noacl"},
1172         {Opt_noload, "norecovery"},
1173         {Opt_noload, "noload"},
1174         {Opt_removed, "nobh"},
1175         {Opt_removed, "bh"},
1176         {Opt_commit, "commit=%u"},
1177         {Opt_min_batch_time, "min_batch_time=%u"},
1178         {Opt_max_batch_time, "max_batch_time=%u"},
1179         {Opt_journal_dev, "journal_dev=%u"},
1180         {Opt_journal_path, "journal_path=%s"},
1181         {Opt_journal_checksum, "journal_checksum"},
1182         {Opt_nojournal_checksum, "nojournal_checksum"},
1183         {Opt_journal_async_commit, "journal_async_commit"},
1184         {Opt_abort, "abort"},
1185         {Opt_data_journal, "data=journal"},
1186         {Opt_data_ordered, "data=ordered"},
1187         {Opt_data_writeback, "data=writeback"},
1188         {Opt_data_err_abort, "data_err=abort"},
1189         {Opt_data_err_ignore, "data_err=ignore"},
1190         {Opt_offusrjquota, "usrjquota="},
1191         {Opt_usrjquota, "usrjquota=%s"},
1192         {Opt_offgrpjquota, "grpjquota="},
1193         {Opt_grpjquota, "grpjquota=%s"},
1194         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1195         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1196         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1197         {Opt_grpquota, "grpquota"},
1198         {Opt_noquota, "noquota"},
1199         {Opt_quota, "quota"},
1200         {Opt_usrquota, "usrquota"},
1201         {Opt_barrier, "barrier=%u"},
1202         {Opt_barrier, "barrier"},
1203         {Opt_nobarrier, "nobarrier"},
1204         {Opt_i_version, "i_version"},
1205         {Opt_dax, "dax"},
1206         {Opt_stripe, "stripe=%u"},
1207         {Opt_delalloc, "delalloc"},
1208         {Opt_lazytime, "lazytime"},
1209         {Opt_nolazytime, "nolazytime"},
1210         {Opt_nodelalloc, "nodelalloc"},
1211         {Opt_removed, "mblk_io_submit"},
1212         {Opt_removed, "nomblk_io_submit"},
1213         {Opt_block_validity, "block_validity"},
1214         {Opt_noblock_validity, "noblock_validity"},
1215         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1216         {Opt_journal_ioprio, "journal_ioprio=%u"},
1217         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1218         {Opt_auto_da_alloc, "auto_da_alloc"},
1219         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1220         {Opt_dioread_nolock, "dioread_nolock"},
1221         {Opt_dioread_lock, "dioread_lock"},
1222         {Opt_discard, "discard"},
1223         {Opt_nodiscard, "nodiscard"},
1224         {Opt_init_itable, "init_itable=%u"},
1225         {Opt_init_itable, "init_itable"},
1226         {Opt_noinit_itable, "noinit_itable"},
1227         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1228         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1229         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1230         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1231         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1232         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1233         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1234         {Opt_err, NULL},
1235 };
1236
1237 static ext4_fsblk_t get_sb_block(void **data)
1238 {
1239         ext4_fsblk_t    sb_block;
1240         char            *options = (char *) *data;
1241
1242         if (!options || strncmp(options, "sb=", 3) != 0)
1243                 return 1;       /* Default location */
1244
1245         options += 3;
1246         /* TODO: use simple_strtoll with >32bit ext4 */
1247         sb_block = simple_strtoul(options, &options, 0);
1248         if (*options && *options != ',') {
1249                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1250                        (char *) *data);
1251                 return 1;
1252         }
1253         if (*options == ',')
1254                 options++;
1255         *data = (void *) options;
1256
1257         return sb_block;
1258 }
1259
1260 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1261 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1262         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1263
1264 #ifdef CONFIG_QUOTA
1265 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1266 {
1267         struct ext4_sb_info *sbi = EXT4_SB(sb);
1268         char *qname;
1269         int ret = -1;
1270
1271         if (sb_any_quota_loaded(sb) &&
1272                 !sbi->s_qf_names[qtype]) {
1273                 ext4_msg(sb, KERN_ERR,
1274                         "Cannot change journaled "
1275                         "quota options when quota turned on");
1276                 return -1;
1277         }
1278         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1279                 ext4_msg(sb, KERN_ERR, "Cannot set journaled quota options "
1280                          "when QUOTA feature is enabled");
1281                 return -1;
1282         }
1283         qname = match_strdup(args);
1284         if (!qname) {
1285                 ext4_msg(sb, KERN_ERR,
1286                         "Not enough memory for storing quotafile name");
1287                 return -1;
1288         }
1289         if (sbi->s_qf_names[qtype]) {
1290                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1291                         ret = 1;
1292                 else
1293                         ext4_msg(sb, KERN_ERR,
1294                                  "%s quota file already specified",
1295                                  QTYPE2NAME(qtype));
1296                 goto errout;
1297         }
1298         if (strchr(qname, '/')) {
1299                 ext4_msg(sb, KERN_ERR,
1300                         "quotafile must be on filesystem root");
1301                 goto errout;
1302         }
1303         sbi->s_qf_names[qtype] = qname;
1304         set_opt(sb, QUOTA);
1305         return 1;
1306 errout:
1307         kfree(qname);
1308         return ret;
1309 }
1310
1311 static int clear_qf_name(struct super_block *sb, int qtype)
1312 {
1313
1314         struct ext4_sb_info *sbi = EXT4_SB(sb);
1315
1316         if (sb_any_quota_loaded(sb) &&
1317                 sbi->s_qf_names[qtype]) {
1318                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1319                         " when quota turned on");
1320                 return -1;
1321         }
1322         kfree(sbi->s_qf_names[qtype]);
1323         sbi->s_qf_names[qtype] = NULL;
1324         return 1;
1325 }
1326 #endif
1327
1328 #define MOPT_SET        0x0001
1329 #define MOPT_CLEAR      0x0002
1330 #define MOPT_NOSUPPORT  0x0004
1331 #define MOPT_EXPLICIT   0x0008
1332 #define MOPT_CLEAR_ERR  0x0010
1333 #define MOPT_GTE0       0x0020
1334 #ifdef CONFIG_QUOTA
1335 #define MOPT_Q          0
1336 #define MOPT_QFMT       0x0040
1337 #else
1338 #define MOPT_Q          MOPT_NOSUPPORT
1339 #define MOPT_QFMT       MOPT_NOSUPPORT
1340 #endif
1341 #define MOPT_DATAJ      0x0080
1342 #define MOPT_NO_EXT2    0x0100
1343 #define MOPT_NO_EXT3    0x0200
1344 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1345 #define MOPT_STRING     0x0400
1346
1347 static const struct mount_opts {
1348         int     token;
1349         int     mount_opt;
1350         int     flags;
1351 } ext4_mount_opts[] = {
1352         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1353         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1354         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1355         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1356         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1357         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1358         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1359          MOPT_EXT4_ONLY | MOPT_SET},
1360         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1361          MOPT_EXT4_ONLY | MOPT_CLEAR},
1362         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1363         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1364         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1365          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1366         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1367          MOPT_EXT4_ONLY | MOPT_CLEAR},
1368         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1369          MOPT_EXT4_ONLY | MOPT_CLEAR},
1370         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1371          MOPT_EXT4_ONLY | MOPT_SET},
1372         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1373                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1374          MOPT_EXT4_ONLY | MOPT_SET},
1375         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1376         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1377         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1378         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1379         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1380          MOPT_NO_EXT2 | MOPT_SET},
1381         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1382          MOPT_NO_EXT2 | MOPT_CLEAR},
1383         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1384         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1385         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1386         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1387         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1388         {Opt_commit, 0, MOPT_GTE0},
1389         {Opt_max_batch_time, 0, MOPT_GTE0},
1390         {Opt_min_batch_time, 0, MOPT_GTE0},
1391         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1392         {Opt_init_itable, 0, MOPT_GTE0},
1393         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1394         {Opt_stripe, 0, MOPT_GTE0},
1395         {Opt_resuid, 0, MOPT_GTE0},
1396         {Opt_resgid, 0, MOPT_GTE0},
1397         {Opt_journal_dev, 0, MOPT_GTE0},
1398         {Opt_journal_path, 0, MOPT_STRING},
1399         {Opt_journal_ioprio, 0, MOPT_GTE0},
1400         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1401         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1402         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1403          MOPT_NO_EXT2 | MOPT_DATAJ},
1404         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1405         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1406 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1407         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1408         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1409 #else
1410         {Opt_acl, 0, MOPT_NOSUPPORT},
1411         {Opt_noacl, 0, MOPT_NOSUPPORT},
1412 #endif
1413         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1414         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1415         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1416         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1417                                                         MOPT_SET | MOPT_Q},
1418         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1419                                                         MOPT_SET | MOPT_Q},
1420         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1421                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1422         {Opt_usrjquota, 0, MOPT_Q},
1423         {Opt_grpjquota, 0, MOPT_Q},
1424         {Opt_offusrjquota, 0, MOPT_Q},
1425         {Opt_offgrpjquota, 0, MOPT_Q},
1426         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1427         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1428         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1429         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1430         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1431         {Opt_err, 0, 0}
1432 };
1433
1434 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1435                             substring_t *args, unsigned long *journal_devnum,
1436                             unsigned int *journal_ioprio, int is_remount)
1437 {
1438         struct ext4_sb_info *sbi = EXT4_SB(sb);
1439         const struct mount_opts *m;
1440         kuid_t uid;
1441         kgid_t gid;
1442         int arg = 0;
1443
1444 #ifdef CONFIG_QUOTA
1445         if (token == Opt_usrjquota)
1446                 return set_qf_name(sb, USRQUOTA, &args[0]);
1447         else if (token == Opt_grpjquota)
1448                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1449         else if (token == Opt_offusrjquota)
1450                 return clear_qf_name(sb, USRQUOTA);
1451         else if (token == Opt_offgrpjquota)
1452                 return clear_qf_name(sb, GRPQUOTA);
1453 #endif
1454         switch (token) {
1455         case Opt_noacl:
1456         case Opt_nouser_xattr:
1457                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1458                 break;
1459         case Opt_sb:
1460                 return 1;       /* handled by get_sb_block() */
1461         case Opt_removed:
1462                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1463                 return 1;
1464         case Opt_abort:
1465                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1466                 return 1;
1467         case Opt_i_version:
1468                 sb->s_flags |= MS_I_VERSION;
1469                 return 1;
1470         case Opt_lazytime:
1471                 sb->s_flags |= MS_LAZYTIME;
1472                 return 1;
1473         case Opt_nolazytime:
1474                 sb->s_flags &= ~MS_LAZYTIME;
1475                 return 1;
1476         }
1477
1478         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1479                 if (token == m->token)
1480                         break;
1481
1482         if (m->token == Opt_err) {
1483                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1484                          "or missing value", opt);
1485                 return -1;
1486         }
1487
1488         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1489                 ext4_msg(sb, KERN_ERR,
1490                          "Mount option \"%s\" incompatible with ext2", opt);
1491                 return -1;
1492         }
1493         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1494                 ext4_msg(sb, KERN_ERR,
1495                          "Mount option \"%s\" incompatible with ext3", opt);
1496                 return -1;
1497         }
1498
1499         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1500                 return -1;
1501         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1502                 return -1;
1503         if (m->flags & MOPT_EXPLICIT)
1504                 set_opt2(sb, EXPLICIT_DELALLOC);
1505         if (m->flags & MOPT_CLEAR_ERR)
1506                 clear_opt(sb, ERRORS_MASK);
1507         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1508                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1509                          "options when quota turned on");
1510                 return -1;
1511         }
1512
1513         if (m->flags & MOPT_NOSUPPORT) {
1514                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1515         } else if (token == Opt_commit) {
1516                 if (arg == 0)
1517                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1518                 sbi->s_commit_interval = HZ * arg;
1519         } else if (token == Opt_max_batch_time) {
1520                 sbi->s_max_batch_time = arg;
1521         } else if (token == Opt_min_batch_time) {
1522                 sbi->s_min_batch_time = arg;
1523         } else if (token == Opt_inode_readahead_blks) {
1524                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1525                         ext4_msg(sb, KERN_ERR,
1526                                  "EXT4-fs: inode_readahead_blks must be "
1527                                  "0 or a power of 2 smaller than 2^31");
1528                         return -1;
1529                 }
1530                 sbi->s_inode_readahead_blks = arg;
1531         } else if (token == Opt_init_itable) {
1532                 set_opt(sb, INIT_INODE_TABLE);
1533                 if (!args->from)
1534                         arg = EXT4_DEF_LI_WAIT_MULT;
1535                 sbi->s_li_wait_mult = arg;
1536         } else if (token == Opt_max_dir_size_kb) {
1537                 sbi->s_max_dir_size_kb = arg;
1538         } else if (token == Opt_stripe) {
1539                 sbi->s_stripe = arg;
1540         } else if (token == Opt_resuid) {
1541                 uid = make_kuid(current_user_ns(), arg);
1542                 if (!uid_valid(uid)) {
1543                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1544                         return -1;
1545                 }
1546                 sbi->s_resuid = uid;
1547         } else if (token == Opt_resgid) {
1548                 gid = make_kgid(current_user_ns(), arg);
1549                 if (!gid_valid(gid)) {
1550                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1551                         return -1;
1552                 }
1553                 sbi->s_resgid = gid;
1554         } else if (token == Opt_journal_dev) {
1555                 if (is_remount) {
1556                         ext4_msg(sb, KERN_ERR,
1557                                  "Cannot specify journal on remount");
1558                         return -1;
1559                 }
1560                 *journal_devnum = arg;
1561         } else if (token == Opt_journal_path) {
1562                 char *journal_path;
1563                 struct inode *journal_inode;
1564                 struct path path;
1565                 int error;
1566
1567                 if (is_remount) {
1568                         ext4_msg(sb, KERN_ERR,
1569                                  "Cannot specify journal on remount");
1570                         return -1;
1571                 }
1572                 journal_path = match_strdup(&args[0]);
1573                 if (!journal_path) {
1574                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1575                                 "journal device string");
1576                         return -1;
1577                 }
1578
1579                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1580                 if (error) {
1581                         ext4_msg(sb, KERN_ERR, "error: could not find "
1582                                 "journal device path: error %d", error);
1583                         kfree(journal_path);
1584                         return -1;
1585                 }
1586
1587                 journal_inode = d_inode(path.dentry);
1588                 if (!S_ISBLK(journal_inode->i_mode)) {
1589                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1590                                 "is not a block device", journal_path);
1591                         path_put(&path);
1592                         kfree(journal_path);
1593                         return -1;
1594                 }
1595
1596                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1597                 path_put(&path);
1598                 kfree(journal_path);
1599         } else if (token == Opt_journal_ioprio) {
1600                 if (arg > 7) {
1601                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1602                                  " (must be 0-7)");
1603                         return -1;
1604                 }
1605                 *journal_ioprio =
1606                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1607         } else if (token == Opt_test_dummy_encryption) {
1608 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1609                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1610                 ext4_msg(sb, KERN_WARNING,
1611                          "Test dummy encryption mode enabled");
1612 #else
1613                 ext4_msg(sb, KERN_WARNING,
1614                          "Test dummy encryption mount option ignored");
1615 #endif
1616         } else if (m->flags & MOPT_DATAJ) {
1617                 if (is_remount) {
1618                         if (!sbi->s_journal)
1619                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1620                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1621                                 ext4_msg(sb, KERN_ERR,
1622                                          "Cannot change data mode on remount");
1623                                 return -1;
1624                         }
1625                 } else {
1626                         clear_opt(sb, DATA_FLAGS);
1627                         sbi->s_mount_opt |= m->mount_opt;
1628                 }
1629 #ifdef CONFIG_QUOTA
1630         } else if (m->flags & MOPT_QFMT) {
1631                 if (sb_any_quota_loaded(sb) &&
1632                     sbi->s_jquota_fmt != m->mount_opt) {
1633                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1634                                  "quota options when quota turned on");
1635                         return -1;
1636                 }
1637                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
1638                                                EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1639                         ext4_msg(sb, KERN_ERR,
1640                                  "Cannot set journaled quota options "
1641                                  "when QUOTA feature is enabled");
1642                         return -1;
1643                 }
1644                 sbi->s_jquota_fmt = m->mount_opt;
1645 #endif
1646 #ifndef CONFIG_FS_DAX
1647         } else if (token == Opt_dax) {
1648                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1649                 return -1;
1650 #endif
1651         } else {
1652                 if (!args->from)
1653                         arg = 1;
1654                 if (m->flags & MOPT_CLEAR)
1655                         arg = !arg;
1656                 else if (unlikely(!(m->flags & MOPT_SET))) {
1657                         ext4_msg(sb, KERN_WARNING,
1658                                  "buggy handling of option %s", opt);
1659                         WARN_ON(1);
1660                         return -1;
1661                 }
1662                 if (arg != 0)
1663                         sbi->s_mount_opt |= m->mount_opt;
1664                 else
1665                         sbi->s_mount_opt &= ~m->mount_opt;
1666         }
1667         return 1;
1668 }
1669
1670 static int parse_options(char *options, struct super_block *sb,
1671                          unsigned long *journal_devnum,
1672                          unsigned int *journal_ioprio,
1673                          int is_remount)
1674 {
1675         struct ext4_sb_info *sbi = EXT4_SB(sb);
1676         char *p;
1677         substring_t args[MAX_OPT_ARGS];
1678         int token;
1679
1680         if (!options)
1681                 return 1;
1682
1683         while ((p = strsep(&options, ",")) != NULL) {
1684                 if (!*p)
1685                         continue;
1686                 /*
1687                  * Initialize args struct so we know whether arg was
1688                  * found; some options take optional arguments.
1689                  */
1690                 args[0].to = args[0].from = NULL;
1691                 token = match_token(p, tokens, args);
1692                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1693                                      journal_ioprio, is_remount) < 0)
1694                         return 0;
1695         }
1696 #ifdef CONFIG_QUOTA
1697         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
1698             (test_opt(sb, USRQUOTA) || test_opt(sb, GRPQUOTA))) {
1699                 ext4_msg(sb, KERN_ERR, "Cannot set quota options when QUOTA "
1700                          "feature is enabled");
1701                 return 0;
1702         }
1703         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1704                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1705                         clear_opt(sb, USRQUOTA);
1706
1707                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1708                         clear_opt(sb, GRPQUOTA);
1709
1710                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1711                         ext4_msg(sb, KERN_ERR, "old and new quota "
1712                                         "format mixing");
1713                         return 0;
1714                 }
1715
1716                 if (!sbi->s_jquota_fmt) {
1717                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1718                                         "not specified");
1719                         return 0;
1720                 }
1721         }
1722 #endif
1723         if (test_opt(sb, DIOREAD_NOLOCK)) {
1724                 int blocksize =
1725                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1726
1727                 if (blocksize < PAGE_CACHE_SIZE) {
1728                         ext4_msg(sb, KERN_ERR, "can't mount with "
1729                                  "dioread_nolock if block size != PAGE_SIZE");
1730                         return 0;
1731                 }
1732         }
1733         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
1734             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
1735                 ext4_msg(sb, KERN_ERR, "can't mount with journal_async_commit "
1736                          "in data=ordered mode");
1737                 return 0;
1738         }
1739         return 1;
1740 }
1741
1742 static inline void ext4_show_quota_options(struct seq_file *seq,
1743                                            struct super_block *sb)
1744 {
1745 #if defined(CONFIG_QUOTA)
1746         struct ext4_sb_info *sbi = EXT4_SB(sb);
1747
1748         if (sbi->s_jquota_fmt) {
1749                 char *fmtname = "";
1750
1751                 switch (sbi->s_jquota_fmt) {
1752                 case QFMT_VFS_OLD:
1753                         fmtname = "vfsold";
1754                         break;
1755                 case QFMT_VFS_V0:
1756                         fmtname = "vfsv0";
1757                         break;
1758                 case QFMT_VFS_V1:
1759                         fmtname = "vfsv1";
1760                         break;
1761                 }
1762                 seq_printf(seq, ",jqfmt=%s", fmtname);
1763         }
1764
1765         if (sbi->s_qf_names[USRQUOTA])
1766                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1767
1768         if (sbi->s_qf_names[GRPQUOTA])
1769                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1770 #endif
1771 }
1772
1773 static const char *token2str(int token)
1774 {
1775         const struct match_token *t;
1776
1777         for (t = tokens; t->token != Opt_err; t++)
1778                 if (t->token == token && !strchr(t->pattern, '='))
1779                         break;
1780         return t->pattern;
1781 }
1782
1783 /*
1784  * Show an option if
1785  *  - it's set to a non-default value OR
1786  *  - if the per-sb default is different from the global default
1787  */
1788 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1789                               int nodefs)
1790 {
1791         struct ext4_sb_info *sbi = EXT4_SB(sb);
1792         struct ext4_super_block *es = sbi->s_es;
1793         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1794         const struct mount_opts *m;
1795         char sep = nodefs ? '\n' : ',';
1796
1797 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1798 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1799
1800         if (sbi->s_sb_block != 1)
1801                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1802
1803         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1804                 int want_set = m->flags & MOPT_SET;
1805                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1806                     (m->flags & MOPT_CLEAR_ERR))
1807                         continue;
1808                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1809                         continue; /* skip if same as the default */
1810                 if ((want_set &&
1811                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1812                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1813                         continue; /* select Opt_noFoo vs Opt_Foo */
1814                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1815         }
1816
1817         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1818             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1819                 SEQ_OPTS_PRINT("resuid=%u",
1820                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1821         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1822             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1823                 SEQ_OPTS_PRINT("resgid=%u",
1824                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1825         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1826         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1827                 SEQ_OPTS_PUTS("errors=remount-ro");
1828         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1829                 SEQ_OPTS_PUTS("errors=continue");
1830         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1831                 SEQ_OPTS_PUTS("errors=panic");
1832         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1833                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1834         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1835                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1836         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1837                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1838         if (sb->s_flags & MS_I_VERSION)
1839                 SEQ_OPTS_PUTS("i_version");
1840         if (nodefs || sbi->s_stripe)
1841                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1842         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1843                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1844                         SEQ_OPTS_PUTS("data=journal");
1845                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1846                         SEQ_OPTS_PUTS("data=ordered");
1847                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1848                         SEQ_OPTS_PUTS("data=writeback");
1849         }
1850         if (nodefs ||
1851             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1852                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1853                                sbi->s_inode_readahead_blks);
1854
1855         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1856                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1857                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1858         if (nodefs || sbi->s_max_dir_size_kb)
1859                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1860
1861         ext4_show_quota_options(seq, sb);
1862         return 0;
1863 }
1864
1865 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1866 {
1867         return _ext4_show_options(seq, root->d_sb, 0);
1868 }
1869
1870 static int options_seq_show(struct seq_file *seq, void *offset)
1871 {
1872         struct super_block *sb = seq->private;
1873         int rc;
1874
1875         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1876         rc = _ext4_show_options(seq, sb, 1);
1877         seq_puts(seq, "\n");
1878         return rc;
1879 }
1880
1881 static int options_open_fs(struct inode *inode, struct file *file)
1882 {
1883         return single_open(file, options_seq_show, PDE_DATA(inode));
1884 }
1885
1886 static const struct file_operations ext4_seq_options_fops = {
1887         .owner = THIS_MODULE,
1888         .open = options_open_fs,
1889         .read = seq_read,
1890         .llseek = seq_lseek,
1891         .release = single_release,
1892 };
1893
1894 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1895                             int read_only)
1896 {
1897         struct ext4_sb_info *sbi = EXT4_SB(sb);
1898         int res = 0;
1899
1900         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1901                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1902                          "forcing read-only mode");
1903                 res = MS_RDONLY;
1904         }
1905         if (read_only)
1906                 goto done;
1907         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1908                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1909                          "running e2fsck is recommended");
1910         else if (sbi->s_mount_state & EXT4_ERROR_FS)
1911                 ext4_msg(sb, KERN_WARNING,
1912                          "warning: mounting fs with errors, "
1913                          "running e2fsck is recommended");
1914         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1915                  le16_to_cpu(es->s_mnt_count) >=
1916                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1917                 ext4_msg(sb, KERN_WARNING,
1918                          "warning: maximal mount count reached, "
1919                          "running e2fsck is recommended");
1920         else if (le32_to_cpu(es->s_checkinterval) &&
1921                 (le32_to_cpu(es->s_lastcheck) +
1922                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1923                 ext4_msg(sb, KERN_WARNING,
1924                          "warning: checktime reached, "
1925                          "running e2fsck is recommended");
1926         if (!sbi->s_journal)
1927                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1928         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1929                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1930         le16_add_cpu(&es->s_mnt_count, 1);
1931         es->s_mtime = cpu_to_le32(get_seconds());
1932         ext4_update_dynamic_rev(sb);
1933         if (sbi->s_journal)
1934                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1935
1936         ext4_commit_super(sb, 1);
1937 done:
1938         if (test_opt(sb, DEBUG))
1939                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1940                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1941                         sb->s_blocksize,
1942                         sbi->s_groups_count,
1943                         EXT4_BLOCKS_PER_GROUP(sb),
1944                         EXT4_INODES_PER_GROUP(sb),
1945                         sbi->s_mount_opt, sbi->s_mount_opt2);
1946
1947         cleancache_init_fs(sb);
1948         return res;
1949 }
1950
1951 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1952 {
1953         struct ext4_sb_info *sbi = EXT4_SB(sb);
1954         struct flex_groups *new_groups;
1955         int size;
1956
1957         if (!sbi->s_log_groups_per_flex)
1958                 return 0;
1959
1960         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1961         if (size <= sbi->s_flex_groups_allocated)
1962                 return 0;
1963
1964         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1965         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1966         if (!new_groups) {
1967                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1968                          size / (int) sizeof(struct flex_groups));
1969                 return -ENOMEM;
1970         }
1971
1972         if (sbi->s_flex_groups) {
1973                 memcpy(new_groups, sbi->s_flex_groups,
1974                        (sbi->s_flex_groups_allocated *
1975                         sizeof(struct flex_groups)));
1976                 kvfree(sbi->s_flex_groups);
1977         }
1978         sbi->s_flex_groups = new_groups;
1979         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1980         return 0;
1981 }
1982
1983 static int ext4_fill_flex_info(struct super_block *sb)
1984 {
1985         struct ext4_sb_info *sbi = EXT4_SB(sb);
1986         struct ext4_group_desc *gdp = NULL;
1987         ext4_group_t flex_group;
1988         int i, err;
1989
1990         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1991         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1992                 sbi->s_log_groups_per_flex = 0;
1993                 return 1;
1994         }
1995
1996         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
1997         if (err)
1998                 goto failed;
1999
2000         for (i = 0; i < sbi->s_groups_count; i++) {
2001                 gdp = ext4_get_group_desc(sb, i, NULL);
2002
2003                 flex_group = ext4_flex_group(sbi, i);
2004                 atomic_add(ext4_free_inodes_count(sb, gdp),
2005                            &sbi->s_flex_groups[flex_group].free_inodes);
2006                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2007                              &sbi->s_flex_groups[flex_group].free_clusters);
2008                 atomic_add(ext4_used_dirs_count(sb, gdp),
2009                            &sbi->s_flex_groups[flex_group].used_dirs);
2010         }
2011
2012         return 1;
2013 failed:
2014         return 0;
2015 }
2016
2017 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
2018                                    struct ext4_group_desc *gdp)
2019 {
2020         int offset;
2021         __u16 crc = 0;
2022         __le32 le_group = cpu_to_le32(block_group);
2023
2024         if (ext4_has_metadata_csum(sbi->s_sb)) {
2025                 /* Use new metadata_csum algorithm */
2026                 __le16 save_csum;
2027                 __u32 csum32;
2028
2029                 save_csum = gdp->bg_checksum;
2030                 gdp->bg_checksum = 0;
2031                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2032                                      sizeof(le_group));
2033                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
2034                                      sbi->s_desc_size);
2035                 gdp->bg_checksum = save_csum;
2036
2037                 crc = csum32 & 0xFFFF;
2038                 goto out;
2039         }
2040
2041         /* old crc16 code */
2042         if (!(sbi->s_es->s_feature_ro_compat &
2043               cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)))
2044                 return 0;
2045
2046         offset = offsetof(struct ext4_group_desc, bg_checksum);
2047
2048         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2049         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2050         crc = crc16(crc, (__u8 *)gdp, offset);
2051         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2052         /* for checksum of struct ext4_group_desc do the rest...*/
2053         if ((sbi->s_es->s_feature_incompat &
2054              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2055             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2056                 crc = crc16(crc, (__u8 *)gdp + offset,
2057                             le16_to_cpu(sbi->s_es->s_desc_size) -
2058                                 offset);
2059
2060 out:
2061         return cpu_to_le16(crc);
2062 }
2063
2064 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2065                                 struct ext4_group_desc *gdp)
2066 {
2067         if (ext4_has_group_desc_csum(sb) &&
2068             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
2069                                                       block_group, gdp)))
2070                 return 0;
2071
2072         return 1;
2073 }
2074
2075 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2076                               struct ext4_group_desc *gdp)
2077 {
2078         if (!ext4_has_group_desc_csum(sb))
2079                 return;
2080         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2081 }
2082
2083 /* Called at mount-time, super-block is locked */
2084 static int ext4_check_descriptors(struct super_block *sb,
2085                                   ext4_group_t *first_not_zeroed)
2086 {
2087         struct ext4_sb_info *sbi = EXT4_SB(sb);
2088         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2089         ext4_fsblk_t last_block;
2090         ext4_fsblk_t block_bitmap;
2091         ext4_fsblk_t inode_bitmap;
2092         ext4_fsblk_t inode_table;
2093         int flexbg_flag = 0;
2094         ext4_group_t i, grp = sbi->s_groups_count;
2095
2096         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2097                 flexbg_flag = 1;
2098
2099         ext4_debug("Checking group descriptors");
2100
2101         for (i = 0; i < sbi->s_groups_count; i++) {
2102                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2103
2104                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2105                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2106                 else
2107                         last_block = first_block +
2108                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2109
2110                 if ((grp == sbi->s_groups_count) &&
2111                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2112                         grp = i;
2113
2114                 block_bitmap = ext4_block_bitmap(sb, gdp);
2115                 if (block_bitmap < first_block || block_bitmap > last_block) {
2116                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2117                                "Block bitmap for group %u not in group "
2118                                "(block %llu)!", i, block_bitmap);
2119                         return 0;
2120                 }
2121                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2122                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2123                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2124                                "Inode bitmap for group %u not in group "
2125                                "(block %llu)!", i, inode_bitmap);
2126                         return 0;
2127                 }
2128                 inode_table = ext4_inode_table(sb, gdp);
2129                 if (inode_table < first_block ||
2130                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2131                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2132                                "Inode table for group %u not in group "
2133                                "(block %llu)!", i, inode_table);
2134                         return 0;
2135                 }
2136                 ext4_lock_group(sb, i);
2137                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2138                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2139                                  "Checksum for group %u failed (%u!=%u)",
2140                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2141                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2142                         if (!(sb->s_flags & MS_RDONLY)) {
2143                                 ext4_unlock_group(sb, i);
2144                                 return 0;
2145                         }
2146                 }
2147                 ext4_unlock_group(sb, i);
2148                 if (!flexbg_flag)
2149                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2150         }
2151         if (NULL != first_not_zeroed)
2152                 *first_not_zeroed = grp;
2153         return 1;
2154 }
2155
2156 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2157  * the superblock) which were deleted from all directories, but held open by
2158  * a process at the time of a crash.  We walk the list and try to delete these
2159  * inodes at recovery time (only with a read-write filesystem).
2160  *
2161  * In order to keep the orphan inode chain consistent during traversal (in
2162  * case of crash during recovery), we link each inode into the superblock
2163  * orphan list_head and handle it the same way as an inode deletion during
2164  * normal operation (which journals the operations for us).
2165  *
2166  * We only do an iget() and an iput() on each inode, which is very safe if we
2167  * accidentally point at an in-use or already deleted inode.  The worst that
2168  * can happen in this case is that we get a "bit already cleared" message from
2169  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2170  * e2fsck was run on this filesystem, and it must have already done the orphan
2171  * inode cleanup for us, so we can safely abort without any further action.
2172  */
2173 static void ext4_orphan_cleanup(struct super_block *sb,
2174                                 struct ext4_super_block *es)
2175 {
2176         unsigned int s_flags = sb->s_flags;
2177         int nr_orphans = 0, nr_truncates = 0;
2178 #ifdef CONFIG_QUOTA
2179         int i;
2180 #endif
2181         if (!es->s_last_orphan) {
2182                 jbd_debug(4, "no orphan inodes to clean up\n");
2183                 return;
2184         }
2185
2186         if (bdev_read_only(sb->s_bdev)) {
2187                 ext4_msg(sb, KERN_ERR, "write access "
2188                         "unavailable, skipping orphan cleanup");
2189                 return;
2190         }
2191
2192         /* Check if feature set would not allow a r/w mount */
2193         if (!ext4_feature_set_ok(sb, 0)) {
2194                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2195                          "unknown ROCOMPAT features");
2196                 return;
2197         }
2198
2199         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2200                 /* don't clear list on RO mount w/ errors */
2201                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2202                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2203                                   "clearing orphan list.\n");
2204                         es->s_last_orphan = 0;
2205                 }
2206                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2207                 return;
2208         }
2209
2210         if (s_flags & MS_RDONLY) {
2211                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2212                 sb->s_flags &= ~MS_RDONLY;
2213         }
2214 #ifdef CONFIG_QUOTA
2215         /* Needed for iput() to work correctly and not trash data */
2216         sb->s_flags |= MS_ACTIVE;
2217         /* Turn on quotas so that they are updated correctly */
2218         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2219                 if (EXT4_SB(sb)->s_qf_names[i]) {
2220                         int ret = ext4_quota_on_mount(sb, i);
2221                         if (ret < 0)
2222                                 ext4_msg(sb, KERN_ERR,
2223                                         "Cannot turn on journaled "
2224                                         "quota: error %d", ret);
2225                 }
2226         }
2227 #endif
2228
2229         while (es->s_last_orphan) {
2230                 struct inode *inode;
2231
2232                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2233                 if (IS_ERR(inode)) {
2234                         es->s_last_orphan = 0;
2235                         break;
2236                 }
2237
2238                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2239                 dquot_initialize(inode);
2240                 if (inode->i_nlink) {
2241                         if (test_opt(sb, DEBUG))
2242                                 ext4_msg(sb, KERN_DEBUG,
2243                                         "%s: truncating inode %lu to %lld bytes",
2244                                         __func__, inode->i_ino, inode->i_size);
2245                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2246                                   inode->i_ino, inode->i_size);
2247                         mutex_lock(&inode->i_mutex);
2248                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2249                         ext4_truncate(inode);
2250                         mutex_unlock(&inode->i_mutex);
2251                         nr_truncates++;
2252                 } else {
2253                         if (test_opt(sb, DEBUG))
2254                                 ext4_msg(sb, KERN_DEBUG,
2255                                         "%s: deleting unreferenced inode %lu",
2256                                         __func__, inode->i_ino);
2257                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2258                                   inode->i_ino);
2259                         nr_orphans++;
2260                 }
2261                 iput(inode);  /* The delete magic happens here! */
2262         }
2263
2264 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2265
2266         if (nr_orphans)
2267                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2268                        PLURAL(nr_orphans));
2269         if (nr_truncates)
2270                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2271                        PLURAL(nr_truncates));
2272 #ifdef CONFIG_QUOTA
2273         /* Turn quotas off */
2274         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2275                 if (sb_dqopt(sb)->files[i])
2276                         dquot_quota_off(sb, i);
2277         }
2278 #endif
2279         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2280 }
2281
2282 /*
2283  * Maximal extent format file size.
2284  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2285  * extent format containers, within a sector_t, and within i_blocks
2286  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2287  * so that won't be a limiting factor.
2288  *
2289  * However there is other limiting factor. We do store extents in the form
2290  * of starting block and length, hence the resulting length of the extent
2291  * covering maximum file size must fit into on-disk format containers as
2292  * well. Given that length is always by 1 unit bigger than max unit (because
2293  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2294  *
2295  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2296  */
2297 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2298 {
2299         loff_t res;
2300         loff_t upper_limit = MAX_LFS_FILESIZE;
2301
2302         /* small i_blocks in vfs inode? */
2303         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2304                 /*
2305                  * CONFIG_LBDAF is not enabled implies the inode
2306                  * i_block represent total blocks in 512 bytes
2307                  * 32 == size of vfs inode i_blocks * 8
2308                  */
2309                 upper_limit = (1LL << 32) - 1;
2310
2311                 /* total blocks in file system block size */
2312                 upper_limit >>= (blkbits - 9);
2313                 upper_limit <<= blkbits;
2314         }
2315
2316         /*
2317          * 32-bit extent-start container, ee_block. We lower the maxbytes
2318          * by one fs block, so ee_len can cover the extent of maximum file
2319          * size
2320          */
2321         res = (1LL << 32) - 1;
2322         res <<= blkbits;
2323
2324         /* Sanity check against vm- & vfs- imposed limits */
2325         if (res > upper_limit)
2326                 res = upper_limit;
2327
2328         return res;
2329 }
2330
2331 /*
2332  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2333  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2334  * We need to be 1 filesystem block less than the 2^48 sector limit.
2335  */
2336 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2337 {
2338         loff_t res = EXT4_NDIR_BLOCKS;
2339         int meta_blocks;
2340         loff_t upper_limit;
2341         /* This is calculated to be the largest file size for a dense, block
2342          * mapped file such that the file's total number of 512-byte sectors,
2343          * including data and all indirect blocks, does not exceed (2^48 - 1).
2344          *
2345          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2346          * number of 512-byte sectors of the file.
2347          */
2348
2349         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2350                 /*
2351                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2352                  * the inode i_block field represents total file blocks in
2353                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2354                  */
2355                 upper_limit = (1LL << 32) - 1;
2356
2357                 /* total blocks in file system block size */
2358                 upper_limit >>= (bits - 9);
2359
2360         } else {
2361                 /*
2362                  * We use 48 bit ext4_inode i_blocks
2363                  * With EXT4_HUGE_FILE_FL set the i_blocks
2364                  * represent total number of blocks in
2365                  * file system block size
2366                  */
2367                 upper_limit = (1LL << 48) - 1;
2368
2369         }
2370
2371         /* indirect blocks */
2372         meta_blocks = 1;
2373         /* double indirect blocks */
2374         meta_blocks += 1 + (1LL << (bits-2));
2375         /* tripple indirect blocks */
2376         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2377
2378         upper_limit -= meta_blocks;
2379         upper_limit <<= bits;
2380
2381         res += 1LL << (bits-2);
2382         res += 1LL << (2*(bits-2));
2383         res += 1LL << (3*(bits-2));
2384         res <<= bits;
2385         if (res > upper_limit)
2386                 res = upper_limit;
2387
2388         if (res > MAX_LFS_FILESIZE)
2389                 res = MAX_LFS_FILESIZE;
2390
2391         return res;
2392 }
2393
2394 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2395                                    ext4_fsblk_t logical_sb_block, int nr)
2396 {
2397         struct ext4_sb_info *sbi = EXT4_SB(sb);
2398         ext4_group_t bg, first_meta_bg;
2399         int has_super = 0;
2400
2401         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2402
2403         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2404             nr < first_meta_bg)
2405                 return logical_sb_block + nr + 1;
2406         bg = sbi->s_desc_per_block * nr;
2407         if (ext4_bg_has_super(sb, bg))
2408                 has_super = 1;
2409
2410         /*
2411          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2412          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2413          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2414          * compensate.
2415          */
2416         if (sb->s_blocksize == 1024 && nr == 0 &&
2417             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2418                 has_super++;
2419
2420         return (has_super + ext4_group_first_block_no(sb, bg));
2421 }
2422
2423 /**
2424  * ext4_get_stripe_size: Get the stripe size.
2425  * @sbi: In memory super block info
2426  *
2427  * If we have specified it via mount option, then
2428  * use the mount option value. If the value specified at mount time is
2429  * greater than the blocks per group use the super block value.
2430  * If the super block value is greater than blocks per group return 0.
2431  * Allocator needs it be less than blocks per group.
2432  *
2433  */
2434 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2435 {
2436         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2437         unsigned long stripe_width =
2438                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2439         int ret;
2440
2441         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2442                 ret = sbi->s_stripe;
2443         else if (stripe_width <= sbi->s_blocks_per_group)
2444                 ret = stripe_width;
2445         else if (stride <= sbi->s_blocks_per_group)
2446                 ret = stride;
2447         else
2448                 ret = 0;
2449
2450         /*
2451          * If the stripe width is 1, this makes no sense and
2452          * we set it to 0 to turn off stripe handling code.
2453          */
2454         if (ret <= 1)
2455                 ret = 0;
2456
2457         return ret;
2458 }
2459
2460 /* sysfs supprt */
2461
2462 struct ext4_attr {
2463         struct attribute attr;
2464         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2465         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2466                          const char *, size_t);
2467         union {
2468                 int offset;
2469                 int deprecated_val;
2470         } u;
2471 };
2472
2473 static int parse_strtoull(const char *buf,
2474                 unsigned long long max, unsigned long long *value)
2475 {
2476         int ret;
2477
2478         ret = kstrtoull(skip_spaces(buf), 0, value);
2479         if (!ret && *value > max)
2480                 ret = -EINVAL;
2481         return ret;
2482 }
2483
2484 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2485                                               struct ext4_sb_info *sbi,
2486                                               char *buf)
2487 {
2488         return snprintf(buf, PAGE_SIZE, "%llu\n",
2489                 (s64) EXT4_C2B(sbi,
2490                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2491 }
2492
2493 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2494                                          struct ext4_sb_info *sbi, char *buf)
2495 {
2496         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2497
2498         if (!sb->s_bdev->bd_part)
2499                 return snprintf(buf, PAGE_SIZE, "0\n");
2500         return snprintf(buf, PAGE_SIZE, "%lu\n",
2501                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2502                          sbi->s_sectors_written_start) >> 1);
2503 }
2504
2505 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2506                                           struct ext4_sb_info *sbi, char *buf)
2507 {
2508         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2509
2510         if (!sb->s_bdev->bd_part)
2511                 return snprintf(buf, PAGE_SIZE, "0\n");
2512         return snprintf(buf, PAGE_SIZE, "%llu\n",
2513                         (unsigned long long)(sbi->s_kbytes_written +
2514                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2515                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2516 }
2517
2518 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2519                                           struct ext4_sb_info *sbi,
2520                                           const char *buf, size_t count)
2521 {
2522         unsigned long t;
2523         int ret;
2524
2525         ret = kstrtoul(skip_spaces(buf), 0, &t);
2526         if (ret)
2527                 return ret;
2528
2529         if (t && (!is_power_of_2(t) || t > 0x40000000))
2530                 return -EINVAL;
2531
2532         sbi->s_inode_readahead_blks = t;
2533         return count;
2534 }
2535
2536 static ssize_t sbi_ui_show(struct ext4_attr *a,
2537                            struct ext4_sb_info *sbi, char *buf)
2538 {
2539         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->u.offset);
2540
2541         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2542 }
2543
2544 static ssize_t sbi_ui_store(struct ext4_attr *a,
2545                             struct ext4_sb_info *sbi,
2546                             const char *buf, size_t count)
2547 {
2548         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->u.offset);
2549         unsigned long t;
2550         int ret;
2551
2552         ret = kstrtoul(skip_spaces(buf), 0, &t);
2553         if (ret)
2554                 return ret;
2555         *ui = t;
2556         return count;
2557 }
2558
2559 static ssize_t es_ui_show(struct ext4_attr *a,
2560                            struct ext4_sb_info *sbi, char *buf)
2561 {
2562
2563         unsigned int *ui = (unsigned int *) (((char *) sbi->s_es) +
2564                            a->u.offset);
2565
2566         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2567 }
2568
2569 static ssize_t reserved_clusters_show(struct ext4_attr *a,
2570                                   struct ext4_sb_info *sbi, char *buf)
2571 {
2572         return snprintf(buf, PAGE_SIZE, "%llu\n",
2573                 (unsigned long long) atomic64_read(&sbi->s_resv_clusters));
2574 }
2575
2576 static ssize_t reserved_clusters_store(struct ext4_attr *a,
2577                                    struct ext4_sb_info *sbi,
2578                                    const char *buf, size_t count)
2579 {
2580         unsigned long long val;
2581         int ret;
2582
2583         if (parse_strtoull(buf, -1ULL, &val))
2584                 return -EINVAL;
2585         ret = ext4_reserve_clusters(sbi, val);
2586
2587         return ret ? ret : count;
2588 }
2589
2590 static ssize_t trigger_test_error(struct ext4_attr *a,
2591                                   struct ext4_sb_info *sbi,
2592                                   const char *buf, size_t count)
2593 {
2594         int len = count;
2595
2596         if (!capable(CAP_SYS_ADMIN))
2597                 return -EPERM;
2598
2599         if (len && buf[len-1] == '\n')
2600                 len--;
2601
2602         if (len)
2603                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2604         return count;
2605 }
2606
2607 static ssize_t sbi_deprecated_show(struct ext4_attr *a,
2608                                    struct ext4_sb_info *sbi, char *buf)
2609 {
2610         return snprintf(buf, PAGE_SIZE, "%d\n", a->u.deprecated_val);
2611 }
2612
2613 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2614 static struct ext4_attr ext4_attr_##_name = {                   \
2615         .attr = {.name = __stringify(_name), .mode = _mode },   \
2616         .show   = _show,                                        \
2617         .store  = _store,                                       \
2618         .u = {                                                  \
2619                 .offset = offsetof(struct ext4_sb_info, _elname),\
2620         },                                                      \
2621 }
2622
2623 #define EXT4_ATTR_OFFSET_ES(_name,_mode,_show,_store,_elname)           \
2624 static struct ext4_attr ext4_attr_##_name = {                           \
2625         .attr = {.name = __stringify(_name), .mode = _mode },           \
2626         .show   = _show,                                                \
2627         .store  = _store,                                               \
2628         .u = {                                                          \
2629                 .offset = offsetof(struct ext4_super_block, _elname),   \
2630         },                                                              \
2631 }
2632
2633 #define EXT4_ATTR(name, mode, show, store) \
2634 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2635
2636 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2637 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2638 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2639
2640 #define EXT4_RO_ATTR_ES_UI(name, elname)        \
2641         EXT4_ATTR_OFFSET_ES(name, 0444, es_ui_show, NULL, elname)
2642 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2643         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2644
2645 #define ATTR_LIST(name) &ext4_attr_##name.attr
2646 #define EXT4_DEPRECATED_ATTR(_name, _val)       \
2647 static struct ext4_attr ext4_attr_##_name = {                   \
2648         .attr = {.name = __stringify(_name), .mode = 0444 },    \
2649         .show   = sbi_deprecated_show,                          \
2650         .u = {                                                  \
2651                 .deprecated_val = _val,                         \
2652         },                                                      \
2653 }
2654
2655 EXT4_RO_ATTR(delayed_allocation_blocks);
2656 EXT4_RO_ATTR(session_write_kbytes);
2657 EXT4_RO_ATTR(lifetime_write_kbytes);
2658 EXT4_RW_ATTR(reserved_clusters);
2659 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2660                  inode_readahead_blks_store, s_inode_readahead_blks);
2661 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2662 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2663 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2664 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2665 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2666 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2667 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2668 EXT4_DEPRECATED_ATTR(max_writeback_mb_bump, 128);
2669 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
2670 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2671 EXT4_RW_ATTR_SBI_UI(err_ratelimit_interval_ms, s_err_ratelimit_state.interval);
2672 EXT4_RW_ATTR_SBI_UI(err_ratelimit_burst, s_err_ratelimit_state.burst);
2673 EXT4_RW_ATTR_SBI_UI(warning_ratelimit_interval_ms, s_warning_ratelimit_state.interval);
2674 EXT4_RW_ATTR_SBI_UI(warning_ratelimit_burst, s_warning_ratelimit_state.burst);
2675 EXT4_RW_ATTR_SBI_UI(msg_ratelimit_interval_ms, s_msg_ratelimit_state.interval);
2676 EXT4_RW_ATTR_SBI_UI(msg_ratelimit_burst, s_msg_ratelimit_state.burst);
2677 EXT4_RO_ATTR_ES_UI(errors_count, s_error_count);
2678 EXT4_RO_ATTR_ES_UI(first_error_time, s_first_error_time);
2679 EXT4_RO_ATTR_ES_UI(last_error_time, s_last_error_time);
2680
2681 static struct attribute *ext4_attrs[] = {
2682         ATTR_LIST(delayed_allocation_blocks),
2683         ATTR_LIST(session_write_kbytes),
2684         ATTR_LIST(lifetime_write_kbytes),
2685         ATTR_LIST(reserved_clusters),
2686         ATTR_LIST(inode_readahead_blks),
2687         ATTR_LIST(inode_goal),
2688         ATTR_LIST(mb_stats),
2689         ATTR_LIST(mb_max_to_scan),
2690         ATTR_LIST(mb_min_to_scan),
2691         ATTR_LIST(mb_order2_req),
2692         ATTR_LIST(mb_stream_req),
2693         ATTR_LIST(mb_group_prealloc),
2694         ATTR_LIST(max_writeback_mb_bump),
2695         ATTR_LIST(extent_max_zeroout_kb),
2696         ATTR_LIST(trigger_fs_error),
2697         ATTR_LIST(err_ratelimit_interval_ms),
2698         ATTR_LIST(err_ratelimit_burst),
2699         ATTR_LIST(warning_ratelimit_interval_ms),
2700         ATTR_LIST(warning_ratelimit_burst),
2701         ATTR_LIST(msg_ratelimit_interval_ms),
2702         ATTR_LIST(msg_ratelimit_burst),
2703         ATTR_LIST(errors_count),
2704         ATTR_LIST(first_error_time),
2705         ATTR_LIST(last_error_time),
2706         NULL,
2707 };
2708
2709 /* Features this copy of ext4 supports */
2710 EXT4_INFO_ATTR(lazy_itable_init);
2711 EXT4_INFO_ATTR(batched_discard);
2712 EXT4_INFO_ATTR(meta_bg_resize);
2713 EXT4_INFO_ATTR(encryption);
2714
2715 static struct attribute *ext4_feat_attrs[] = {
2716         ATTR_LIST(lazy_itable_init),
2717         ATTR_LIST(batched_discard),
2718         ATTR_LIST(meta_bg_resize),
2719         ATTR_LIST(encryption),
2720         NULL,
2721 };
2722
2723 static ssize_t ext4_attr_show(struct kobject *kobj,
2724                               struct attribute *attr, char *buf)
2725 {
2726         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2727                                                 s_kobj);
2728         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2729
2730         return a->show ? a->show(a, sbi, buf) : 0;
2731 }
2732
2733 static ssize_t ext4_attr_store(struct kobject *kobj,
2734                                struct attribute *attr,
2735                                const char *buf, size_t len)
2736 {
2737         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2738                                                 s_kobj);
2739         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2740
2741         return a->store ? a->store(a, sbi, buf, len) : 0;
2742 }
2743
2744 static void ext4_sb_release(struct kobject *kobj)
2745 {
2746         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2747                                                 s_kobj);
2748         complete(&sbi->s_kobj_unregister);
2749 }
2750
2751 static const struct sysfs_ops ext4_attr_ops = {
2752         .show   = ext4_attr_show,
2753         .store  = ext4_attr_store,
2754 };
2755
2756 static struct kobj_type ext4_ktype = {
2757         .default_attrs  = ext4_attrs,
2758         .sysfs_ops      = &ext4_attr_ops,
2759         .release        = ext4_sb_release,
2760 };
2761
2762 static void ext4_feat_release(struct kobject *kobj)
2763 {
2764         complete(&ext4_feat->f_kobj_unregister);
2765 }
2766
2767 static ssize_t ext4_feat_show(struct kobject *kobj,
2768                               struct attribute *attr, char *buf)
2769 {
2770         return snprintf(buf, PAGE_SIZE, "supported\n");
2771 }
2772
2773 /*
2774  * We can not use ext4_attr_show/store because it relies on the kobject
2775  * being embedded in the ext4_sb_info structure which is definitely not
2776  * true in this case.
2777  */
2778 static const struct sysfs_ops ext4_feat_ops = {
2779         .show   = ext4_feat_show,
2780         .store  = NULL,
2781 };
2782
2783 static struct kobj_type ext4_feat_ktype = {
2784         .default_attrs  = ext4_feat_attrs,
2785         .sysfs_ops      = &ext4_feat_ops,
2786         .release        = ext4_feat_release,
2787 };
2788
2789 /*
2790  * Check whether this filesystem can be mounted based on
2791  * the features present and the RDONLY/RDWR mount requested.
2792  * Returns 1 if this filesystem can be mounted as requested,
2793  * 0 if it cannot be.
2794  */
2795 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2796 {
2797         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2798                 ext4_msg(sb, KERN_ERR,
2799                         "Couldn't mount because of "
2800                         "unsupported optional features (%x)",
2801                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2802                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2803                 return 0;
2804         }
2805
2806         if (readonly)
2807                 return 1;
2808
2809         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_READONLY)) {
2810                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2811                 sb->s_flags |= MS_RDONLY;
2812                 return 1;
2813         }
2814
2815         /* Check that feature set is OK for a read-write mount */
2816         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2817                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2818                          "unsupported optional features (%x)",
2819                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2820                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2821                 return 0;
2822         }
2823         /*
2824          * Large file size enabled file system can only be mounted
2825          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2826          */
2827         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2828                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2829                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2830                                  "cannot be mounted RDWR without "
2831                                  "CONFIG_LBDAF");
2832                         return 0;
2833                 }
2834         }
2835         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2836             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2837                 ext4_msg(sb, KERN_ERR,
2838                          "Can't support bigalloc feature without "
2839                          "extents feature\n");
2840                 return 0;
2841         }
2842
2843 #ifndef CONFIG_QUOTA
2844         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2845             !readonly) {
2846                 ext4_msg(sb, KERN_ERR,
2847                          "Filesystem with quota feature cannot be mounted RDWR "
2848                          "without CONFIG_QUOTA");
2849                 return 0;
2850         }
2851 #endif  /* CONFIG_QUOTA */
2852         return 1;
2853 }
2854
2855 /*
2856  * This function is called once a day if we have errors logged
2857  * on the file system
2858  */
2859 static void print_daily_error_info(unsigned long arg)
2860 {
2861         struct super_block *sb = (struct super_block *) arg;
2862         struct ext4_sb_info *sbi;
2863         struct ext4_super_block *es;
2864
2865         sbi = EXT4_SB(sb);
2866         es = sbi->s_es;
2867
2868         if (es->s_error_count)
2869                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2870                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2871                          le32_to_cpu(es->s_error_count));
2872         if (es->s_first_error_time) {
2873                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2874                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2875                        (int) sizeof(es->s_first_error_func),
2876                        es->s_first_error_func,
2877                        le32_to_cpu(es->s_first_error_line));
2878                 if (es->s_first_error_ino)
2879                         printk(": inode %u",
2880                                le32_to_cpu(es->s_first_error_ino));
2881                 if (es->s_first_error_block)
2882                         printk(": block %llu", (unsigned long long)
2883                                le64_to_cpu(es->s_first_error_block));
2884                 printk("\n");
2885         }
2886         if (es->s_last_error_time) {
2887                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2888                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2889                        (int) sizeof(es->s_last_error_func),
2890                        es->s_last_error_func,
2891                        le32_to_cpu(es->s_last_error_line));
2892                 if (es->s_last_error_ino)
2893                         printk(": inode %u",
2894                                le32_to_cpu(es->s_last_error_ino));
2895                 if (es->s_last_error_block)
2896                         printk(": block %llu", (unsigned long long)
2897                                le64_to_cpu(es->s_last_error_block));
2898                 printk("\n");
2899         }
2900         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2901 }
2902
2903 /* Find next suitable group and run ext4_init_inode_table */
2904 static int ext4_run_li_request(struct ext4_li_request *elr)
2905 {
2906         struct ext4_group_desc *gdp = NULL;
2907         ext4_group_t group, ngroups;
2908         struct super_block *sb;
2909         unsigned long timeout = 0;
2910         int ret = 0;
2911
2912         sb = elr->lr_super;
2913         ngroups = EXT4_SB(sb)->s_groups_count;
2914
2915         sb_start_write(sb);
2916         for (group = elr->lr_next_group; group < ngroups; group++) {
2917                 gdp = ext4_get_group_desc(sb, group, NULL);
2918                 if (!gdp) {
2919                         ret = 1;
2920                         break;
2921                 }
2922
2923                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2924                         break;
2925         }
2926
2927         if (group >= ngroups)
2928                 ret = 1;
2929
2930         if (!ret) {
2931                 timeout = jiffies;
2932                 ret = ext4_init_inode_table(sb, group,
2933                                             elr->lr_timeout ? 0 : 1);
2934                 if (elr->lr_timeout == 0) {
2935                         timeout = (jiffies - timeout) *
2936                                   elr->lr_sbi->s_li_wait_mult;
2937                         elr->lr_timeout = timeout;
2938                 }
2939                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2940                 elr->lr_next_group = group + 1;
2941         }
2942         sb_end_write(sb);
2943
2944         return ret;
2945 }
2946
2947 /*
2948  * Remove lr_request from the list_request and free the
2949  * request structure. Should be called with li_list_mtx held
2950  */
2951 static void ext4_remove_li_request(struct ext4_li_request *elr)
2952 {
2953         struct ext4_sb_info *sbi;
2954
2955         if (!elr)
2956                 return;
2957
2958         sbi = elr->lr_sbi;
2959
2960         list_del(&elr->lr_request);
2961         sbi->s_li_request = NULL;
2962         kfree(elr);
2963 }
2964
2965 static void ext4_unregister_li_request(struct super_block *sb)
2966 {
2967         mutex_lock(&ext4_li_mtx);
2968         if (!ext4_li_info) {
2969                 mutex_unlock(&ext4_li_mtx);
2970                 return;
2971         }
2972
2973         mutex_lock(&ext4_li_info->li_list_mtx);
2974         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2975         mutex_unlock(&ext4_li_info->li_list_mtx);
2976         mutex_unlock(&ext4_li_mtx);
2977 }
2978
2979 static struct task_struct *ext4_lazyinit_task;
2980
2981 /*
2982  * This is the function where ext4lazyinit thread lives. It walks
2983  * through the request list searching for next scheduled filesystem.
2984  * When such a fs is found, run the lazy initialization request
2985  * (ext4_rn_li_request) and keep track of the time spend in this
2986  * function. Based on that time we compute next schedule time of
2987  * the request. When walking through the list is complete, compute
2988  * next waking time and put itself into sleep.
2989  */
2990 static int ext4_lazyinit_thread(void *arg)
2991 {
2992         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2993         struct list_head *pos, *n;
2994         struct ext4_li_request *elr;
2995         unsigned long next_wakeup, cur;
2996
2997         BUG_ON(NULL == eli);
2998
2999 cont_thread:
3000         while (true) {
3001                 next_wakeup = MAX_JIFFY_OFFSET;
3002
3003                 mutex_lock(&eli->li_list_mtx);
3004                 if (list_empty(&eli->li_request_list)) {
3005                         mutex_unlock(&eli->li_list_mtx);
3006                         goto exit_thread;
3007                 }
3008
3009                 list_for_each_safe(pos, n, &eli->li_request_list) {
3010                         elr = list_entry(pos, struct ext4_li_request,
3011                                          lr_request);
3012
3013                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
3014                                 if (ext4_run_li_request(elr) != 0) {
3015                                         /* error, remove the lazy_init job */
3016                                         ext4_remove_li_request(elr);
3017                                         continue;
3018                                 }
3019                         }
3020
3021                         if (time_before(elr->lr_next_sched, next_wakeup))
3022                                 next_wakeup = elr->lr_next_sched;
3023                 }
3024                 mutex_unlock(&eli->li_list_mtx);
3025
3026                 try_to_freeze();
3027
3028                 cur = jiffies;
3029                 if ((time_after_eq(cur, next_wakeup)) ||
3030                     (MAX_JIFFY_OFFSET == next_wakeup)) {
3031                         cond_resched();
3032                         continue;
3033                 }
3034
3035                 schedule_timeout_interruptible(next_wakeup - cur);
3036
3037                 if (kthread_should_stop()) {
3038                         ext4_clear_request_list();
3039                         goto exit_thread;
3040                 }
3041         }
3042
3043 exit_thread:
3044         /*
3045          * It looks like the request list is empty, but we need
3046          * to check it under the li_list_mtx lock, to prevent any
3047          * additions into it, and of course we should lock ext4_li_mtx
3048          * to atomically free the list and ext4_li_info, because at
3049          * this point another ext4 filesystem could be registering
3050          * new one.
3051          */
3052         mutex_lock(&ext4_li_mtx);
3053         mutex_lock(&eli->li_list_mtx);
3054         if (!list_empty(&eli->li_request_list)) {
3055                 mutex_unlock(&eli->li_list_mtx);
3056                 mutex_unlock(&ext4_li_mtx);
3057                 goto cont_thread;
3058         }
3059         mutex_unlock(&eli->li_list_mtx);
3060         kfree(ext4_li_info);
3061         ext4_li_info = NULL;
3062         mutex_unlock(&ext4_li_mtx);
3063
3064         return 0;
3065 }
3066
3067 static void ext4_clear_request_list(void)
3068 {
3069         struct list_head *pos, *n;
3070         struct ext4_li_request *elr;
3071
3072         mutex_lock(&ext4_li_info->li_list_mtx);
3073         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3074                 elr = list_entry(pos, struct ext4_li_request,
3075                                  lr_request);
3076                 ext4_remove_li_request(elr);
3077         }
3078         mutex_unlock(&ext4_li_info->li_list_mtx);
3079 }
3080
3081 static int ext4_run_lazyinit_thread(void)
3082 {
3083         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3084                                          ext4_li_info, "ext4lazyinit");
3085         if (IS_ERR(ext4_lazyinit_task)) {
3086                 int err = PTR_ERR(ext4_lazyinit_task);
3087                 ext4_clear_request_list();
3088                 kfree(ext4_li_info);
3089                 ext4_li_info = NULL;
3090                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3091                                  "initialization thread\n",
3092                                  err);
3093                 return err;
3094         }
3095         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3096         return 0;
3097 }
3098
3099 /*
3100  * Check whether it make sense to run itable init. thread or not.
3101  * If there is at least one uninitialized inode table, return
3102  * corresponding group number, else the loop goes through all
3103  * groups and return total number of groups.
3104  */
3105 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3106 {
3107         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3108         struct ext4_group_desc *gdp = NULL;
3109
3110         for (group = 0; group < ngroups; group++) {
3111                 gdp = ext4_get_group_desc(sb, group, NULL);
3112                 if (!gdp)
3113                         continue;
3114
3115                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3116                         break;
3117         }
3118
3119         return group;
3120 }
3121
3122 static int ext4_li_info_new(void)
3123 {
3124         struct ext4_lazy_init *eli = NULL;
3125
3126         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3127         if (!eli)
3128                 return -ENOMEM;
3129
3130         INIT_LIST_HEAD(&eli->li_request_list);
3131         mutex_init(&eli->li_list_mtx);
3132
3133         eli->li_state |= EXT4_LAZYINIT_QUIT;
3134
3135         ext4_li_info = eli;
3136
3137         return 0;
3138 }
3139
3140 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3141                                             ext4_group_t start)
3142 {
3143         struct ext4_sb_info *sbi = EXT4_SB(sb);
3144         struct ext4_li_request *elr;
3145
3146         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3147         if (!elr)
3148                 return NULL;
3149
3150         elr->lr_super = sb;
3151         elr->lr_sbi = sbi;
3152         elr->lr_next_group = start;
3153
3154         /*
3155          * Randomize first schedule time of the request to
3156          * spread the inode table initialization requests
3157          * better.
3158          */
3159         elr->lr_next_sched = jiffies + (prandom_u32() %
3160                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3161         return elr;
3162 }
3163
3164 int ext4_register_li_request(struct super_block *sb,
3165                              ext4_group_t first_not_zeroed)
3166 {
3167         struct ext4_sb_info *sbi = EXT4_SB(sb);
3168         struct ext4_li_request *elr = NULL;
3169         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3170         int ret = 0;
3171
3172         mutex_lock(&ext4_li_mtx);
3173         if (sbi->s_li_request != NULL) {
3174                 /*
3175                  * Reset timeout so it can be computed again, because
3176                  * s_li_wait_mult might have changed.
3177                  */
3178                 sbi->s_li_request->lr_timeout = 0;
3179                 goto out;
3180         }
3181
3182         if (first_not_zeroed == ngroups ||
3183             (sb->s_flags & MS_RDONLY) ||
3184             !test_opt(sb, INIT_INODE_TABLE))
3185                 goto out;
3186
3187         elr = ext4_li_request_new(sb, first_not_zeroed);
3188         if (!elr) {
3189                 ret = -ENOMEM;
3190                 goto out;
3191         }
3192
3193         if (NULL == ext4_li_info) {
3194                 ret = ext4_li_info_new();
3195                 if (ret)
3196                         goto out;
3197         }
3198
3199         mutex_lock(&ext4_li_info->li_list_mtx);
3200         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3201         mutex_unlock(&ext4_li_info->li_list_mtx);
3202
3203         sbi->s_li_request = elr;
3204         /*
3205          * set elr to NULL here since it has been inserted to
3206          * the request_list and the removal and free of it is
3207          * handled by ext4_clear_request_list from now on.
3208          */
3209         elr = NULL;
3210
3211         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3212                 ret = ext4_run_lazyinit_thread();
3213                 if (ret)
3214                         goto out;
3215         }
3216 out:
3217         mutex_unlock(&ext4_li_mtx);
3218         if (ret)
3219                 kfree(elr);
3220         return ret;
3221 }
3222
3223 /*
3224  * We do not need to lock anything since this is called on
3225  * module unload.
3226  */
3227 static void ext4_destroy_lazyinit_thread(void)
3228 {
3229         /*
3230          * If thread exited earlier
3231          * there's nothing to be done.
3232          */
3233         if (!ext4_li_info || !ext4_lazyinit_task)
3234                 return;
3235
3236         kthread_stop(ext4_lazyinit_task);
3237 }
3238
3239 static int set_journal_csum_feature_set(struct super_block *sb)
3240 {
3241         int ret = 1;
3242         int compat, incompat;
3243         struct ext4_sb_info *sbi = EXT4_SB(sb);
3244
3245         if (ext4_has_metadata_csum(sb)) {
3246                 /* journal checksum v3 */
3247                 compat = 0;
3248                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3249         } else {
3250                 /* journal checksum v1 */
3251                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3252                 incompat = 0;
3253         }
3254
3255         jbd2_journal_clear_features(sbi->s_journal,
3256                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3257                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3258                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3259         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3260                 ret = jbd2_journal_set_features(sbi->s_journal,
3261                                 compat, 0,
3262                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3263                                 incompat);
3264         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3265                 ret = jbd2_journal_set_features(sbi->s_journal,
3266                                 compat, 0,
3267                                 incompat);
3268                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3269                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3270         } else {
3271                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3272                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3273         }
3274
3275         return ret;
3276 }
3277
3278 /*
3279  * Note: calculating the overhead so we can be compatible with
3280  * historical BSD practice is quite difficult in the face of
3281  * clusters/bigalloc.  This is because multiple metadata blocks from
3282  * different block group can end up in the same allocation cluster.
3283  * Calculating the exact overhead in the face of clustered allocation
3284  * requires either O(all block bitmaps) in memory or O(number of block
3285  * groups**2) in time.  We will still calculate the superblock for
3286  * older file systems --- and if we come across with a bigalloc file
3287  * system with zero in s_overhead_clusters the estimate will be close to
3288  * correct especially for very large cluster sizes --- but for newer
3289  * file systems, it's better to calculate this figure once at mkfs
3290  * time, and store it in the superblock.  If the superblock value is
3291  * present (even for non-bigalloc file systems), we will use it.
3292  */
3293 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3294                           char *buf)
3295 {
3296         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3297         struct ext4_group_desc  *gdp;
3298         ext4_fsblk_t            first_block, last_block, b;
3299         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3300         int                     s, j, count = 0;
3301
3302         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3303                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3304                         sbi->s_itb_per_group + 2);
3305
3306         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3307                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3308         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3309         for (i = 0; i < ngroups; i++) {
3310                 gdp = ext4_get_group_desc(sb, i, NULL);
3311                 b = ext4_block_bitmap(sb, gdp);
3312                 if (b >= first_block && b <= last_block) {
3313                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3314                         count++;
3315                 }
3316                 b = ext4_inode_bitmap(sb, gdp);
3317                 if (b >= first_block && b <= last_block) {
3318                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3319                         count++;
3320                 }
3321                 b = ext4_inode_table(sb, gdp);
3322                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3323                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3324                                 int c = EXT4_B2C(sbi, b - first_block);
3325                                 ext4_set_bit(c, buf);
3326                                 count++;
3327                         }
3328                 if (i != grp)
3329                         continue;
3330                 s = 0;
3331                 if (ext4_bg_has_super(sb, grp)) {
3332                         ext4_set_bit(s++, buf);
3333                         count++;
3334                 }
3335                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3336                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3337                         count++;
3338                 }
3339         }
3340         if (!count)
3341                 return 0;
3342         return EXT4_CLUSTERS_PER_GROUP(sb) -
3343                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3344 }
3345
3346 /*
3347  * Compute the overhead and stash it in sbi->s_overhead
3348  */
3349 int ext4_calculate_overhead(struct super_block *sb)
3350 {
3351         struct ext4_sb_info *sbi = EXT4_SB(sb);
3352         struct ext4_super_block *es = sbi->s_es;
3353         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3354         ext4_fsblk_t overhead = 0;
3355         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3356
3357         if (!buf)
3358                 return -ENOMEM;
3359
3360         /*
3361          * Compute the overhead (FS structures).  This is constant
3362          * for a given filesystem unless the number of block groups
3363          * changes so we cache the previous value until it does.
3364          */
3365
3366         /*
3367          * All of the blocks before first_data_block are overhead
3368          */
3369         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3370
3371         /*
3372          * Add the overhead found in each block group
3373          */
3374         for (i = 0; i < ngroups; i++) {
3375                 int blks;
3376
3377                 blks = count_overhead(sb, i, buf);
3378                 overhead += blks;
3379                 if (blks)
3380                         memset(buf, 0, PAGE_SIZE);
3381                 cond_resched();
3382         }
3383         /* Add the internal journal blocks as well */
3384         if (sbi->s_journal && !sbi->journal_bdev)
3385                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3386
3387         sbi->s_overhead = overhead;
3388         smp_wmb();
3389         free_page((unsigned long) buf);
3390         return 0;
3391 }
3392
3393
3394 static ext4_fsblk_t ext4_calculate_resv_clusters(struct super_block *sb)
3395 {
3396         ext4_fsblk_t resv_clusters;
3397
3398         /*
3399          * There's no need to reserve anything when we aren't using extents.
3400          * The space estimates are exact, there are no unwritten extents,
3401          * hole punching doesn't need new metadata... This is needed especially
3402          * to keep ext2/3 backward compatibility.
3403          */
3404         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
3405                 return 0;
3406         /*
3407          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3408          * This should cover the situations where we can not afford to run
3409          * out of space like for example punch hole, or converting
3410          * unwritten extents in delalloc path. In most cases such
3411          * allocation would require 1, or 2 blocks, higher numbers are
3412          * very rare.
3413          */
3414         resv_clusters = ext4_blocks_count(EXT4_SB(sb)->s_es) >>
3415                         EXT4_SB(sb)->s_cluster_bits;
3416
3417         do_div(resv_clusters, 50);
3418         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3419
3420         return resv_clusters;
3421 }
3422
3423
3424 static int ext4_reserve_clusters(struct ext4_sb_info *sbi, ext4_fsblk_t count)
3425 {
3426         ext4_fsblk_t clusters = ext4_blocks_count(sbi->s_es) >>
3427                                 sbi->s_cluster_bits;
3428
3429         if (count >= clusters)
3430                 return -EINVAL;
3431
3432         atomic64_set(&sbi->s_resv_clusters, count);
3433         return 0;
3434 }
3435
3436 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3437 {
3438         char *orig_data = kstrdup(data, GFP_KERNEL);
3439         struct buffer_head *bh;
3440         struct ext4_super_block *es = NULL;
3441         struct ext4_sb_info *sbi;
3442         ext4_fsblk_t block;
3443         ext4_fsblk_t sb_block = get_sb_block(&data);
3444         ext4_fsblk_t logical_sb_block;
3445         unsigned long offset = 0;
3446         unsigned long journal_devnum = 0;
3447         unsigned long def_mount_opts;
3448         struct inode *root;
3449         const char *descr;
3450         int ret = -ENOMEM;
3451         int blocksize, clustersize;
3452         unsigned int db_count;
3453         unsigned int i;
3454         int needs_recovery, has_huge_files, has_bigalloc;
3455         __u64 blocks_count;
3456         int err = 0;
3457         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3458         ext4_group_t first_not_zeroed;
3459
3460         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3461         if (!sbi)
3462                 goto out_free_orig;
3463
3464         sbi->s_blockgroup_lock =
3465                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3466         if (!sbi->s_blockgroup_lock) {
3467                 kfree(sbi);
3468                 goto out_free_orig;
3469         }
3470         sb->s_fs_info = sbi;
3471         sbi->s_sb = sb;
3472         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3473         sbi->s_sb_block = sb_block;
3474         if (sb->s_bdev->bd_part)
3475                 sbi->s_sectors_written_start =
3476                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3477
3478         /* Cleanup superblock name */
3479         strreplace(sb->s_id, '/', '!');
3480
3481         /* -EINVAL is default */
3482         ret = -EINVAL;
3483         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3484         if (!blocksize) {
3485                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3486                 goto out_fail;
3487         }
3488
3489         /*
3490          * The ext4 superblock will not be buffer aligned for other than 1kB
3491          * block sizes.  We need to calculate the offset from buffer start.
3492          */
3493         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3494                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3495                 offset = do_div(logical_sb_block, blocksize);
3496         } else {
3497                 logical_sb_block = sb_block;
3498         }
3499
3500         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3501                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3502                 goto out_fail;
3503         }
3504         /*
3505          * Note: s_es must be initialized as soon as possible because
3506          *       some ext4 macro-instructions depend on its value
3507          */
3508         es = (struct ext4_super_block *) (bh->b_data + offset);
3509         sbi->s_es = es;
3510         sb->s_magic = le16_to_cpu(es->s_magic);
3511         if (sb->s_magic != EXT4_SUPER_MAGIC)
3512                 goto cantfind_ext4;
3513         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3514
3515         /* Warn if metadata_csum and gdt_csum are both set. */
3516         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3517                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3518             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3519                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3520                              "redundant flags; please run fsck.");
3521
3522         /* Check for a known checksum algorithm */
3523         if (!ext4_verify_csum_type(sb, es)) {
3524                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3525                          "unknown checksum algorithm.");
3526                 silent = 1;
3527                 goto cantfind_ext4;
3528         }
3529
3530         /* Load the checksum driver */
3531         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3532                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3533                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3534                 if (IS_ERR(sbi->s_chksum_driver)) {
3535                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3536                         ret = PTR_ERR(sbi->s_chksum_driver);
3537                         sbi->s_chksum_driver = NULL;
3538                         goto failed_mount;
3539                 }
3540         }
3541
3542         /* Check superblock checksum */
3543         if (!ext4_superblock_csum_verify(sb, es)) {
3544                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3545                          "invalid superblock checksum.  Run e2fsck?");
3546                 silent = 1;
3547                 goto cantfind_ext4;
3548         }
3549
3550         /* Precompute checksum seed for all metadata */
3551         if (ext4_has_metadata_csum(sb))
3552                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3553                                                sizeof(es->s_uuid));
3554
3555         /* Set defaults before we parse the mount options */
3556         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3557         set_opt(sb, INIT_INODE_TABLE);
3558         if (def_mount_opts & EXT4_DEFM_DEBUG)
3559                 set_opt(sb, DEBUG);
3560         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3561                 set_opt(sb, GRPID);
3562         if (def_mount_opts & EXT4_DEFM_UID16)
3563                 set_opt(sb, NO_UID32);
3564         /* xattr user namespace & acls are now defaulted on */
3565         set_opt(sb, XATTR_USER);
3566 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3567         set_opt(sb, POSIX_ACL);
3568 #endif
3569         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3570         if (ext4_has_metadata_csum(sb))
3571                 set_opt(sb, JOURNAL_CHECKSUM);
3572
3573         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3574                 set_opt(sb, JOURNAL_DATA);
3575         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3576                 set_opt(sb, ORDERED_DATA);
3577         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3578                 set_opt(sb, WRITEBACK_DATA);
3579
3580         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3581                 set_opt(sb, ERRORS_PANIC);
3582         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3583                 set_opt(sb, ERRORS_CONT);
3584         else
3585                 set_opt(sb, ERRORS_RO);
3586         /* block_validity enabled by default; disable with noblock_validity */
3587         set_opt(sb, BLOCK_VALIDITY);
3588         if (def_mount_opts & EXT4_DEFM_DISCARD)
3589                 set_opt(sb, DISCARD);
3590
3591         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3592         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3593         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3594         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3595         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3596
3597         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3598                 set_opt(sb, BARRIER);
3599
3600         /*
3601          * enable delayed allocation by default
3602          * Use -o nodelalloc to turn it off
3603          */
3604         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3605             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3606                 set_opt(sb, DELALLOC);
3607
3608         /*
3609          * set default s_li_wait_mult for lazyinit, for the case there is
3610          * no mount option specified.
3611          */
3612         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3613
3614         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3615                            &journal_devnum, &journal_ioprio, 0)) {
3616                 ext4_msg(sb, KERN_WARNING,
3617                          "failed to parse options in superblock: %s",
3618                          sbi->s_es->s_mount_opts);
3619         }
3620         sbi->s_def_mount_opt = sbi->s_mount_opt;
3621         if (!parse_options((char *) data, sb, &journal_devnum,
3622                            &journal_ioprio, 0))
3623                 goto failed_mount;
3624
3625         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3626                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3627                             "with data=journal disables delayed "
3628                             "allocation and O_DIRECT support!\n");
3629                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3630                         ext4_msg(sb, KERN_ERR, "can't mount with "
3631                                  "both data=journal and delalloc");
3632                         goto failed_mount;
3633                 }
3634                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3635                         ext4_msg(sb, KERN_ERR, "can't mount with "
3636                                  "both data=journal and dioread_nolock");
3637                         goto failed_mount;
3638                 }
3639                 if (test_opt(sb, DAX)) {
3640                         ext4_msg(sb, KERN_ERR, "can't mount with "
3641                                  "both data=journal and dax");
3642                         goto failed_mount;
3643                 }
3644                 if (test_opt(sb, DELALLOC))
3645                         clear_opt(sb, DELALLOC);
3646         }
3647
3648         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3649                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3650
3651         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3652             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3653              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3654              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3655                 ext4_msg(sb, KERN_WARNING,
3656                        "feature flags set on rev 0 fs, "
3657                        "running e2fsck is recommended");
3658
3659         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3660                 set_opt2(sb, HURD_COMPAT);
3661                 if (EXT4_HAS_INCOMPAT_FEATURE(sb,
3662                                               EXT4_FEATURE_INCOMPAT_64BIT)) {
3663                         ext4_msg(sb, KERN_ERR,
3664                                  "The Hurd can't support 64-bit file systems");
3665                         goto failed_mount;
3666                 }
3667         }
3668
3669         if (IS_EXT2_SB(sb)) {
3670                 if (ext2_feature_set_ok(sb))
3671                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3672                                  "using the ext4 subsystem");
3673                 else {
3674                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3675                                  "to feature incompatibilities");
3676                         goto failed_mount;
3677                 }
3678         }
3679
3680         if (IS_EXT3_SB(sb)) {
3681                 if (ext3_feature_set_ok(sb))
3682                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3683                                  "using the ext4 subsystem");
3684                 else {
3685                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3686                                  "to feature incompatibilities");
3687                         goto failed_mount;
3688                 }
3689         }
3690
3691         /*
3692          * Check feature flags regardless of the revision level, since we
3693          * previously didn't change the revision level when setting the flags,
3694          * so there is a chance incompat flags are set on a rev 0 filesystem.
3695          */
3696         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3697                 goto failed_mount;
3698
3699         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3700         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3701             blocksize > EXT4_MAX_BLOCK_SIZE) {
3702                 ext4_msg(sb, KERN_ERR,
3703                        "Unsupported filesystem blocksize %d", blocksize);
3704                 goto failed_mount;
3705         }
3706
3707         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3708                 if (blocksize != PAGE_SIZE) {
3709                         ext4_msg(sb, KERN_ERR,
3710                                         "error: unsupported blocksize for dax");
3711                         goto failed_mount;
3712                 }
3713                 if (!sb->s_bdev->bd_disk->fops->direct_access) {
3714                         ext4_msg(sb, KERN_ERR,
3715                                         "error: device does not support dax");
3716                         goto failed_mount;
3717                 }
3718         }
3719
3720         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_ENCRYPT) &&
3721             es->s_encryption_level) {
3722                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3723                          es->s_encryption_level);
3724                 goto failed_mount;
3725         }
3726
3727         if (sb->s_blocksize != blocksize) {
3728                 /* Validate the filesystem blocksize */
3729                 if (!sb_set_blocksize(sb, blocksize)) {
3730                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3731                                         blocksize);
3732                         goto failed_mount;
3733                 }
3734
3735                 brelse(bh);
3736                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3737                 offset = do_div(logical_sb_block, blocksize);
3738                 bh = sb_bread_unmovable(sb, logical_sb_block);
3739                 if (!bh) {
3740                         ext4_msg(sb, KERN_ERR,
3741                                "Can't read superblock on 2nd try");
3742                         goto failed_mount;
3743                 }
3744                 es = (struct ext4_super_block *)(bh->b_data + offset);
3745                 sbi->s_es = es;
3746                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3747                         ext4_msg(sb, KERN_ERR,
3748                                "Magic mismatch, very weird!");
3749                         goto failed_mount;
3750                 }
3751         }
3752
3753         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3754                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3755         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3756                                                       has_huge_files);
3757         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3758
3759         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3760                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3761                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3762         } else {
3763                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3764                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3765                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3766                     (!is_power_of_2(sbi->s_inode_size)) ||
3767                     (sbi->s_inode_size > blocksize)) {
3768                         ext4_msg(sb, KERN_ERR,
3769                                "unsupported inode size: %d",
3770                                sbi->s_inode_size);
3771                         goto failed_mount;
3772                 }
3773                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3774                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3775         }
3776
3777         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3778         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3779                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3780                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3781                     !is_power_of_2(sbi->s_desc_size)) {
3782                         ext4_msg(sb, KERN_ERR,
3783                                "unsupported descriptor size %lu",
3784                                sbi->s_desc_size);
3785                         goto failed_mount;
3786                 }
3787         } else
3788                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3789
3790         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3791         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3792         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3793                 goto cantfind_ext4;
3794
3795         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3796         if (sbi->s_inodes_per_block == 0)
3797                 goto cantfind_ext4;
3798         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3799                                         sbi->s_inodes_per_block;
3800         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3801         sbi->s_sbh = bh;
3802         sbi->s_mount_state = le16_to_cpu(es->s_state);
3803         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3804         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3805
3806         for (i = 0; i < 4; i++)
3807                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3808         sbi->s_def_hash_version = es->s_def_hash_version;
3809         if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX)) {
3810                 i = le32_to_cpu(es->s_flags);
3811                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3812                         sbi->s_hash_unsigned = 3;
3813                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3814 #ifdef __CHAR_UNSIGNED__
3815                         if (!(sb->s_flags & MS_RDONLY))
3816                                 es->s_flags |=
3817                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3818                         sbi->s_hash_unsigned = 3;
3819 #else
3820                         if (!(sb->s_flags & MS_RDONLY))
3821                                 es->s_flags |=
3822                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3823 #endif
3824                 }
3825         }
3826
3827         /* Handle clustersize */
3828         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3829         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3830                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3831         if (has_bigalloc) {
3832                 if (clustersize < blocksize) {
3833                         ext4_msg(sb, KERN_ERR,
3834                                  "cluster size (%d) smaller than "
3835                                  "block size (%d)", clustersize, blocksize);
3836                         goto failed_mount;
3837                 }
3838                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3839                         le32_to_cpu(es->s_log_block_size);
3840                 sbi->s_clusters_per_group =
3841                         le32_to_cpu(es->s_clusters_per_group);
3842                 if (sbi->s_clusters_per_group > blocksize * 8) {
3843                         ext4_msg(sb, KERN_ERR,
3844                                  "#clusters per group too big: %lu",
3845                                  sbi->s_clusters_per_group);
3846                         goto failed_mount;
3847                 }
3848                 if (sbi->s_blocks_per_group !=
3849                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3850                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3851                                  "clusters per group (%lu) inconsistent",
3852                                  sbi->s_blocks_per_group,
3853                                  sbi->s_clusters_per_group);
3854                         goto failed_mount;
3855                 }
3856         } else {
3857                 if (clustersize != blocksize) {
3858                         ext4_warning(sb, "fragment/cluster size (%d) != "
3859                                      "block size (%d)", clustersize,
3860                                      blocksize);
3861                         clustersize = blocksize;
3862                 }
3863                 if (sbi->s_blocks_per_group > blocksize * 8) {
3864                         ext4_msg(sb, KERN_ERR,
3865                                  "#blocks per group too big: %lu",
3866                                  sbi->s_blocks_per_group);
3867                         goto failed_mount;
3868                 }
3869                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3870                 sbi->s_cluster_bits = 0;
3871         }
3872         sbi->s_cluster_ratio = clustersize / blocksize;
3873
3874         if (sbi->s_inodes_per_group > blocksize * 8) {
3875                 ext4_msg(sb, KERN_ERR,
3876                        "#inodes per group too big: %lu",
3877                        sbi->s_inodes_per_group);
3878                 goto failed_mount;
3879         }
3880
3881         /* Do we have standard group size of clustersize * 8 blocks ? */
3882         if (sbi->s_blocks_per_group == clustersize << 3)
3883                 set_opt2(sb, STD_GROUP_SIZE);
3884
3885         /*
3886          * Test whether we have more sectors than will fit in sector_t,
3887          * and whether the max offset is addressable by the page cache.
3888          */
3889         err = generic_check_addressable(sb->s_blocksize_bits,
3890                                         ext4_blocks_count(es));
3891         if (err) {
3892                 ext4_msg(sb, KERN_ERR, "filesystem"
3893                          " too large to mount safely on this system");
3894                 if (sizeof(sector_t) < 8)
3895                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3896                 goto failed_mount;
3897         }
3898
3899         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3900                 goto cantfind_ext4;
3901
3902         /* check blocks count against device size */
3903         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3904         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3905                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3906                        "exceeds size of device (%llu blocks)",
3907                        ext4_blocks_count(es), blocks_count);
3908                 goto failed_mount;
3909         }
3910
3911         /*
3912          * It makes no sense for the first data block to be beyond the end
3913          * of the filesystem.
3914          */
3915         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3916                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3917                          "block %u is beyond end of filesystem (%llu)",
3918                          le32_to_cpu(es->s_first_data_block),
3919                          ext4_blocks_count(es));
3920                 goto failed_mount;
3921         }
3922         blocks_count = (ext4_blocks_count(es) -
3923                         le32_to_cpu(es->s_first_data_block) +
3924                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3925         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3926         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3927                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3928                        "(block count %llu, first data block %u, "
3929                        "blocks per group %lu)", sbi->s_groups_count,
3930                        ext4_blocks_count(es),
3931                        le32_to_cpu(es->s_first_data_block),
3932                        EXT4_BLOCKS_PER_GROUP(sb));
3933                 goto failed_mount;
3934         }
3935         sbi->s_groups_count = blocks_count;
3936         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3937                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3938         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3939                    EXT4_DESC_PER_BLOCK(sb);
3940         sbi->s_group_desc = ext4_kvmalloc(db_count *
3941                                           sizeof(struct buffer_head *),
3942                                           GFP_KERNEL);
3943         if (sbi->s_group_desc == NULL) {
3944                 ext4_msg(sb, KERN_ERR, "not enough memory");
3945                 ret = -ENOMEM;
3946                 goto failed_mount;
3947         }
3948
3949         if (ext4_proc_root)
3950                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3951
3952         if (sbi->s_proc)
3953                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3954                                  &ext4_seq_options_fops, sb);
3955
3956         bgl_lock_init(sbi->s_blockgroup_lock);
3957
3958         for (i = 0; i < db_count; i++) {
3959                 block = descriptor_loc(sb, logical_sb_block, i);
3960                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
3961                 if (!sbi->s_group_desc[i]) {
3962                         ext4_msg(sb, KERN_ERR,
3963                                "can't read group descriptor %d", i);
3964                         db_count = i;
3965                         goto failed_mount2;
3966                 }
3967         }
3968         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3969                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3970                 goto failed_mount2;
3971         }
3972
3973         sbi->s_gdb_count = db_count;
3974         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3975         spin_lock_init(&sbi->s_next_gen_lock);
3976
3977         setup_timer(&sbi->s_err_report, print_daily_error_info,
3978                 (unsigned long) sb);
3979
3980         /* Register extent status tree shrinker */
3981         if (ext4_es_register_shrinker(sbi))
3982                 goto failed_mount3;
3983
3984         sbi->s_stripe = ext4_get_stripe_size(sbi);
3985         sbi->s_extent_max_zeroout_kb = 32;
3986
3987         /*
3988          * set up enough so that it can read an inode
3989          */
3990         sb->s_op = &ext4_sops;
3991         sb->s_export_op = &ext4_export_ops;
3992         sb->s_xattr = ext4_xattr_handlers;
3993 #ifdef CONFIG_QUOTA
3994         sb->dq_op = &ext4_quota_operations;
3995         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
3996                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
3997         else
3998                 sb->s_qcop = &ext4_qctl_operations;
3999         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
4000 #endif
4001         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
4002
4003         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
4004         mutex_init(&sbi->s_orphan_lock);
4005
4006         sb->s_root = NULL;
4007
4008         needs_recovery = (es->s_last_orphan != 0 ||
4009                           EXT4_HAS_INCOMPAT_FEATURE(sb,
4010                                     EXT4_FEATURE_INCOMPAT_RECOVER));
4011
4012         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
4013             !(sb->s_flags & MS_RDONLY))
4014                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
4015                         goto failed_mount3a;
4016
4017         /*
4018          * The first inode we look at is the journal inode.  Don't try
4019          * root first: it may be modified in the journal!
4020          */
4021         if (!test_opt(sb, NOLOAD) &&
4022             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4023                 if (ext4_load_journal(sb, es, journal_devnum))
4024                         goto failed_mount3a;
4025         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
4026               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4027                 ext4_msg(sb, KERN_ERR, "required journal recovery "
4028                        "suppressed and not mounted read-only");
4029                 goto failed_mount_wq;
4030         } else {
4031                 clear_opt(sb, DATA_FLAGS);
4032                 sbi->s_journal = NULL;
4033                 needs_recovery = 0;
4034                 goto no_journal;
4035         }
4036
4037         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
4038             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4039                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4040                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4041                 goto failed_mount_wq;
4042         }
4043
4044         if (!set_journal_csum_feature_set(sb)) {
4045                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4046                          "feature set");
4047                 goto failed_mount_wq;
4048         }
4049
4050         /* We have now updated the journal if required, so we can
4051          * validate the data journaling mode. */
4052         switch (test_opt(sb, DATA_FLAGS)) {
4053         case 0:
4054                 /* No mode set, assume a default based on the journal
4055                  * capabilities: ORDERED_DATA if the journal can
4056                  * cope, else JOURNAL_DATA
4057                  */
4058                 if (jbd2_journal_check_available_features
4059                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
4060                         set_opt(sb, ORDERED_DATA);
4061                 else
4062                         set_opt(sb, JOURNAL_DATA);
4063                 break;
4064
4065         case EXT4_MOUNT_ORDERED_DATA:
4066         case EXT4_MOUNT_WRITEBACK_DATA:
4067                 if (!jbd2_journal_check_available_features
4068                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4069                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4070                                "requested data journaling mode");
4071                         goto failed_mount_wq;
4072                 }
4073         default:
4074                 break;
4075         }
4076         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4077
4078         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4079
4080 no_journal:
4081         if (ext4_mballoc_ready) {
4082                 sbi->s_mb_cache = ext4_xattr_create_cache(sb->s_id);
4083                 if (!sbi->s_mb_cache) {
4084                         ext4_msg(sb, KERN_ERR, "Failed to create an mb_cache");
4085                         goto failed_mount_wq;
4086                 }
4087         }
4088
4089         if ((DUMMY_ENCRYPTION_ENABLED(sbi) ||
4090              EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_ENCRYPT)) &&
4091             (blocksize != PAGE_CACHE_SIZE)) {
4092                 ext4_msg(sb, KERN_ERR,
4093                          "Unsupported blocksize for fs encryption");
4094                 goto failed_mount_wq;
4095         }
4096
4097         if (DUMMY_ENCRYPTION_ENABLED(sbi) &&
4098             !(sb->s_flags & MS_RDONLY) &&
4099             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_ENCRYPT)) {
4100                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_ENCRYPT);
4101                 ext4_commit_super(sb, 1);
4102         }
4103
4104         /*
4105          * Get the # of file system overhead blocks from the
4106          * superblock if present.
4107          */
4108         if (es->s_overhead_clusters)
4109                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4110         else {
4111                 err = ext4_calculate_overhead(sb);
4112                 if (err)
4113                         goto failed_mount_wq;
4114         }
4115
4116         /*
4117          * The maximum number of concurrent works can be high and
4118          * concurrency isn't really necessary.  Limit it to 1.
4119          */
4120         EXT4_SB(sb)->rsv_conversion_wq =
4121                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4122         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4123                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4124                 ret = -ENOMEM;
4125                 goto failed_mount4;
4126         }
4127
4128         /*
4129          * The jbd2_journal_load will have done any necessary log recovery,
4130          * so we can safely mount the rest of the filesystem now.
4131          */
4132
4133         root = ext4_iget(sb, EXT4_ROOT_INO);
4134         if (IS_ERR(root)) {
4135                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4136                 ret = PTR_ERR(root);
4137                 root = NULL;
4138                 goto failed_mount4;
4139         }
4140         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4141                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4142                 iput(root);
4143                 goto failed_mount4;
4144         }
4145         sb->s_root = d_make_root(root);
4146         if (!sb->s_root) {
4147                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4148                 ret = -ENOMEM;
4149                 goto failed_mount4;
4150         }
4151
4152         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
4153                 sb->s_flags |= MS_RDONLY;
4154
4155         /* determine the minimum size of new large inodes, if present */
4156         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
4157                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4158                                                      EXT4_GOOD_OLD_INODE_SIZE;
4159                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4160                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
4161                         if (sbi->s_want_extra_isize <
4162                             le16_to_cpu(es->s_want_extra_isize))
4163                                 sbi->s_want_extra_isize =
4164                                         le16_to_cpu(es->s_want_extra_isize);
4165                         if (sbi->s_want_extra_isize <
4166                             le16_to_cpu(es->s_min_extra_isize))
4167                                 sbi->s_want_extra_isize =
4168                                         le16_to_cpu(es->s_min_extra_isize);
4169                 }
4170         }
4171         /* Check if enough inode space is available */
4172         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4173                                                         sbi->s_inode_size) {
4174                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4175                                                        EXT4_GOOD_OLD_INODE_SIZE;
4176                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4177                          "available");
4178         }
4179
4180         err = ext4_reserve_clusters(sbi, ext4_calculate_resv_clusters(sb));
4181         if (err) {
4182                 ext4_msg(sb, KERN_ERR, "failed to reserve %llu clusters for "
4183                          "reserved pool", ext4_calculate_resv_clusters(sb));
4184                 goto failed_mount4a;
4185         }
4186
4187         err = ext4_setup_system_zone(sb);
4188         if (err) {
4189                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4190                          "zone (%d)", err);
4191                 goto failed_mount4a;
4192         }
4193
4194         ext4_ext_init(sb);
4195         err = ext4_mb_init(sb);
4196         if (err) {
4197                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4198                          err);
4199                 goto failed_mount5;
4200         }
4201
4202         block = ext4_count_free_clusters(sb);
4203         ext4_free_blocks_count_set(sbi->s_es,
4204                                    EXT4_C2B(sbi, block));
4205         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4206                                   GFP_KERNEL);
4207         if (!err) {
4208                 unsigned long freei = ext4_count_free_inodes(sb);
4209                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4210                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4211                                           GFP_KERNEL);
4212         }
4213         if (!err)
4214                 err = percpu_counter_init(&sbi->s_dirs_counter,
4215                                           ext4_count_dirs(sb), GFP_KERNEL);
4216         if (!err)
4217                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4218                                           GFP_KERNEL);
4219         if (err) {
4220                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4221                 goto failed_mount6;
4222         }
4223
4224         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
4225                 if (!ext4_fill_flex_info(sb)) {
4226                         ext4_msg(sb, KERN_ERR,
4227                                "unable to initialize "
4228                                "flex_bg meta info!");
4229                         goto failed_mount6;
4230                 }
4231
4232         err = ext4_register_li_request(sb, first_not_zeroed);
4233         if (err)
4234                 goto failed_mount6;
4235
4236         sbi->s_kobj.kset = ext4_kset;
4237         init_completion(&sbi->s_kobj_unregister);
4238         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
4239                                    "%s", sb->s_id);
4240         if (err)
4241                 goto failed_mount7;
4242
4243 #ifdef CONFIG_QUOTA
4244         /* Enable quota usage during mount. */
4245         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
4246             !(sb->s_flags & MS_RDONLY)) {
4247                 err = ext4_enable_quotas(sb);
4248                 if (err)
4249                         goto failed_mount8;
4250         }
4251 #endif  /* CONFIG_QUOTA */
4252
4253         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4254         ext4_orphan_cleanup(sb, es);
4255         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4256         if (needs_recovery) {
4257                 ext4_msg(sb, KERN_INFO, "recovery complete");
4258                 ext4_mark_recovery_complete(sb, es);
4259         }
4260         if (EXT4_SB(sb)->s_journal) {
4261                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4262                         descr = " journalled data mode";
4263                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4264                         descr = " ordered data mode";
4265                 else
4266                         descr = " writeback data mode";
4267         } else
4268                 descr = "out journal";
4269
4270         if (test_opt(sb, DISCARD)) {
4271                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4272                 if (!blk_queue_discard(q))
4273                         ext4_msg(sb, KERN_WARNING,
4274                                  "mounting with \"discard\" option, but "
4275                                  "the device does not support discard");
4276         }
4277
4278         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4279                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
4280                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4281
4282         if (es->s_error_count)
4283                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4284
4285         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4286         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4287         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4288         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4289
4290         kfree(orig_data);
4291         return 0;
4292
4293 cantfind_ext4:
4294         if (!silent)
4295                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4296         goto failed_mount;
4297
4298 #ifdef CONFIG_QUOTA
4299 failed_mount8:
4300         kobject_del(&sbi->s_kobj);
4301 #endif
4302 failed_mount7:
4303         ext4_unregister_li_request(sb);
4304 failed_mount6:
4305         ext4_mb_release(sb);
4306         if (sbi->s_flex_groups)
4307                 kvfree(sbi->s_flex_groups);
4308         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4309         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4310         percpu_counter_destroy(&sbi->s_dirs_counter);
4311         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4312 failed_mount5:
4313         ext4_ext_release(sb);
4314         ext4_release_system_zone(sb);
4315 failed_mount4a:
4316         dput(sb->s_root);
4317         sb->s_root = NULL;
4318 failed_mount4:
4319         ext4_msg(sb, KERN_ERR, "mount failed");
4320         if (EXT4_SB(sb)->rsv_conversion_wq)
4321                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4322 failed_mount_wq:
4323         if (sbi->s_journal) {
4324                 jbd2_journal_destroy(sbi->s_journal);
4325                 sbi->s_journal = NULL;
4326         }
4327 failed_mount3a:
4328         ext4_es_unregister_shrinker(sbi);
4329 failed_mount3:
4330         del_timer_sync(&sbi->s_err_report);
4331         if (sbi->s_mmp_tsk)
4332                 kthread_stop(sbi->s_mmp_tsk);
4333 failed_mount2:
4334         for (i = 0; i < db_count; i++)
4335                 brelse(sbi->s_group_desc[i]);
4336         kvfree(sbi->s_group_desc);
4337 failed_mount:
4338         if (sbi->s_chksum_driver)
4339                 crypto_free_shash(sbi->s_chksum_driver);
4340         if (sbi->s_proc) {
4341                 remove_proc_entry("options", sbi->s_proc);
4342                 remove_proc_entry(sb->s_id, ext4_proc_root);
4343         }
4344 #ifdef CONFIG_QUOTA
4345         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4346                 kfree(sbi->s_qf_names[i]);
4347 #endif
4348         ext4_blkdev_remove(sbi);
4349         brelse(bh);
4350 out_fail:
4351         sb->s_fs_info = NULL;
4352         kfree(sbi->s_blockgroup_lock);
4353         kfree(sbi);
4354 out_free_orig:
4355         kfree(orig_data);
4356         return err ? err : ret;
4357 }
4358
4359 /*
4360  * Setup any per-fs journal parameters now.  We'll do this both on
4361  * initial mount, once the journal has been initialised but before we've
4362  * done any recovery; and again on any subsequent remount.
4363  */
4364 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4365 {
4366         struct ext4_sb_info *sbi = EXT4_SB(sb);
4367
4368         journal->j_commit_interval = sbi->s_commit_interval;
4369         journal->j_min_batch_time = sbi->s_min_batch_time;
4370         journal->j_max_batch_time = sbi->s_max_batch_time;
4371
4372         write_lock(&journal->j_state_lock);
4373         if (test_opt(sb, BARRIER))
4374                 journal->j_flags |= JBD2_BARRIER;
4375         else
4376                 journal->j_flags &= ~JBD2_BARRIER;
4377         if (test_opt(sb, DATA_ERR_ABORT))
4378                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4379         else
4380                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4381         write_unlock(&journal->j_state_lock);
4382 }
4383
4384 static journal_t *ext4_get_journal(struct super_block *sb,
4385                                    unsigned int journal_inum)
4386 {
4387         struct inode *journal_inode;
4388         journal_t *journal;
4389
4390         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4391
4392         /* First, test for the existence of a valid inode on disk.  Bad
4393          * things happen if we iget() an unused inode, as the subsequent
4394          * iput() will try to delete it. */
4395
4396         journal_inode = ext4_iget(sb, journal_inum);
4397         if (IS_ERR(journal_inode)) {
4398                 ext4_msg(sb, KERN_ERR, "no journal found");
4399                 return NULL;
4400         }
4401         if (!journal_inode->i_nlink) {
4402                 make_bad_inode(journal_inode);
4403                 iput(journal_inode);
4404                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4405                 return NULL;
4406         }
4407
4408         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4409                   journal_inode, journal_inode->i_size);
4410         if (!S_ISREG(journal_inode->i_mode)) {
4411                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4412                 iput(journal_inode);
4413                 return NULL;
4414         }
4415
4416         journal = jbd2_journal_init_inode(journal_inode);
4417         if (!journal) {
4418                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4419                 iput(journal_inode);
4420                 return NULL;
4421         }
4422         journal->j_private = sb;
4423         ext4_init_journal_params(sb, journal);
4424         return journal;
4425 }
4426
4427 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4428                                        dev_t j_dev)
4429 {
4430         struct buffer_head *bh;
4431         journal_t *journal;
4432         ext4_fsblk_t start;
4433         ext4_fsblk_t len;
4434         int hblock, blocksize;
4435         ext4_fsblk_t sb_block;
4436         unsigned long offset;
4437         struct ext4_super_block *es;
4438         struct block_device *bdev;
4439
4440         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4441
4442         bdev = ext4_blkdev_get(j_dev, sb);
4443         if (bdev == NULL)
4444                 return NULL;
4445
4446         blocksize = sb->s_blocksize;
4447         hblock = bdev_logical_block_size(bdev);
4448         if (blocksize < hblock) {
4449                 ext4_msg(sb, KERN_ERR,
4450                         "blocksize too small for journal device");
4451                 goto out_bdev;
4452         }
4453
4454         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4455         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4456         set_blocksize(bdev, blocksize);
4457         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4458                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4459                        "external journal");
4460                 goto out_bdev;
4461         }
4462
4463         es = (struct ext4_super_block *) (bh->b_data + offset);
4464         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4465             !(le32_to_cpu(es->s_feature_incompat) &
4466               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4467                 ext4_msg(sb, KERN_ERR, "external journal has "
4468                                         "bad superblock");
4469                 brelse(bh);
4470                 goto out_bdev;
4471         }
4472
4473         if ((le32_to_cpu(es->s_feature_ro_compat) &
4474              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4475             es->s_checksum != ext4_superblock_csum(sb, es)) {
4476                 ext4_msg(sb, KERN_ERR, "external journal has "
4477                                        "corrupt superblock");
4478                 brelse(bh);
4479                 goto out_bdev;
4480         }
4481
4482         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4483                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4484                 brelse(bh);
4485                 goto out_bdev;
4486         }
4487
4488         len = ext4_blocks_count(es);
4489         start = sb_block + 1;
4490         brelse(bh);     /* we're done with the superblock */
4491
4492         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4493                                         start, len, blocksize);
4494         if (!journal) {
4495                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4496                 goto out_bdev;
4497         }
4498         journal->j_private = sb;
4499         ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4500         wait_on_buffer(journal->j_sb_buffer);
4501         if (!buffer_uptodate(journal->j_sb_buffer)) {
4502                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4503                 goto out_journal;
4504         }
4505         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4506                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4507                                         "user (unsupported) - %d",
4508                         be32_to_cpu(journal->j_superblock->s_nr_users));
4509                 goto out_journal;
4510         }
4511         EXT4_SB(sb)->journal_bdev = bdev;
4512         ext4_init_journal_params(sb, journal);
4513         return journal;
4514
4515 out_journal:
4516         jbd2_journal_destroy(journal);
4517 out_bdev:
4518         ext4_blkdev_put(bdev);
4519         return NULL;
4520 }
4521
4522 static int ext4_load_journal(struct super_block *sb,
4523                              struct ext4_super_block *es,
4524                              unsigned long journal_devnum)
4525 {
4526         journal_t *journal;
4527         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4528         dev_t journal_dev;
4529         int err = 0;
4530         int really_read_only;
4531
4532         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4533
4534         if (journal_devnum &&
4535             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4536                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4537                         "numbers have changed");
4538                 journal_dev = new_decode_dev(journal_devnum);
4539         } else
4540                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4541
4542         really_read_only = bdev_read_only(sb->s_bdev);
4543
4544         /*
4545          * Are we loading a blank journal or performing recovery after a
4546          * crash?  For recovery, we need to check in advance whether we
4547          * can get read-write access to the device.
4548          */
4549         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4550                 if (sb->s_flags & MS_RDONLY) {
4551                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4552                                         "required on readonly filesystem");
4553                         if (really_read_only) {
4554                                 ext4_msg(sb, KERN_ERR, "write access "
4555                                         "unavailable, cannot proceed");
4556                                 return -EROFS;
4557                         }
4558                         ext4_msg(sb, KERN_INFO, "write access will "
4559                                "be enabled during recovery");
4560                 }
4561         }
4562
4563         if (journal_inum && journal_dev) {
4564                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4565                        "and inode journals!");
4566                 return -EINVAL;
4567         }
4568
4569         if (journal_inum) {
4570                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4571                         return -EINVAL;
4572         } else {
4573                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4574                         return -EINVAL;
4575         }
4576
4577         if (!(journal->j_flags & JBD2_BARRIER))
4578                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4579
4580         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4581                 err = jbd2_journal_wipe(journal, !really_read_only);
4582         if (!err) {
4583                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4584                 if (save)
4585                         memcpy(save, ((char *) es) +
4586                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4587                 err = jbd2_journal_load(journal);
4588                 if (save)
4589                         memcpy(((char *) es) + EXT4_S_ERR_START,
4590                                save, EXT4_S_ERR_LEN);
4591                 kfree(save);
4592         }
4593
4594         if (err) {
4595                 ext4_msg(sb, KERN_ERR, "error loading journal");
4596                 jbd2_journal_destroy(journal);
4597                 return err;
4598         }
4599
4600         EXT4_SB(sb)->s_journal = journal;
4601         ext4_clear_journal_err(sb, es);
4602
4603         if (!really_read_only && journal_devnum &&
4604             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4605                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4606
4607                 /* Make sure we flush the recovery flag to disk. */
4608                 ext4_commit_super(sb, 1);
4609         }
4610
4611         return 0;
4612 }
4613
4614 static int ext4_commit_super(struct super_block *sb, int sync)
4615 {
4616         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4617         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4618         int error = 0;
4619
4620         if (!sbh)
4621                 return error;
4622         if (buffer_write_io_error(sbh)) {
4623                 /*
4624                  * Oh, dear.  A previous attempt to write the
4625                  * superblock failed.  This could happen because the
4626                  * USB device was yanked out.  Or it could happen to
4627                  * be a transient write error and maybe the block will
4628                  * be remapped.  Nothing we can do but to retry the
4629                  * write and hope for the best.
4630                  */
4631                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4632                        "superblock detected");
4633                 clear_buffer_write_io_error(sbh);
4634                 set_buffer_uptodate(sbh);
4635         }
4636         /*
4637          * If the file system is mounted read-only, don't update the
4638          * superblock write time.  This avoids updating the superblock
4639          * write time when we are mounting the root file system
4640          * read/only but we need to replay the journal; at that point,
4641          * for people who are east of GMT and who make their clock
4642          * tick in localtime for Windows bug-for-bug compatibility,
4643          * the clock is set in the future, and this will cause e2fsck
4644          * to complain and force a full file system check.
4645          */
4646         if (!(sb->s_flags & MS_RDONLY))
4647                 es->s_wtime = cpu_to_le32(get_seconds());
4648         if (sb->s_bdev->bd_part)
4649                 es->s_kbytes_written =
4650                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4651                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4652                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4653         else
4654                 es->s_kbytes_written =
4655                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4656         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4657                 ext4_free_blocks_count_set(es,
4658                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4659                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4660         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4661                 es->s_free_inodes_count =
4662                         cpu_to_le32(percpu_counter_sum_positive(
4663                                 &EXT4_SB(sb)->s_freeinodes_counter));
4664         BUFFER_TRACE(sbh, "marking dirty");
4665         ext4_superblock_csum_set(sb);
4666         mark_buffer_dirty(sbh);
4667         if (sync) {
4668                 error = sync_dirty_buffer(sbh);
4669                 if (error)
4670                         return error;
4671
4672                 error = buffer_write_io_error(sbh);
4673                 if (error) {
4674                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4675                                "superblock");
4676                         clear_buffer_write_io_error(sbh);
4677                         set_buffer_uptodate(sbh);
4678                 }
4679         }
4680         return error;
4681 }
4682
4683 /*
4684  * Have we just finished recovery?  If so, and if we are mounting (or
4685  * remounting) the filesystem readonly, then we will end up with a
4686  * consistent fs on disk.  Record that fact.
4687  */
4688 static void ext4_mark_recovery_complete(struct super_block *sb,
4689                                         struct ext4_super_block *es)
4690 {
4691         journal_t *journal = EXT4_SB(sb)->s_journal;
4692
4693         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4694                 BUG_ON(journal != NULL);
4695                 return;
4696         }
4697         jbd2_journal_lock_updates(journal);
4698         if (jbd2_journal_flush(journal) < 0)
4699                 goto out;
4700
4701         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4702             sb->s_flags & MS_RDONLY) {
4703                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4704                 ext4_commit_super(sb, 1);
4705         }
4706
4707 out:
4708         jbd2_journal_unlock_updates(journal);
4709 }
4710
4711 /*
4712  * If we are mounting (or read-write remounting) a filesystem whose journal
4713  * has recorded an error from a previous lifetime, move that error to the
4714  * main filesystem now.
4715  */
4716 static void ext4_clear_journal_err(struct super_block *sb,
4717                                    struct ext4_super_block *es)
4718 {
4719         journal_t *journal;
4720         int j_errno;
4721         const char *errstr;
4722
4723         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4724
4725         journal = EXT4_SB(sb)->s_journal;
4726
4727         /*
4728          * Now check for any error status which may have been recorded in the
4729          * journal by a prior ext4_error() or ext4_abort()
4730          */
4731
4732         j_errno = jbd2_journal_errno(journal);
4733         if (j_errno) {
4734                 char nbuf[16];
4735
4736                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4737                 ext4_warning(sb, "Filesystem error recorded "
4738                              "from previous mount: %s", errstr);
4739                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4740
4741                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4742                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4743                 ext4_commit_super(sb, 1);
4744
4745                 jbd2_journal_clear_err(journal);
4746                 jbd2_journal_update_sb_errno(journal);
4747         }
4748 }
4749
4750 /*
4751  * Force the running and committing transactions to commit,
4752  * and wait on the commit.
4753  */
4754 int ext4_force_commit(struct super_block *sb)
4755 {
4756         journal_t *journal;
4757
4758         if (sb->s_flags & MS_RDONLY)
4759                 return 0;
4760
4761         journal = EXT4_SB(sb)->s_journal;
4762         return ext4_journal_force_commit(journal);
4763 }
4764
4765 static int ext4_sync_fs(struct super_block *sb, int wait)
4766 {
4767         int ret = 0;
4768         tid_t target;
4769         bool needs_barrier = false;
4770         struct ext4_sb_info *sbi = EXT4_SB(sb);
4771
4772         trace_ext4_sync_fs(sb, wait);
4773         flush_workqueue(sbi->rsv_conversion_wq);
4774         /*
4775          * Writeback quota in non-journalled quota case - journalled quota has
4776          * no dirty dquots
4777          */
4778         dquot_writeback_dquots(sb, -1);
4779         /*
4780          * Data writeback is possible w/o journal transaction, so barrier must
4781          * being sent at the end of the function. But we can skip it if
4782          * transaction_commit will do it for us.
4783          */
4784         if (sbi->s_journal) {
4785                 target = jbd2_get_latest_transaction(sbi->s_journal);
4786                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4787                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4788                         needs_barrier = true;
4789
4790                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4791                         if (wait)
4792                                 ret = jbd2_log_wait_commit(sbi->s_journal,
4793                                                            target);
4794                 }
4795         } else if (wait && test_opt(sb, BARRIER))
4796                 needs_barrier = true;
4797         if (needs_barrier) {
4798                 int err;
4799                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4800                 if (!ret)
4801                         ret = err;
4802         }
4803
4804         return ret;
4805 }
4806
4807 /*
4808  * LVM calls this function before a (read-only) snapshot is created.  This
4809  * gives us a chance to flush the journal completely and mark the fs clean.
4810  *
4811  * Note that only this function cannot bring a filesystem to be in a clean
4812  * state independently. It relies on upper layer to stop all data & metadata
4813  * modifications.
4814  */
4815 static int ext4_freeze(struct super_block *sb)
4816 {
4817         int error = 0;
4818         journal_t *journal;
4819
4820         if (sb->s_flags & MS_RDONLY)
4821                 return 0;
4822
4823         journal = EXT4_SB(sb)->s_journal;
4824
4825         if (journal) {
4826                 /* Now we set up the journal barrier. */
4827                 jbd2_journal_lock_updates(journal);
4828
4829                 /*
4830                  * Don't clear the needs_recovery flag if we failed to
4831                  * flush the journal.
4832                  */
4833                 error = jbd2_journal_flush(journal);
4834                 if (error < 0)
4835                         goto out;
4836         }
4837
4838         /* Journal blocked and flushed, clear needs_recovery flag. */
4839         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4840         error = ext4_commit_super(sb, 1);
4841 out:
4842         if (journal)
4843                 /* we rely on upper layer to stop further updates */
4844                 jbd2_journal_unlock_updates(journal);
4845         return error;
4846 }
4847
4848 /*
4849  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4850  * flag here, even though the filesystem is not technically dirty yet.
4851  */
4852 static int ext4_unfreeze(struct super_block *sb)
4853 {
4854         if (sb->s_flags & MS_RDONLY)
4855                 return 0;
4856
4857         /* Reset the needs_recovery flag before the fs is unlocked. */
4858         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4859         ext4_commit_super(sb, 1);
4860         return 0;
4861 }
4862
4863 /*
4864  * Structure to save mount options for ext4_remount's benefit
4865  */
4866 struct ext4_mount_options {
4867         unsigned long s_mount_opt;
4868         unsigned long s_mount_opt2;
4869         kuid_t s_resuid;
4870         kgid_t s_resgid;
4871         unsigned long s_commit_interval;
4872         u32 s_min_batch_time, s_max_batch_time;
4873 #ifdef CONFIG_QUOTA
4874         int s_jquota_fmt;
4875         char *s_qf_names[EXT4_MAXQUOTAS];
4876 #endif
4877 };
4878
4879 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4880 {
4881         struct ext4_super_block *es;
4882         struct ext4_sb_info *sbi = EXT4_SB(sb);
4883         unsigned long old_sb_flags;
4884         struct ext4_mount_options old_opts;
4885         int enable_quota = 0;
4886         ext4_group_t g;
4887         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4888         int err = 0;
4889 #ifdef CONFIG_QUOTA
4890         int i, j;
4891 #endif
4892         char *orig_data = kstrdup(data, GFP_KERNEL);
4893
4894         /* Store the original options */
4895         old_sb_flags = sb->s_flags;
4896         old_opts.s_mount_opt = sbi->s_mount_opt;
4897         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4898         old_opts.s_resuid = sbi->s_resuid;
4899         old_opts.s_resgid = sbi->s_resgid;
4900         old_opts.s_commit_interval = sbi->s_commit_interval;
4901         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4902         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4903 #ifdef CONFIG_QUOTA
4904         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4905         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4906                 if (sbi->s_qf_names[i]) {
4907                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4908                                                          GFP_KERNEL);
4909                         if (!old_opts.s_qf_names[i]) {
4910                                 for (j = 0; j < i; j++)
4911                                         kfree(old_opts.s_qf_names[j]);
4912                                 kfree(orig_data);
4913                                 return -ENOMEM;
4914                         }
4915                 } else
4916                         old_opts.s_qf_names[i] = NULL;
4917 #endif
4918         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4919                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4920
4921         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4922                 err = -EINVAL;
4923                 goto restore_opts;
4924         }
4925
4926         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
4927             test_opt(sb, JOURNAL_CHECKSUM)) {
4928                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
4929                          "during remount not supported; ignoring");
4930                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
4931         }
4932
4933         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4934                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4935                         ext4_msg(sb, KERN_ERR, "can't mount with "
4936                                  "both data=journal and delalloc");
4937                         err = -EINVAL;
4938                         goto restore_opts;
4939                 }
4940                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4941                         ext4_msg(sb, KERN_ERR, "can't mount with "
4942                                  "both data=journal and dioread_nolock");
4943                         err = -EINVAL;
4944                         goto restore_opts;
4945                 }
4946                 if (test_opt(sb, DAX)) {
4947                         ext4_msg(sb, KERN_ERR, "can't mount with "
4948                                  "both data=journal and dax");
4949                         err = -EINVAL;
4950                         goto restore_opts;
4951                 }
4952         }
4953
4954         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
4955                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
4956                         "dax flag with busy inodes while remounting");
4957                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
4958         }
4959
4960         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4961                 ext4_abort(sb, "Abort forced by user");
4962
4963         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4964                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4965
4966         es = sbi->s_es;
4967
4968         if (sbi->s_journal) {
4969                 ext4_init_journal_params(sb, sbi->s_journal);
4970                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4971         }
4972
4973         if (*flags & MS_LAZYTIME)
4974                 sb->s_flags |= MS_LAZYTIME;
4975
4976         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4977                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4978                         err = -EROFS;
4979                         goto restore_opts;
4980                 }
4981
4982                 if (*flags & MS_RDONLY) {
4983                         err = sync_filesystem(sb);
4984                         if (err < 0)
4985                                 goto restore_opts;
4986                         err = dquot_suspend(sb, -1);
4987                         if (err < 0)
4988                                 goto restore_opts;
4989
4990                         /*
4991                          * First of all, the unconditional stuff we have to do
4992                          * to disable replay of the journal when we next remount
4993                          */
4994                         sb->s_flags |= MS_RDONLY;
4995
4996                         /*
4997                          * OK, test if we are remounting a valid rw partition
4998                          * readonly, and if so set the rdonly flag and then
4999                          * mark the partition as valid again.
5000                          */
5001                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5002                             (sbi->s_mount_state & EXT4_VALID_FS))
5003                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5004
5005                         if (sbi->s_journal)
5006                                 ext4_mark_recovery_complete(sb, es);
5007                 } else {
5008                         /* Make sure we can mount this feature set readwrite */
5009                         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
5010                                         EXT4_FEATURE_RO_COMPAT_READONLY) ||
5011                             !ext4_feature_set_ok(sb, 0)) {
5012                                 err = -EROFS;
5013                                 goto restore_opts;
5014                         }
5015                         /*
5016                          * Make sure the group descriptor checksums
5017                          * are sane.  If they aren't, refuse to remount r/w.
5018                          */
5019                         for (g = 0; g < sbi->s_groups_count; g++) {
5020                                 struct ext4_group_desc *gdp =
5021                                         ext4_get_group_desc(sb, g, NULL);
5022
5023                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5024                                         ext4_msg(sb, KERN_ERR,
5025                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5026                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
5027                                                le16_to_cpu(gdp->bg_checksum));
5028                                         err = -EINVAL;
5029                                         goto restore_opts;
5030                                 }
5031                         }
5032
5033                         /*
5034                          * If we have an unprocessed orphan list hanging
5035                          * around from a previously readonly bdev mount,
5036                          * require a full umount/remount for now.
5037                          */
5038                         if (es->s_last_orphan) {
5039                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5040                                        "remount RDWR because of unprocessed "
5041                                        "orphan inode list.  Please "
5042                                        "umount/remount instead");
5043                                 err = -EINVAL;
5044                                 goto restore_opts;
5045                         }
5046
5047                         /*
5048                          * Mounting a RDONLY partition read-write, so reread
5049                          * and store the current valid flag.  (It may have
5050                          * been changed by e2fsck since we originally mounted
5051                          * the partition.)
5052                          */
5053                         if (sbi->s_journal)
5054                                 ext4_clear_journal_err(sb, es);
5055                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5056                         if (!ext4_setup_super(sb, es, 0))
5057                                 sb->s_flags &= ~MS_RDONLY;
5058                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
5059                                                      EXT4_FEATURE_INCOMPAT_MMP))
5060                                 if (ext4_multi_mount_protect(sb,
5061                                                 le64_to_cpu(es->s_mmp_block))) {
5062                                         err = -EROFS;
5063                                         goto restore_opts;
5064                                 }
5065                         enable_quota = 1;
5066                 }
5067         }
5068
5069         /*
5070          * Reinitialize lazy itable initialization thread based on
5071          * current settings
5072          */
5073         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
5074                 ext4_unregister_li_request(sb);
5075         else {
5076                 ext4_group_t first_not_zeroed;
5077                 first_not_zeroed = ext4_has_uninit_itable(sb);
5078                 ext4_register_li_request(sb, first_not_zeroed);
5079         }
5080
5081         ext4_setup_system_zone(sb);
5082         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
5083                 ext4_commit_super(sb, 1);
5084
5085 #ifdef CONFIG_QUOTA
5086         /* Release old quota file names */
5087         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5088                 kfree(old_opts.s_qf_names[i]);
5089         if (enable_quota) {
5090                 if (sb_any_quota_suspended(sb))
5091                         dquot_resume(sb, -1);
5092                 else if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
5093                                         EXT4_FEATURE_RO_COMPAT_QUOTA)) {
5094                         err = ext4_enable_quotas(sb);
5095                         if (err)
5096                                 goto restore_opts;
5097                 }
5098         }
5099 #endif
5100
5101         *flags = (*flags & ~MS_LAZYTIME) | (sb->s_flags & MS_LAZYTIME);
5102         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5103         kfree(orig_data);
5104         return 0;
5105
5106 restore_opts:
5107         sb->s_flags = old_sb_flags;
5108         sbi->s_mount_opt = old_opts.s_mount_opt;
5109         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5110         sbi->s_resuid = old_opts.s_resuid;
5111         sbi->s_resgid = old_opts.s_resgid;
5112         sbi->s_commit_interval = old_opts.s_commit_interval;
5113         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5114         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5115 #ifdef CONFIG_QUOTA
5116         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5117         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5118                 kfree(sbi->s_qf_names[i]);
5119                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
5120         }
5121 #endif
5122         kfree(orig_data);
5123         return err;
5124 }
5125
5126 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5127 {
5128         struct super_block *sb = dentry->d_sb;
5129         struct ext4_sb_info *sbi = EXT4_SB(sb);
5130         struct ext4_super_block *es = sbi->s_es;
5131         ext4_fsblk_t overhead = 0, resv_blocks;
5132         u64 fsid;
5133         s64 bfree;
5134         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5135
5136         if (!test_opt(sb, MINIX_DF))
5137                 overhead = sbi->s_overhead;
5138
5139         buf->f_type = EXT4_SUPER_MAGIC;
5140         buf->f_bsize = sb->s_blocksize;
5141         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5142         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5143                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5144         /* prevent underflow in case that few free space is available */
5145         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5146         buf->f_bavail = buf->f_bfree -
5147                         (ext4_r_blocks_count(es) + resv_blocks);
5148         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5149                 buf->f_bavail = 0;
5150         buf->f_files = le32_to_cpu(es->s_inodes_count);
5151         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5152         buf->f_namelen = EXT4_NAME_LEN;
5153         fsid = le64_to_cpup((void *)es->s_uuid) ^
5154                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5155         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5156         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5157
5158         return 0;
5159 }
5160
5161 /* Helper function for writing quotas on sync - we need to start transaction
5162  * before quota file is locked for write. Otherwise the are possible deadlocks:
5163  * Process 1                         Process 2
5164  * ext4_create()                     quota_sync()
5165  *   jbd2_journal_start()                  write_dquot()
5166  *   dquot_initialize()                         down(dqio_mutex)
5167  *     down(dqio_mutex)                    jbd2_journal_start()
5168  *
5169  */
5170
5171 #ifdef CONFIG_QUOTA
5172
5173 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5174 {
5175         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5176 }
5177
5178 static int ext4_write_dquot(struct dquot *dquot)
5179 {
5180         int ret, err;
5181         handle_t *handle;
5182         struct inode *inode;
5183
5184         inode = dquot_to_inode(dquot);
5185         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5186                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5187         if (IS_ERR(handle))
5188                 return PTR_ERR(handle);
5189         ret = dquot_commit(dquot);
5190         err = ext4_journal_stop(handle);
5191         if (!ret)
5192                 ret = err;
5193         return ret;
5194 }
5195
5196 static int ext4_acquire_dquot(struct dquot *dquot)
5197 {
5198         int ret, err;
5199         handle_t *handle;
5200
5201         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5202                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5203         if (IS_ERR(handle))
5204                 return PTR_ERR(handle);
5205         ret = dquot_acquire(dquot);
5206         err = ext4_journal_stop(handle);
5207         if (!ret)
5208                 ret = err;
5209         return ret;
5210 }
5211
5212 static int ext4_release_dquot(struct dquot *dquot)
5213 {
5214         int ret, err;
5215         handle_t *handle;
5216
5217         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5218                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5219         if (IS_ERR(handle)) {
5220                 /* Release dquot anyway to avoid endless cycle in dqput() */
5221                 dquot_release(dquot);
5222                 return PTR_ERR(handle);
5223         }
5224         ret = dquot_release(dquot);
5225         err = ext4_journal_stop(handle);
5226         if (!ret)
5227                 ret = err;
5228         return ret;
5229 }
5230
5231 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5232 {
5233         struct super_block *sb = dquot->dq_sb;
5234         struct ext4_sb_info *sbi = EXT4_SB(sb);
5235
5236         /* Are we journaling quotas? */
5237         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) ||
5238             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5239                 dquot_mark_dquot_dirty(dquot);
5240                 return ext4_write_dquot(dquot);
5241         } else {
5242                 return dquot_mark_dquot_dirty(dquot);
5243         }
5244 }
5245
5246 static int ext4_write_info(struct super_block *sb, int type)
5247 {
5248         int ret, err;
5249         handle_t *handle;
5250
5251         /* Data block + inode block */
5252         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5253         if (IS_ERR(handle))
5254                 return PTR_ERR(handle);
5255         ret = dquot_commit_info(sb, type);
5256         err = ext4_journal_stop(handle);
5257         if (!ret)
5258                 ret = err;
5259         return ret;
5260 }
5261
5262 /*
5263  * Turn on quotas during mount time - we need to find
5264  * the quota file and such...
5265  */
5266 static int ext4_quota_on_mount(struct super_block *sb, int type)
5267 {
5268         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5269                                         EXT4_SB(sb)->s_jquota_fmt, type);
5270 }
5271
5272 /*
5273  * Standard function to be called on quota_on
5274  */
5275 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5276                          struct path *path)
5277 {
5278         int err;
5279
5280         if (!test_opt(sb, QUOTA))
5281                 return -EINVAL;
5282
5283         /* Quotafile not on the same filesystem? */
5284         if (path->dentry->d_sb != sb)
5285                 return -EXDEV;
5286         /* Journaling quota? */
5287         if (EXT4_SB(sb)->s_qf_names[type]) {
5288                 /* Quotafile not in fs root? */
5289                 if (path->dentry->d_parent != sb->s_root)
5290                         ext4_msg(sb, KERN_WARNING,
5291                                 "Quota file not on filesystem root. "
5292                                 "Journaled quota will not work");
5293         }
5294
5295         /*
5296          * When we journal data on quota file, we have to flush journal to see
5297          * all updates to the file when we bypass pagecache...
5298          */
5299         if (EXT4_SB(sb)->s_journal &&
5300             ext4_should_journal_data(d_inode(path->dentry))) {
5301                 /*
5302                  * We don't need to lock updates but journal_flush() could
5303                  * otherwise be livelocked...
5304                  */
5305                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5306                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5307                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5308                 if (err)
5309                         return err;
5310         }
5311
5312         return dquot_quota_on(sb, type, format_id, path);
5313 }
5314
5315 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5316                              unsigned int flags)
5317 {
5318         int err;
5319         struct inode *qf_inode;
5320         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5321                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5322                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5323         };
5324
5325         BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA));
5326
5327         if (!qf_inums[type])
5328                 return -EPERM;
5329
5330         qf_inode = ext4_iget(sb, qf_inums[type]);
5331         if (IS_ERR(qf_inode)) {
5332                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5333                 return PTR_ERR(qf_inode);
5334         }
5335
5336         /* Don't account quota for quota files to avoid recursion */
5337         qf_inode->i_flags |= S_NOQUOTA;
5338         err = dquot_enable(qf_inode, type, format_id, flags);
5339         iput(qf_inode);
5340
5341         return err;
5342 }
5343
5344 /* Enable usage tracking for all quota types. */
5345 static int ext4_enable_quotas(struct super_block *sb)
5346 {
5347         int type, err = 0;
5348         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5349                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5350                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5351         };
5352
5353         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5354         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5355                 if (qf_inums[type]) {
5356                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5357                                                 DQUOT_USAGE_ENABLED);
5358                         if (err) {
5359                                 ext4_warning(sb,
5360                                         "Failed to enable quota tracking "
5361                                         "(type=%d, err=%d). Please run "
5362                                         "e2fsck to fix.", type, err);
5363                                 return err;
5364                         }
5365                 }
5366         }
5367         return 0;
5368 }
5369
5370 static int ext4_quota_off(struct super_block *sb, int type)
5371 {
5372         struct inode *inode = sb_dqopt(sb)->files[type];
5373         handle_t *handle;
5374
5375         /* Force all delayed allocation blocks to be allocated.
5376          * Caller already holds s_umount sem */
5377         if (test_opt(sb, DELALLOC))
5378                 sync_filesystem(sb);
5379
5380         if (!inode)
5381                 goto out;
5382
5383         /* Update modification times of quota files when userspace can
5384          * start looking at them */
5385         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5386         if (IS_ERR(handle))
5387                 goto out;
5388         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5389         ext4_mark_inode_dirty(handle, inode);
5390         ext4_journal_stop(handle);
5391
5392 out:
5393         return dquot_quota_off(sb, type);
5394 }
5395
5396 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5397  * acquiring the locks... As quota files are never truncated and quota code
5398  * itself serializes the operations (and no one else should touch the files)
5399  * we don't have to be afraid of races */
5400 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5401                                size_t len, loff_t off)
5402 {
5403         struct inode *inode = sb_dqopt(sb)->files[type];
5404         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5405         int offset = off & (sb->s_blocksize - 1);
5406         int tocopy;
5407         size_t toread;
5408         struct buffer_head *bh;
5409         loff_t i_size = i_size_read(inode);
5410
5411         if (off > i_size)
5412                 return 0;
5413         if (off+len > i_size)
5414                 len = i_size-off;
5415         toread = len;
5416         while (toread > 0) {
5417                 tocopy = sb->s_blocksize - offset < toread ?
5418                                 sb->s_blocksize - offset : toread;
5419                 bh = ext4_bread(NULL, inode, blk, 0);
5420                 if (IS_ERR(bh))
5421                         return PTR_ERR(bh);
5422                 if (!bh)        /* A hole? */
5423                         memset(data, 0, tocopy);
5424                 else
5425                         memcpy(data, bh->b_data+offset, tocopy);
5426                 brelse(bh);
5427                 offset = 0;
5428                 toread -= tocopy;
5429                 data += tocopy;
5430                 blk++;
5431         }
5432         return len;
5433 }
5434
5435 /* Write to quotafile (we know the transaction is already started and has
5436  * enough credits) */
5437 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5438                                 const char *data, size_t len, loff_t off)
5439 {
5440         struct inode *inode = sb_dqopt(sb)->files[type];
5441         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5442         int err, offset = off & (sb->s_blocksize - 1);
5443         int retries = 0;
5444         struct buffer_head *bh;
5445         handle_t *handle = journal_current_handle();
5446
5447         if (EXT4_SB(sb)->s_journal && !handle) {
5448                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5449                         " cancelled because transaction is not started",
5450                         (unsigned long long)off, (unsigned long long)len);
5451                 return -EIO;
5452         }
5453         /*
5454          * Since we account only one data block in transaction credits,
5455          * then it is impossible to cross a block boundary.
5456          */
5457         if (sb->s_blocksize - offset < len) {
5458                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5459                         " cancelled because not block aligned",
5460                         (unsigned long long)off, (unsigned long long)len);
5461                 return -EIO;
5462         }
5463
5464         do {
5465                 bh = ext4_bread(handle, inode, blk,
5466                                 EXT4_GET_BLOCKS_CREATE |
5467                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5468         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5469                  ext4_should_retry_alloc(inode->i_sb, &retries));
5470         if (IS_ERR(bh))
5471                 return PTR_ERR(bh);
5472         if (!bh)
5473                 goto out;
5474         BUFFER_TRACE(bh, "get write access");
5475         err = ext4_journal_get_write_access(handle, bh);
5476         if (err) {
5477                 brelse(bh);
5478                 return err;
5479         }
5480         lock_buffer(bh);
5481         memcpy(bh->b_data+offset, data, len);
5482         flush_dcache_page(bh->b_page);
5483         unlock_buffer(bh);
5484         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5485         brelse(bh);
5486 out:
5487         if (inode->i_size < off + len) {
5488                 i_size_write(inode, off + len);
5489                 EXT4_I(inode)->i_disksize = inode->i_size;
5490                 ext4_mark_inode_dirty(handle, inode);
5491         }
5492         return len;
5493 }
5494
5495 #endif
5496
5497 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5498                        const char *dev_name, void *data)
5499 {
5500         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5501 }
5502
5503 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5504 static inline void register_as_ext2(void)
5505 {
5506         int err = register_filesystem(&ext2_fs_type);
5507         if (err)
5508                 printk(KERN_WARNING
5509                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5510 }
5511
5512 static inline void unregister_as_ext2(void)
5513 {
5514         unregister_filesystem(&ext2_fs_type);
5515 }
5516
5517 static inline int ext2_feature_set_ok(struct super_block *sb)
5518 {
5519         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5520                 return 0;
5521         if (sb->s_flags & MS_RDONLY)
5522                 return 1;
5523         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5524                 return 0;
5525         return 1;
5526 }
5527 #else
5528 static inline void register_as_ext2(void) { }
5529 static inline void unregister_as_ext2(void) { }
5530 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5531 #endif
5532
5533 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5534 static inline void register_as_ext3(void)
5535 {
5536         int err = register_filesystem(&ext3_fs_type);
5537         if (err)
5538                 printk(KERN_WARNING
5539                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5540 }
5541
5542 static inline void unregister_as_ext3(void)
5543 {
5544         unregister_filesystem(&ext3_fs_type);
5545 }
5546
5547 static inline int ext3_feature_set_ok(struct super_block *sb)
5548 {
5549         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5550                 return 0;
5551         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5552                 return 0;
5553         if (sb->s_flags & MS_RDONLY)
5554                 return 1;
5555         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5556                 return 0;
5557         return 1;
5558 }
5559 #else
5560 static inline void register_as_ext3(void) { }
5561 static inline void unregister_as_ext3(void) { }
5562 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5563 #endif
5564
5565 static struct file_system_type ext4_fs_type = {
5566         .owner          = THIS_MODULE,
5567         .name           = "ext4",
5568         .mount          = ext4_mount,
5569         .kill_sb        = kill_block_super,
5570         .fs_flags       = FS_REQUIRES_DEV,
5571 };
5572 MODULE_ALIAS_FS("ext4");
5573
5574 static int __init ext4_init_feat_adverts(void)
5575 {
5576         struct ext4_features *ef;
5577         int ret = -ENOMEM;
5578
5579         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5580         if (!ef)
5581                 goto out;
5582
5583         ef->f_kobj.kset = ext4_kset;
5584         init_completion(&ef->f_kobj_unregister);
5585         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5586                                    "features");
5587         if (ret) {
5588                 kfree(ef);
5589                 goto out;
5590         }
5591
5592         ext4_feat = ef;
5593         ret = 0;
5594 out:
5595         return ret;
5596 }
5597
5598 static void ext4_exit_feat_adverts(void)
5599 {
5600         kobject_put(&ext4_feat->f_kobj);
5601         wait_for_completion(&ext4_feat->f_kobj_unregister);
5602         kfree(ext4_feat);
5603 }
5604
5605 /* Shared across all ext4 file systems */
5606 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5607 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5608
5609 static int __init ext4_init_fs(void)
5610 {
5611         int i, err;
5612
5613         ext4_li_info = NULL;
5614         mutex_init(&ext4_li_mtx);
5615
5616         /* Build-time check for flags consistency */
5617         ext4_check_flag_values();
5618
5619         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5620                 mutex_init(&ext4__aio_mutex[i]);
5621                 init_waitqueue_head(&ext4__ioend_wq[i]);
5622         }
5623
5624         err = ext4_init_es();
5625         if (err)
5626                 return err;
5627
5628         err = ext4_init_pageio();
5629         if (err)
5630                 goto out7;
5631
5632         err = ext4_init_system_zone();
5633         if (err)
5634                 goto out6;
5635         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5636         if (!ext4_kset) {
5637                 err = -ENOMEM;
5638                 goto out5;
5639         }
5640         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5641
5642         err = ext4_init_feat_adverts();
5643         if (err)
5644                 goto out4;
5645
5646         err = ext4_init_mballoc();
5647         if (err)
5648                 goto out2;
5649         else
5650                 ext4_mballoc_ready = 1;
5651         err = init_inodecache();
5652         if (err)
5653                 goto out1;
5654         register_as_ext3();
5655         register_as_ext2();
5656         err = register_filesystem(&ext4_fs_type);
5657         if (err)
5658                 goto out;
5659
5660         return 0;
5661 out:
5662         unregister_as_ext2();
5663         unregister_as_ext3();
5664         destroy_inodecache();
5665 out1:
5666         ext4_mballoc_ready = 0;
5667         ext4_exit_mballoc();
5668 out2:
5669         ext4_exit_feat_adverts();
5670 out4:
5671         if (ext4_proc_root)
5672                 remove_proc_entry("fs/ext4", NULL);
5673         kset_unregister(ext4_kset);
5674 out5:
5675         ext4_exit_system_zone();
5676 out6:
5677         ext4_exit_pageio();
5678 out7:
5679         ext4_exit_es();
5680
5681         return err;
5682 }
5683
5684 static void __exit ext4_exit_fs(void)
5685 {
5686         ext4_exit_crypto();
5687         ext4_destroy_lazyinit_thread();
5688         unregister_as_ext2();
5689         unregister_as_ext3();
5690         unregister_filesystem(&ext4_fs_type);
5691         destroy_inodecache();
5692         ext4_exit_mballoc();
5693         ext4_exit_feat_adverts();
5694         remove_proc_entry("fs/ext4", NULL);
5695         kset_unregister(ext4_kset);
5696         ext4_exit_system_zone();
5697         ext4_exit_pageio();
5698         ext4_exit_es();
5699 }
5700
5701 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5702 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5703 MODULE_LICENSE("GPL");
5704 module_init(ext4_init_fs)
5705 module_exit(ext4_exit_fs)